Related: Why the U.S. needs a $17 minimum wage • Why eliminate the tipped minimum wage

Interactive

Many colleges and universities now require educators to have a Bachelor’s degree prior to entering an education program. After getting a Bachelor’s degree future teachers have one more year to get teaching credentials or in many cases they can spend two more years getting a Master’s degree.

In other words: teachers face the unenviable choice of incurring greater debt prior to entering the workforce or changing majors and entering the workforce after only four years with less debt but also less credentials. This is a significant problem since students average $30,000 in college debt. Some of my colleagues owe something closer to $60,000 in debt. It is the passion, the call of teaching, the desire to make a difference that leads people into education not the paychecks.

When you consider teacher’s salaries you have to ponder how someone with this much debt can afford to take a starting position with the national average starting salary less than $40,000 in 2017! Why would anyone become a teacher?

It is not surprising that education programs are now considering changing course in Virginia to make education once again a four-year degree program. If we want the best people in education we need to make it affordable to get a degree. We also need to consider the portability of that degree. Some states work with surrounding states for reciprocity of licensure, however; a teacher usually has to take additional courses if he/she relocates too far away. This presents yet another drawback.

Many teachers who enter the field quickly leave. The number of teachers leaving the field with less than five years of experience keeps growing. Why? Everyone has their own reasons but some of the reasons cut across schools, school divisions, states, and our nation. Teachers do not feel supported and the role and responsibilities of teachers just keep increasing. When I was growing up in the 70s and 80s my teachers taught us, supported us, disciplined us, attended a few meetings and our testing was a nationally normed standardized test that was given a few times throughout my K-12 education.

Today, teachers teach, discipline, support, remediate, attend countless trainings, prepare students for dozens of evaluations at the local and state level and are told to do more with less. This mantra grew louder during the recession and continues today. During the recession our responsibilities and accountability grew while our support and financial assistance shrunk. It is time that those making laws and regulations that impact educators and students start having conversations with those teachers and students.

This is starting to happen in a few places, but it needs to happen across the nation at the local, state, and federal level. It is amazing that people who have never set foot in a public school are setting policy and regulations that impact our day-to-day practice with no knowledge of the impact and no dollars provided. Our new educators need innovative and effective mentoring programs, meaningful professional development, supports to help them grow into great teachers all of which require funding. When they feel supported, they will stay.

Many localities across the country will never be able to offer competitive salaries compared to places a short drive away due to how schools are funded in some states. I have colleagues in Virginia with over 20 years of experience still making less than $50,000. We need to find a way to show educators that they make a difference and acknowledge their skills and trainings.

Until that happens we need to find other ways to compensate educators. Our school divisions need to get creative. Some are starting to offer one type of leave and teachers do not have to worry about sick or personal days. Others are making professional development more meaningful by letting individual teachers determine what skills or knowledge they need to be effective. Some are making mentoring programs more effective so our young teachers feel supported and stay in the field of teaching. Still others provide work from home days instead of requiring teachers to make an appearance on a teacher work day. There is still more that could be done.

To teach is to uplift lives, but it is the teachers and public school employees who need a boost. It is the field of education that needs support. Public education should be something we all believe in, support, and want funded.

Chairman Neal, Ranking Member Brady, and members of the committee, thank you for the opportunity to testify today on rising inequality in the United States.

My name is Elise Gould and I am an economist at the Economic Policy Institute (EPI) in Washington, D.C. EPI is a nonprofit, nonpartisan think tank that believes every working person deserves a good job with fair pay, affordable health care, and retirement security. To achieve this goal, EPI conducts research and analysis on the economic status of working America. I am an economist with particular expertise on wages and wage inequality.

My testimony establishes that the poor performance of American workers’ wages in recent decades—particularly the failure of workers’ wages to grow at anywhere near the pace of overall productivity—is one of the country’s central economic challenges. Indeed, it’s hard to think of a more important economic development in recent decades. It is at the root of the large rise in overall income inequality that has attracted so much attention in recent years. A range of other economic challenges—reducing poverty, increasing mobility, closing racial and gender wage gaps, and spurring a more complete recovery from the Great Recession—also rely largely on boosting hourly wage growth for the vast majority.

The main points of this testimony are as follows:

1. Income inequality is the primary reason why the vast majority of Americans experienced disappointing growth in their living standards over the last four decades. In other words, most Americans are seeing slow income growth because most of overall income growth is going to households at the top.
2. Labor market income represents the largest source of income for most Americans and that is why we cannot tackle income inequality without tackling wage growth.
3. Wage growth in the last four decades has been uneven, with notable growth only at the top while wages for most workers have failed to rise with productivity growth.
4. This uneven wage growth—what we can call growing wage inequality—continued through the 2000s, as wage gaps between demographic groups persisted, and, in some cases, worsened. Further, the growth in inequality cannot be explained by growing demand for college-educated workers.
5. Recent wage gains for the lowest wage workers can be explained by tight labor markets and the institution of a number of state-level minimum wage increases.
6. Going forward, policymakers should prioritize keeping labor markets tight while also strengthening institutions and policies that provide workers the leverage they will need to achieve decent wage growth even when the economy is not at full employment. These policies and institutions include strengthening and enforcing labor standards, making it easier for workers to collectively bargain, and raising the federal minimum wage.

Rising inequality helps explain the disappointing living standards growth for the vast majority

In recent decades, the vast majority of Americans have experienced disappointing growth in their living standards—despite economic growth that could have easily generated faster gains in their living standards had it been broadly shared. Figure A helps us assess the economic performance for different groups by charting the cumulative percentage increase in household income for the top 1 percent compared with the bottom 90 percent. Breaking the top 1 percent down even further would show nearly as dramatic an increase in inequality just within this top group, but it would also stretch the vertical axis so much as to make it nearly unreadable. What this shows is that income grew swiftly for a small sliver of the population while living standards for most grew far more slowly.

Figure A measures the change in comprehensive income—including cash, market-based incomes (wages and salaries, dividends, rent, capital gains, and business income); noncash income, such as employer contributions to health insurance premiums; and cash and noncash government transfers like Social Security, food stamps, Medicare, and Medicaid. It is easy to see that the rise in American inequality is extreme even when using these comprehensive income measures, which include taxes and transfers.

One striking aspect of the figure is the large decline in top 1 percent incomes following the onset of the Great Recession after 2007. However, a similarly large fall in top 1 percent incomes resulted from stock market declines following the 2001 recession as well, and as the figure shows, as of 2015, these incomes mostly recovered. Even with these losses, the top 1 percent of household income has grown 229 percent since 1979, far in excess of the slower 46 percent growth—just 1.0 percent annualized growth—for the bottom 90 percent of households.

Figure A

Most income for the vast majority of households comes from their wages

Among the bottom 90 percent of American households, labor income—including wages and wage-related income such as employer contributions to health insurance benefits for workers and Social Security and Medicare for retired workers—represents the vast majority of income. Figure B illustrates the share of total income that is composed of wages and wage-related incomes for the bottom 90 percent and the top 1 percent of household incomes. What’s clear from the figure is that the vast majority of American households get the vast majority of their incomes from wages and wage-related sources while a much smaller share of incomes for the top 1 percent comes from these sources. Over the entire period, contributions of wages and wage-related income for the top 1 percent averaged just under 40 percent, while it averaged 86 percent for the bottom 90 percent of households, more than twice as high.

In 1979, 86.9 percent of household income for the bottom 90 percent came from wages and wage-related sources. By 2015, the share had fallen slightly to 84.0 percent. Over this period, much of the rise in earnings for most households came from increasing work hours and not increasing hourly wages (Bivens et al. 2014). Because the vast majority of household income for the bottom 90 percent comes from labor income, it is clear that growing wage inequality is at the root of slow growing incomes for the vast majority of American households.

Figure B

Broad wage suppression underlies sluggish living standards growth for the vast majority

Because wages are their primary source of income, the rise in income inequality that has blocked living standards growth for the vast majority since 1979 has been driven by a pronounced reduction in the collective and individual bargaining power of ordinary American workers. As a result of their eroded bargaining power, their wages have grown agonizingly slow over the past generation. Rising wage inequality—anemic wage growth for the vast majority, combined with substantial wage gains for those at the very top—has left most Americans with an ever-shrinking portion of the overall wage bill. It is also the case that if labor incomes—i.e., wages—had not grown so unequally, then the share of total output available to be claimed by capital owners, again concentrated at the top of the income distribution, would have been significantly smaller. It is the combination of these two factors—driven by wages for the vast majority lagging productivity—that has led to the erosion of most Americans’ living standards. The resulting lackluster wage growth and inequality have afflicted men and women, and people at all levels of education; even the college educated are just treading water.

Figure C demonstrates that since 1979, “real” (inflation-adjusted) hourly pay for the vast majority of American workers has diverged from economy-wide productivity. After tracking rather closely in the three decades following World War II, growing productivity and typical worker compensation diverged. From 1979 to 2017, productivity grew 70.3 percent, while hourly compensation of production and nonsupervisory workers grew just 11.1 percent. Productivity thus grew six times faster than typical worker compensation.

Figure C

A natural question that arises from this story is just where did the “excess” productivity go? A significant portion of it went to higher corporate profits and increased income accruing to capital and business owners (Bivens et al. 2014). But much of it went to those at the very top of the wage distribution, as shown in Figure D. The top 1 percent of earners saw cumulative gains in annual wages of 157.3 percent between 1979 and 2017—far in excess of economywide productivity growth and nearly four times faster than average wage growth (40.1 percent, not shown). Over the same period, top 0.1 percent earnings grew 343.2 percent, with the latest spike reflecting the sharp increase in executive compensation (Mishel and Wolfe 2018). Over the same period, despite a growing economy and increases in productivity, the earnings for the bottom 90 percent only rose 22.2 percent. It’s important to remember this disparity. When policymakers consider policies to improve productivity growth, they also should consider ways that growth could better translate into wage growth for most workers and not just for those at the very top.

Figure D

While the CPS-ORG—the primary data set used in the remainder of this testimony—does not allow disaggregation within the top 5 percent of the earnings distribution, it is still instructive for measuring the growth in wage inequality over the last 40-odd years. Figure E illustrates that for all but the highest earners, hourly wage growth has been weak. If it hadn’t been for a period of strong across-the-board wage growth in the late 1990s, wages for most would have fallen outright. Median hourly wages rose 14.0 percent between 1979 and 2018, compared with an increase of 4.1 percent for the 10th-percentile worker (i.e., the worker who earns more than only 10 percent of workers). Over the same period, the 95th-percentile worker saw growth of 56.1 percent.

Figure E

Wage inequality continued through the 2000s

Figure F illustrates the trends in wages for select deciles (and the 95th percentile), showing the cumulative percent change in real hourly wages from 2000 to 2018. The continuing overall story of inequality is clear. From 2000 to 2018, the 95th-percentile wage grew over three times as fast as wages at the median. Additional details on recent wage trends can be found in Gould 2019a, also submitted into the written record. During this period, wage inequality among men grew more than wage inequality among women, and the gap between men and women at the top continued to widen in part because men are more likely to occupy jobs at the top of the wage distribution. Black–white wage gaps also widened between 2000 and 2018 as white wages grew more than four times as fast as black wages across most of the wage distribution (Gould 2019b).

Steep and rising wage inequality is too often blamed on growing demand for workers with higher levels of educational attainment—the more schooling you have, the more you’ll be paid, the theory goes. But research has shown that rising inequality cannot be explained by rising wages for those with more educational attainment. The more salient story between 2000 and 2018 is not one of a growing differential of wages between college and high school graduates, but one of growing wage inequality between the top relative to the vast majority of workers, as shown in Figure F. Wage inequality is driven by changes within education groups (among people with the same education) and not between education groups. Among college graduates, there has been a significant pulling away at the very top of the wage distribution. In fact, the bottom 60 percent of workers with a college degree still have lower wages than they did in 2000 (Gould 2019c).

Increases in inequality over the last 18 years clearly cannot be explained away by claims that employers face a growing shortage of college graduates and that, correspondingly, wage inequality is some unfortunate side effect of the positive gains from automation that we neither can nor would want to alter. There are plenty of good reasons to provide widespread access to college educations and skill development, but expanding college enrollment and graduation is not an answer to escalating wage inequality.

Figure F

Recent wage gains among lower-wage workers

After years of wage losses, the lowest-wage workers finally exceeded their 1979 wage levels in 2017 (as shown in Figure E.) These recent wage gains can essentially be fully explained by tight labor markets and state-level minimum wage increases.

Because the lowest-wage workers are often the most vulnerable in economic downturns, it often takes them longer to recover in economic expansions. Achieving genuine full employment is one way that low- and moderate-wage workers gain enough bargaining power to increase their wages; employers have to pay more to attract and retain the workers they need when idle workers are scarce. The “lever” for higher wages that comes from full employment is most important for workers at the bottom of the wage distribution: For a given fall in the unemployment rate, wage growth rises more for low-wage workers, and in the absence of stronger labor standards like a strong minimum wage, it is often only in the tightest of labor markets that low-wage workers see stronger wage growth (Bivens and Zipperer 2018).

Figure G illustrates how the wages of low-, middle-, and high-wage workers change in response to labor market conditions. Each bar shows the percentage-point change in the growth rate of inflation-adjusted wages following a 1-percentage-point increase in the state-specific unemployment rate, employment-to-population ratio, and prime-age employment to population ratio (among 25- to 54-year-olds), respectively. The blocks of bars show results for the 10th, 50th, and 90th percentiles of wages, corresponding to low-, middle-, and high-wage workers.

The results indicate that a 1-percentage-point drop in unemployment results in annual wage growth for workers at the 10th percentile of the wage distribution that is 0.5 percentage points faster. For example, if annual real wage growth is at 1.0 percent, then a 1-percentage-point fall in unemployment would result in annual real wage growth rising to 1.5 percent. For workers near the median of the wage distribution, wage growth is faster by 0.4 percentage points following a 1-percentage-point decline in the unemployment rate. For workers at the 90th percentile of the wage distribution, wage growth is faster by 0.3 percentage points following a 1-percentage-point decline in the unemployment rate. There are similar findings for the other measures of the labor market shown: stronger effects for low- and moderate-wage workers than for the highest-wage workers. What this tells us is that low- and moderate-wage workers do relatively worse in bad times, but also see a relatively larger boost in good times. That alone can explain the recent rise in wages for the lowest-wage and for middle-wage workers over the last few years as shown in Figure F.

Another policy lever was pulled in a number of states over the last few years. In 2018, the minimum wage was increased in 13 states and the District of Columbia through legislation or referendum, and in eight states because the minimum wage is indexed to inflation in those states. And, these changes in state minimum wages came on the heels of other recent changes to minimum wages in many of the same states over the previous couple of years. In fact, when we compare states that have had any minimum wage change since 2013 with states that did not have a minimum wage change during that time, the results—as shown in Figure H—are highly suggestive. Wage growth at the 10th percentile in states with at least one minimum wage increase from 2013 to 2018 was more than 50 percent faster than in states without any minimum wage increases (13.0 percent vs. 8.4 percent). As expected, given women’s lower wages in general, this result is even stronger for women (13.0 percent vs. 6.0 percent), though men also experienced much faster 10th-percentile wage growth in states with minimum wage increases than in those without (12.0 percent vs. 8.6 percent).

Figure G

Figure H

Policies to increase wage growth for the vast majority will increase Americans’ living standards

Beyond seeking to keep labor markets tight, policymakers could take other steps to foster strong broad-based wage growth, such as raising the federal minimum wage, expanding eligibility for overtime pay, addressing gender and racial pay disparities, and protecting and strengthening workers’ rights to bargain collectively for higher wages and benefits. Going forward, policymakers should do two things. They should prioritize wage growth by continuing to push toward genuine full employment. And they should provide workers with the leverage to achieve decent wage growth even when the economy is not at full employment by strengthening and enforcing labor standards and making it easier for workers to collectively bargain.

The right to collectively bargain is tightly linked to wages and incomes. In fact, the spread of collective bargaining that followed the passage of the National Labor Relations Act in 1935 led to decades of faster and fairer economic growth that persisted until the late 1970s. But since the 1970s, declining unionization has fueled rising inequality and stalled economic progress for the broad American middle class. Figure I shows that when unions are weak, the highest incomes go up even more, but when unions are strong, the bottom 90 percent enjoy more income growth.

Figure I

This correlation is no accident. Unions have strong positive effects not only on the wages of union workers but also on the wages of comparable nonunion workers, as unions set standards for entire industries and occupations (Rosenfeld, Denice, and Laird 2016). Further, the union wage boost is largest for low-wage workers and larger at the middle than at the highest wage levels, larger for black and Hispanic workers than for white workers, and larger for those with lower levels of education—wage increases for these groups help narrow wage inequalities.

We know how big a force for equality unions are by looking at how much their decline has contributed to inequality between middle- and high-wage workers: union decline can explain one-third of the rise in wage inequality among men and one-fifth of the rise in wage inequality among women from 1973 to 2007. Among men, the erosion of collective bargaining has been the largest single factor driving a wedge between middle- and high-wage workers (Western and Rosenfeld 2011).

For a more thorough analysis of how collective bargaining affects worker living standards see How Today’s Unions Help Working People (Bivens et al. 2017). For more policies that will raise wages, see EPI’s Policy Agenda (EPI 2018).

Sources

Bivens, Josh, Lora Engdahl, Elise Gould, Teresa Kroeger, Celine McNicholas, Lawrence Mishel, Zane Mokhiber, Heidi Shierholz, Marni von Wilpert, Valerie Wilson, and Ben Zipperer. 2017. How Today’s Unions Help Working People: Giving Workers the Power to Improve Their Jobs and Unrig the Economy. Economic Policy Institute, August 2017.

Bivens, Josh, Elise Gould, Lawrence Mishel, and Heidi Shierholz. 2014. Raising America’s Pay: Why It’s Our Central Economic Policy Challenge. Economic Policy Institute, Briefing Paper No. 378, June 2014.

Bivens, Josh, and Ben Zipperer. 2018. The Importance of Locking in Full Employment for the Long Haul. Economic Policy Institute, August 2018.

Economic Policy Institute (EPI). 2018. Policy Agenda. December 2018.

Gould, Elise. 2019a. State of Working America Wages 2018: Wage Inequality Marches On—and Is Even Threatening Data Reliability. Economic policy Institute, February 2019.

Gould, Elise. 2019b. “Stark Black–white Divide in Wages Is Widening Further.” Working Economics (Economic Policy Institute blog), February 27, 2019.

Gould, Elise. 2019c. “Higher Returns on Education Can’t Explain Growing Wage Inequality.” Working Economics (Economic Policy Institute blog), March 15, 2019.

Mishel, Lawrence, and Julia Wolfe. 2018. “Top 1.0 Percent Reaches Highest Wages Ever—Up 157 Percent Since 1979.” Working Economics (Economic Policy Institute blog), October 18, 2018.

Rosenfeld, Jake, Patrick Denice, and Jennifer Laird. 2016. Union Decline Lowers Wages of Nonunion Workers: The Overlooked Reason Why Wages Are Stuck and Inequality Is Growing. Economic Policy Institute, August 2016.

Western, Bruce, and Jake Rosenfeld, “Unions, Norms, and the Rise in U.S. Wage Inequality,” American Sociological Review vol. 76 (2011), 513–37.

This report was produced in collaboration with The Century Foundation’s Bernard L. Schwartz Rediscovering Government Initiative.

This report lays out the basic facts of the wage crisis. In the first section, we document trends in earnings over the last 70 years. We focus on three periods: First, we look at the immediate postwar period, from 1947 through 1979, when wage growth was rapid for all workers, including workers at the bottom and the middle. Next, we look at the period from 1979 through the present, when wage growth was much slower, especially for low- and middle-wage workers. Finally, we take a closer look at the period since 2009, which marked the beginning of the economic recovery from the Great Recession.

Our analysis centers primarily on the hourly wage: the payment workers receive, excluding benefits, for an hour of work.2 We also examine data on hourly compensation: the payment workers receive for an hour of work, including nonwage benefits. We do so primarily to demonstrate that including benefits does not significantly alter any of the conclusions we draw about wage trends over the last four decades. Unless otherwise noted, all numbers in the text, tables, and figures have been adjusted for inflation.3

In the second section of this report, we set these wage trends since the late 1970s against the backdrop of enormous changes in the composition of the U.S. workforce over the same period. American workers today are, on average, older (and so potentially more experienced) and much better educated than their earlier counterparts. Women, workers of color, and immigrants now make up a much larger share of the workforce.4 The workforce has also largely shifted out of manufacturing and into the service industries; out of unionized workplaces and into nonunionized workplaces; and out of the Northeast and Midwest and into the West and South.

In the final section, we examine differences in wages and employment across U.S. regions, categorized to emphasize several parts of the country where manufacturing remains particularly important. The most striking feature of the regional analysis is how varied the workforce and wage rates are across the regions. There are substantial regional differences in the educational attainment and racial/ethnic makeup of the workforce; the share of immigrants in the workforce; unionization levels; the importance of manufacturing; and the wages paid to workers. Of particular interest, the biggest declines in relative employment occurred in the Great Lakes region as manufacturing declined.

Wage trends

The postwar period, 1947–1979: A period of strong, shared wage growth

As we document below, wage growth has been slow and unequal for the last four decades, a tendency that has continued through most of the economic recovery following the Great Recession. But the current pattern of slow and unequal growth is far from an inevitable feature of the U.S. economy. In the three decades from the end of World War II through the 1970s, wage growth was much more rapid and much more widely shared.

U.S. statistical agencies have collected a wide range of data on workers’ pay. These data are particularly rich for roughly the last four decades, allowing us to paint a detailed picture of trends since the late 1970s, including differences by gender, race and ethnicity, educational attainment, and other worker characteristics. For the immediate postwar period through the 1960s, however, the available data are more limited and generally don’t allow us to distinguish between different kinds of workers.

Nevertheless, the available data, summarized in Table 1, are sufficient to confirm that wage growth was both more rapid and more equally shared during the three decades after World War II than has been the case over the four most recent decades. Given these data constraints in the earlier period, we present four different measures of earnings, each of which allows us to see different aspects of workers earnings.

Table 1

The first measure is an estimate of the average inflation-adjusted hourly earnings for all workers in the economy (row 1), based on data taken from the Bureau of Economic Analysis’s National Income and Product Accounts.5 Between 1947 and 1979, the average real hourly wage grew 2.2 percent per year. From 1979 to the present, average growth fell to 0.7 percent per year, only one-third of the average rate in the earlier postwar period.

The second measure is the average hourly earnings of production and nonsupervisory workers (row 2), which is tracked by the Bureau of Labor Statistics in a monthly survey of employers. This group makes up about 80 percent of all workers in the economy and is a good proxy for the wage experience of the bottom 80 percent of workers.6 From 1947 through 1979, the hourly wage of nonsupervisory workers increased at a 2.0 percent annual rate. Since 1979, however, the rate has decelerated sharply, to just 0.3 percent per year, about 13 percent of its earlier rate.

These first two measures of hourly wage data both show a steep decline in wage growth in recent decades. A comparison of the growth rate for all workers (row 1) with the corresponding rate for nonsupervisory workers (row 2) also lets us draw some conclusions about developments in wage inequality in the two different periods. In the immediate postwar period, the annual growth rate in wages for nonsupervisory workers was 2.0 percent, not far behind the 2.2 percent annual average for all workers, which includes generally better-paid supervisors. In the more recent period, wage growth decelerated for all workers and for nonsupervisory workers, but the falloff was steeper for nonsupervisory workers, suggesting that wage growth has been substantially more unequal in recent years.

The third pay measure in Table 1 is average annual earnings for all workers (row 3), based on data collected by the Social Security Administration.7 These data are not adjusted for the hours worked in a year, but they provide some independent corroboration of the hourly wage trends in the first two series. Between 1947 and 1979, average growth in annual earnings was 2.1 percent, right in the middle of the average rates in the first two series (2.0 to 2.2 percent per year). Between 1979 and 2016, however, the average growth rate fell by more than half, to 0.9 percent per year. Published earnings tables from the Social Security Administration allow us to look separately at the earnings growth over the same two periods for the bottom 90 percent of earners (row 4). The average growth rate for the bottom 90 percent of workers also declined substantially, from 2.0 percent per year between 1947 and 1979, to 0.5 percent per year from 1979 to 2016. As with the earlier comparison of the experience of the total workforce with roughly the bottom 80 percent, the Social Security data also suggest that earnings growth has been substantially more unequal in recent decades than in the earlier postwar period. In the recent period, earnings growth for the bottom 90 percent of earners fell more steeply (to 26.2 percent of its earlier rate) than it did for the workforce as a whole (42.5 percent), a group that includes the experience of the highest earners.

The final pay measure in Table 1 is an estimate of growth in annual earnings from work for all households (row 5) and the bottom 90 percent of households (row 6), based on an analysis by Thomas Piketty and Emmanuel Saez of income tax return data from the Internal Revenue Service covering the period 1947 through 2011.8 The IRS data tell a similar story. The growth rate for annual earnings from work dropped substantially for all working households (from 1.9 percent per year in the 1947–1979 period, to 0.9 percent per year for 1979–2011) and even more steeply for working households in the bottom 90 percent (from 1.8 percent per year in the first period to 0.6 percent per year in the second period). Once again, wage growth was much faster and much more equally shared until about 1979, when a prolonged period of slow and unequal growth became the norm.

The last four decades, 1979–present: A period of slow and unequal wage growth

For three decades after World War II wages grew rapidly across the board. But since the end of the 1970s, wages have grown slowly for most of the workforce, and the gap between the best-paid workers and the rest of the workforce has widened significantly. We can illustrate both of these trends by showing the cumulative inflation-adjusted growth since 1979 in the hourly wages paid to three different kinds of workers (Figure A): low-wage workers, which we define here as workers in the 10th percentile of the wage distribution (that is, workers who make more than the bottom 10 percent of workers but less than the top 90 percent of workers); middle-wage workers (which we define as 50th-percentile or median workers, who make more than the bottom half of workers and less than the top half of workers); and high-wage or 90th-percentile workers (who earn more than 90 percent of all workers but less than the top 10 percent).

Figure A

We draw the data on low-, middle-, and high-wage workers—and much of the rest of the analysis presented here—from the Current Population Survey (CPS), a monthly survey of about 60,000 U.S. households conducted by the Census Bureau for the Bureau of Labor Statistics. A major feature of the CPS is that it provides a careful measure of workers’ weekly and hourly earnings along with detailed information on workers’ characteristics, including age, gender, race and ethnicity, educational attainment, state of residence, country of birth, industry, occupation, and union membership.9

The first key trend since 1979 is the historically slow growth in real wages. In 2017, middle-wage workers earned just 16.8 percent more than their counterparts almost four decades earlier. This corresponds to an annualized inflation-adjusted growth rate over the 38-year period of just 0.4 percent per year. The real wage increase for low-wage workers (those at the 10th percentile) was even slower: 8.9 percent over 38 years, or a 0.2 percent annualized growth rate. As noted earlier, we do not have comparable wage data for the 1947–1979 period. But if the deceleration in wage growth at the 10th and the 50th percentiles between the immediate postwar period and the most recent four decades was proportional to what happened to the deceleration in hourly wages experienced by nonsupervisory workers (or in annual earnings of the bottom 90 percent of wage and salary workers in the Social Security or IRS data), then we can estimate that average real wage growth for low- and middle-wage workers would have been between three and seven times faster between 1947 and 1979 than it has been since.

This slow growth is particularly disappointing for two reasons. First, as we will see in the next section, U.S. workers today are generally older (and hence potentially more experienced) and substantially better educated than workers were at the end of the 1970s.10 Second, for workers at the bottom and the middle, most of the increase in real wages over the entire period took place in the short window between 1996 and the early 2000s. For the large majority of workers over the last four decades, wages were essentially flat or falling apart from a few short bursts of growth.11

The second key trend that stands out in Figure A is the widening gap between workers at the top and the rest of workforce. Wage growth for workers at the 90th percentile was much faster—a cumulative increase of 46.9 percent, or about 1.0 percent per year on a compounded basis—than it was for those at the 50th and 10th percentiles. In 1979, the 90th-percentile worker already made significantly more than the worker at the middle (2.0 times more, $31.10 at the top, compared with $15.41 at the middle, both in 2017 dollars), but the faster growth at the top than at the middle meant that the 90th-percentile worker made 2.5 times more than the middle-wage worker by 2017 ($45.67 at the top, compared with $18.00 at the middle).12 Relative to low-wage workers at the 10th percentile, the earnings for those at top increased from a ratio of 3.5 in 1979 to a ratio of 4.7 in 2017.

The Current Population Survey data, however, almost certainly understate the actual increase in wage inequality. One limitation of the Current Population Survey data analyzed here is that the survey does not adequately capture the earnings of very highly paid workers, such as many CEOs and top corporate managers, lawyers, doctors, and other highly paid professionals. Research on top 1 percent incomes (which include wages, but also other forms of income) by Thomas Piketty and Emmanuel Saez demonstrates that income growth was much more robust for the top 1 percent than it was for the rest of the top 10 percent. If the CPS were able to collect wage data accurately at the very top, the gap between the top and the rest would almost certainly be even higher than what appears in Figure A.13

So far, the wage data we’ve seen have referred to the workforce as a whole, but trends differ in important ways by gender. Figures B and C present wage growth in the same format as Figure A, but now separately for men and women. For both men and women, we see the same widening of wage inequality over time. The biggest difference, however, is that real wage growth was much higher for the typical woman (up 33.8 percent between 1979 and 2017) than for the typical man (up 8.1 percent) and for high-wage women (up 73.0 percent) relative to high-wage men (up 37.0 percent). To be clear, throughout the entire period, men at the bottom, middle, and top of the male wage distribution consistently earned more than women in the same position of the female wage distribution. In 2017, for example, men at the 90th percentile of men’s wages earned about 25 percent more than women at the 90th percentile of women’s wages ($50.00 for men, $40.00 for women); at the median, the difference was about 21 percent ($20.00 for men, $16.50 for women); and at the 10th percentile, about 11 percent ($10.00 for men, $9.00 for women). What has happened over the last four decades is that each position in the women’s wage distribution has moved somewhat closer to the corresponding position in the men’s distribution, likely reflecting a combination of much greater participation of women in paid work, their substantial increase in educational attainment relative to men (women as a group are now better educated then men), and women’s entry into higher-paying professions.14

Figure B

Figure C

The growth of wage inequality is also evident in the pattern of wage growth by education. Figure D summarizes trends in wages for the typical worker at five different levels of educational attainment. For workers with less than a four-year college degree (over 60 percent of the workforce in 2017), real wages for the typical (median) worker were lower in 2017 than they had been in 1979. In 2017, workers with less than a high school diploma made 9.6 percent less than what similar workers made in 1979; for workers with a high school diploma, the decline was 2.7 percent relative to their 1979 counterparts; and for workers with some college, but not a four-year degree, the decline was 1.1 percent. Meanwhile, the median real wage grew for workers with a four-year college degree (up 15.9 percent) and for those with an advanced degree (up 30.0 percent).

Figure D

The wage trends in Figure D reveal a clear increase in inequality. But focusing on the median wage earner in each education category masks another important dimension of growing wage inequality. It isn’t simply that wage inequality grew between workers with different levels of formal education; it is also the case that wage inequality grew within groups of workers that had the same educational qualifications. To illustrate this point, we can measure the spread of the wage distribution just among college graduates by comparing the wage of the 90th-percentile college graduate with that of the 10th-percentile college graduate, first in 1979 and then in 2017. Between 1979 and 2017, the ratio of what a high-paid college graduate (at the 90th percentile of college graduates) earned relative to what a low-paid college graduate (at the 10th percentile) earned increased substantially, from 3.6 to 4.8. In real terms, the spread between the 10th- and 90th-percentile wages grew from $29.53 in 1979 to $45.69 in 2017, with the 10th-percentile college graduate’s wage growing by less than $0.50 in total over the entire 38-year period. As a result, inequality increased markedly, even if we limit our analysis just to college graduates (with no advanced degree).

Wage inequality also rose sharply across racial and ethnic groups. Figure E displays wage trends for the median worker in each of four mutually exclusive race and ethnicity groups: white non-Hispanic, black non-Hispanic, Hispanic of any race, and Asian American/Pacific Islander non-Hispanic.15 Wage growth for the median African American and Hispanic worker was slowest: up 10.1 percent for African American workers and 11.9 percent for Hispanic workers. The increase in median wages was roughly double (up 23.3 percent) for the median white worker and more than three times higher for the median Asian American/Pacific Islander worker (up 36.7 percent).

Figure E

The current economic recovery, 2009–present: A period of persistent slow wage growth

The period since 2009, which coincides with the economic recovery from the Great Recession, is of particular interest. Between 2009 and 2014, inflation-adjusted wages were flat or falling across a range of available wage measures. In the more recent period, real wages grew, but, as we shall see, growth rates for recovery as a whole still trail far behind the 2.0–2.2 percent annual rates of the earlier postwar period from 1947–1979.

For the shorter time period since 2009, we can supplement the CPS data we have focused on so far with several wage (and compensation) measures not available for the full period since 1979. Table 2 summarizes inflation-adjusted wage and compensation data for several different measures covering the period from 2009 to 2017. The first three columns reproduce the CPS hourly wage data for low-, middle-, and high-wage workers, the same series referenced in Figure A. The fourth column shows another series we have already seen—hourly wage data for nonsupervisory workers from the Current Employment Statistics (CES) survey of employers. The fifth column presents hourly wages from the same CES source, now for all workers, including supervisory workers; we haven’t used this wage series before because it is only available from 2006. The sixth column shows the growth in wages and salaries for all workers in the Bureau of Labor Statistics’ Employer Costs for Employee Compensation (ECEC) data. (The final column displays ECEC data including nonwage benefits, which we will turn to later.) Figure F presents the cumulative, inflation-adjusted growth for each of these measures, relative to the start of the recovery in 2009.

Table 2

Figure F

The first striking feature of the data for the current recovery is that all of the wage series were at or below their 2009 level continuously from 2010 through 2014. Over this period, wage declines were the steepest for low- and middle-wage workers (those at the 10th and the 50th percentiles), but even high-wage workers (at the 90th percentile) saw no real wage growth over these first five years of economic recovery. A second feature of the data in Figure F is the faster wage growth as the recovery continued, particularly in the period between 2014 and 2017. Wages for workers at the top (90th percentile), the average worker (which includes top earners) from the CES, the average nonsupervisory worker, the average worker from the ECEC, and even low-wage workers, all accelerated sharply after 2013 or 2014. The timing of this acceleration underscores the importance of sustained tight labor markets for generating rapid wage growth.16 The especially strong performance of wages for low-wage workers (10th percentile) likely also reflects the impact of a series of increases in the minimum wage over the same period.17 A final feature of the data is how slow wage growth has been in the current recovery. Even the long economic expansion since 2009 has not been sufficient to raise wages anywhere near the rates experienced on average across economic upturns and downturns in the three decades immediately following World War II.

Including benefits does not change the story

Most of the data we’ve presented so far have focused on wages, primarily drawn from an analysis of microdata from the Current Population Survey. The CPS is one of the most widely used and authoritative sources of information on workers’ wages. Unfortunately, the CPS contains little information on employee benefits, which constitute an important part of workers’ total compensation.

A careful examination of the available data, however, suggests that incorporating benefits would do little to change the picture. The main reason is that total inflation-adjusted compensation for the typical (median) worker has not increased that much faster than wages for the same worker. A recent analysis by the Economic Policy Institute, for example, estimates that total compensation—wages plus nonwage benefits—for the median worker rose only 10.1 percent between 1979 and 2016, compared with a 9.2 percent increase for median wages over the same period using a compatible wage measure.18 A similar pattern holds if we focus only on the current economic recovery. If we factor in the cost of nonwage benefits, between 2009 and 2017 total compensation for workers in the ECEC data did grow faster (Table 2, last column) than wages alone for the same group (Table 2, next-to-last column), but the difference was small (0.7 percent per year vs. 0.4 percent per year) and the average growth rate was still well under 1 percent per year.

How can growth in benefits add so little to the growth in total compensation of the median worker—especially given the rising cost of health care benefits? The cost of employer-provided health insurance has indeed risen dramatically since the end of the 1970s, but the increase in costs has been at least partly offset by a large decline in the share of workers receiving employer-provided health insurance, along with an increase in the share of those rising costs borne by workers themselves through higher deductibles and co-pays. The decline in employer-provided coverage has also been much greater for low- and middle-wage workers than it has been for high-wage workers, reinforcing rather than mitigating the rise in inequality.19

A similar story holds for retirement plans, another major form of nonwage compensation. Employers have shifted retirement plans from defined-benefit (traditional pension) plans to defined-contribution (401(k)-style) plans,20 and workers at the middle and the bottom have seen far bigger declines in access to employer-sponsored retirement plans than workers at the top.21 Both of these developments have slowed the growth of compensation, especially for low- and middle-wage workers.

The decline in manufacturing and unionization contributes to wage stagnation, inequality

Manufacturing has traditionally paid higher wages and benefits than other sectors, especially for workers with less than a college degree. The long-term decline in manufacturing employment, both as a share of total employment and, more recently, including the absolute number of workers in manufacturing (see Figure G), has contributed to wage stagnation and widening wage inequality. In a recent analysis, Lawrence Mishel documents that manufacturing still pays wages that are about 10 percent higher, and total compensation (including benefits) that is about 15 percent higher, than the nonmanufacturing private sector, even after controlling for key worker characteristics. But Mishel also notes that the manufacturing premium has declined by about 25 percent relative to what it was in the 1980s.22

Figure G

Collective bargaining also raises wages and benefits substantially.23 As a result, the long-term decline in union membership and representation has played an important role in slow and unequal wage growth. On average, workers covered by a union contract earn about 13 percent more than similar workers in nonunion jobs.24 Unions also help to raise wages of nonunion workers in areas where unionization rates are higher, primarily by setting wage standards that even nonunion employers feel obligated to follow for market reasons or in response to developing norms surrounding pay and benefits.25

The changing composition of the U.S. workforce

The disturbing wage trends we highlight in the preceding section take place against a backdrop of major changes in the structure of employment in the United States. In this section, we describe these changes in the composition of the U.S. workforce by age, education, gender, race and ethnicity, nativity, industry, and union status. For the purposes of understanding wage trends, the most notable development of the last four decades is the large increase in the age and educational attainment of the workforce. The large increases in potential work experience and educational attainment that we document in this section are difficult to reconcile with the slow and unequal real wage growth that has characterized the last four decades. The data also show that the workforce today includes much larger shares of women, people of color, and immigrants. The workforce has also shifted out of manufacturing and into nonunion workplaces.

Age

In 1979, almost one-fourth (22.7 percent) of workers were between the ages of 16 and 24. By 2017, the share of young workers had dropped to just one in eight (12.5 percent) (Figure H). The share of what economists call “prime-age” workers (workers ages 25–54) increased from 62.5 percent in 1979 to 64.4 percent in 2017, but the biggest increases were among workers ages 55–64 (up to 17.1 percent in 2017, from 11.8 percent in 1979) and 65 and older (doubling their share, from 3.0 percent to 6.0 percent).

Figure H

Education

Today’s workforce is also much better educated. In 1979, almost one-fourth (23.2 percent) of workers had less than a high school diploma (Figure I). By 2017, the share without a high school diploma had fallen to only 7.2 percent. The share of workers with a four-year college degree almost doubled, from 12.5 percent in 1979 to 23.7 percent in 2017; and the share with an advanced degree (master’s degree, doctorate, or law, dental, medical, or similar degree) jumped from 5.9 percent to 13.4 percent over the same period.

Figure I

Gender, race, and ethnicity

The U.S. workforce is also remarkably more diverse in terms of gender, race, and ethnicity than it was in 1979. Women now make up close to half of all workers (46.9 percent in 2017, up from 41.7 percent in 1979) (Figure J). In 1979, workers of color made up 16.3 percent of the workforce; today they make up 36.6 percent (Figure K). Among workers of color, the largest increases have been among Hispanic workers, whose share in the workforce has more than tripled, from 4.8 percent in 1979 to 16.9 percent by 2017. The share of African American workers in the workforce is also up, from 9.4 percent in 1979 to 11.9 percent in the most recent data.

Figure J

Figure K

The Current Population Survey, which is the source of the data we analyze here, did not allow respondents to identify as Asian American or Pacific Islander until 1989, but in 1979, 2.2 percent of the workforce identified as “Other” (not white, black, or Hispanic). This “Other” category included Asian Americans, Native Americans, Pacific Islanders, and other groups not captured in the three larger categories. From 1989 on, however, the survey did allow respondents to identify as Asian American/Pacific Islander. In Figure K, for 1979, we report the share in the broad “Other” category, which includes Asian American, Pacific Islander, Native American, and other groups; for 2017, we separate out the Asian American and Pacific Islander respondents from the “Other” category, which is now primarily Native Americans. The broadly defined “Other” group—which combines “Asian American/Pacific Islander” and “Other” in 2017—increased from 2.2 percent in 1979 to 7.9 percent in 2017, with a large share of the increase attributable to a rise in immigration from Asian countries.26

Nativity

The share of immigrants in the U.S. workforce has increased substantially. In 1994 (the earliest year the Current Population Survey asked respondents if they were born outside the country), immigrants made up 9.7 percent of the workforce (Figure L). By 2017, the immigrant share had increased to 17.1 percent.27

Figure L

Industry

The structure of employment across industries has changed markedly over the last four decades. Perhaps most importantly, the share of workers in manufacturing has plummeted. In 1979, almost one-fourth (23.6 percent) of the workforce was in manufacturing (Figure M). Today, the manufacturing sector employs only a little over one in 10 (10.4 percent of) workers. Employment in the public sector (federal, state, and local government) has declined, although less sharply, from 15.8 percent in 1979 to 13.7 percent in 2017. Meanwhile, nonmanufacturing employment—overwhelmingly jobs in the service sector—rose from 60.5 percent to over three-quarters (75.9 percent) of all employment in 2017.

Figure M

Union status

Workers today are also far less likely to be members of a union or to be represented by a union at their workplace (Figure N). In 1983 (the earliest year for which union status is available in the data), almost one-fourth (23.3 percent) of workers were unionized; today, the unionization rate (11.9 percent) is about half the 1983 level.28

Figure N

A regional perspective

So far, we have paid attention exclusively to trends for the national workforce. In this final section, we take a closer look at differences in employment and wages across regions.

States

Employment is heavily concentrated in a relatively small number of states. In 2017, California (11.9 percent) and Texas (8.4 percent) were the states with the largest share of the U.S. workforce (Figure O). Together, just nine states (California, Texas, Florida, New York, Illinois, Pennsylvania, Ohio, Georgia, and North Carolina) account for more than half of total employment. Twenty-one states (including the District of Columbia) each have 1.0 percent or less of the national workforce.

Figure O

Between 1979 and 2017, the distribution of employment shifted across the states. In broad terms, employment shares fell in states in the Northeast and Midwest and rose in states in the South and West (Figure P). The biggest gainers were Florida (up 2.7 percentage points) and Texas (up 2.3). The biggest declines were in New York (down 1.6 percentage points), Ohio (down 1.3), Illinois (down 1.2), and Pennsylvania and Michigan (both down 1.1).

Figure P

Regions

In what follows, we modify the standard Census regions and divisions to focus better on long-run trends related to manufacturing employment. The Census Bureau divides the states into four regions (Northeast, Midwest, South, and West) and nine “divisions” (Pacific, Mountain, West North Central, West South Central, East North Central, East South Central, New England, Middle Atlantic, and South Atlantic).29 Table 3 defines our alternative regional breakdown and lists the states that belong to each. The first region, following research by Andrew Stettner, Joel Yudken, and Michael McCormack,30 is the “Great Lakes” region, which consists of the manufacturing-heavy states that border any of the Great Lakes (Illinois, Indiana, Michigan, Minnesota, New York, Ohio, Pennsylvania, and Wisconsin). The second group is what we label the “Manufacturing Southeast” states (Alabama, Georgia, Kentucky, Mississippi, North Carolina, South Carolina, and Tennessee), where manufacturing employment is high or rising relative to other states in the Southeast. Given their importance in national manufacturing, we treat California and Texas as their own regions. The remaining categories are Other Pacific (minus California), Other West (minus Texas), Other Midwest (minus the Great Lakes states), Other Southeast (minus the Manufacturing Southeast states), and Other Northeast (minus the Great Lakes states).

Table 3

In 2017, the Great Lakes states together accounted for over one-fourth (26.7 percent) of total employment in the country (Table 4). The Manufacturing Southeast region (13.1 percent) had the second-largest workforce among our regions, followed closely by the Other Southeast states (12.9 percent). California, on its own, contained 11.9 percent of the U.S. workforce. The Other Northeast states’ share was almost as big (10.3 percent) as California’s. Texas (8.4 percent) and the Other West states (7.4 percent) also had a substantial share of the nation’s workforce. The two smallest groups in our categories were Other Midwest (5.1 percent) and Other Pacific (4.3 percent).

Table 4

Organizing the states into these regions helps us see stark changes in the distribution of national employment over the last four decades (Figure Q). Between 1979 and 2017, the share of the U.S. workforce in the Great Lakes region fell 7.1 percentage points, more than four times the decline in Other Northeast (down 1.7) and Other Midwest (down 1.2). Meanwhile, employment shares grew in all other regions: Other West (up 2.7 percentage points), Texas (up 2.3), Other Southeast (up 2.2), California (up 1.3), Manufacturing Southeast (up 0.9), and Other Pacific (up 0.7).

Figure Q

Wages by regions

The hourly wage received by the typical worker varies by about 25 percent across the nine regions, from a low of $16.20 per hour in the Manufacturing Southeast region to $20.25 per hour in the Other Northeast region (Figure R). Typical wages are also relatively high in Other Pacific ($19.00), California ($18.69), and Great Lakes ($18.00). Typical wages are lower in Texas ($17.00), Other West ($17.00), Other Southeast ($16.88), and Other Midwest ($16.71). (Note that in our discussion of the wage and employment trends in these regions, we have pooled the data from the Current Population Survey for 2015 through 2017 in order to improve the accuracy of our analysis; for simplicity, we refer to this period as “2017” in the text.)

Figure R

The wage spread across regions is even higher in manufacturing (Figure S). The typical manufacturing worker in the Other Northeast region earns $23.46 per hour, which is 36 percent higher than the $17.21 per hour paid to the typical manufacturing worker in the Manufacturing Southeast region. The manufacturing-intensive Great Lakes region ($19.47) is closer to the Manufacturing Southeast region at the bottom than it is to the Other Northeast states.

Figure S

Regional workforce composition

As we noted earlier, the national workforce is considerably older today than it was four decades ago (Figure H). In 2017, our nine regions show some differences across regions in the age structure of the workforce, but the differences are relatively small (Figure T).

Figure T

The national workforce is also much better educated now than it was in 1979 (Figure I), but fairly large differences in educational attainment persist across regions (Figure U). The Other Northeast region has the best-educated workforce, by a considerable margin. This region has the highest share of workers with an advanced degree (18.6 percent, compared with the Great Lakes region in second place at 13.8 percent); the highest share with a college degree (27.0 percent), compared with California and Other Pacific (tied for second place at 24.1 percent); and the lowest share of the workforce with less than a high school diploma (5.3 percent). Three regions—Great Lakes, California, and Other Pacific—cluster a bit behind the Other Northeast region, with about 37 percent of their workforce having either a four-year college degree or an advanced degree; of these regions, the Great Lakes and Other Pacific regions also have among the lowest shares of workers with less than a high school diploma.

Figure U

Texas stands out as the regional workforce with the lowest level of formal education. Only one-third (33.3 percent) of Texas workers have a four-year college degree or more, and 11.2 percent have less than a high school diploma. The Other West and Manufacturing Southeast regions also trail the rest of the regions with respect to educational attainment. In both of these regions, only about one-third of their workforce have a four-year college degree or more; 8.0 percent in Other West and 7.5 percent in Manufacturing Southeast have not completed a high school education.

California simultaneously has high shares of workers with college and advanced degrees (24.1 percent and 13.3 percent, respectively) and a high share of workers with less than a high school diploma (10.0 percent).

Across most of the regions, the share of women in the paid workforce is around 47–48 percent over the 2015–2017 period. The three regions where the share of women workers is lower are Texas (44.7 percent), California (45.2 percent), and Other West (45.4 percent) (Figure V).

Figure V

The racial and ethnic composition of the workforce differs substantially across regions (Figure W). Most of the regions have a majority-white workforce; the largest shares of white workers are in the Other Midwest and Great Lakes regions (83.0 percent and 74.4 percent, respectively). In California and Texas, workers of color make up a majority of the workforce.

Figure W

The distribution of African American, Hispanic, and Asian American/Pacific Islander workers varies widely across the country. The share of African American workers ranges from levels well below the national average (11.7 percent in 2015–2017) in Other Pacific (3.4 percent), Other West (4.1 percent), and California (5.6 percent), to almost one-fourth of the workforce in the Manufacturing Southeast region (23.1 percent). The range for Hispanic workers is even larger, from under 7 percent in the Manufacturing Southeast (6.9 percent) and Other Midwest (6.4 percent) regions, to over one-third of the workforce in California (35.7 percent) and Texas (36.1 percent). Asian American and Pacific Islander workers make up a substantial portion of the workforce in California (17.0 percent) and the Other Pacific region (15.4 percent), but a much smaller share in the rest of the country.

Almost one-third of workers in California (32.4 percent), almost one-fourth of workers in Texas (22.5 percent), and almost one-fifth in Other Northeast (19.9 percent) are immigrants (Figure X). The immigrant share of the workforce is also above the national average (17.0 percent) in Other Southeast (18.0 percent). Other Midwest (6.7 percent) and Manufacturing Southeast (9.3 percent) are the regions with the lowest share of immigrant workers.

Figure X

In all nine regions, manufacturing is a major employer (Figure Y), but the share of regional employment in manufacturing varies from a low of about 8 percent in the Other Southeast (7.6 percent), Other Northeast (8.0 percent), and Other West (8.2 percent) regions, to about 13 percent in the Manufacturing Southeast (12.6 percent) and Great Lakes (13.2 percent) regions. Even excluding the Great Lakes states, the Other Midwest region also has a high share of manufacturing employment (11.7 percent).

Figure Y

Large differences in the working populations of the nine regions obscure an important aspect of the distribution of manufacturing employment across the country. Only 13.2 percent of employment in the Great Lakes region is in manufacturing, but because of the size of the collective economies of the states in this region, this region accounts for almost one-third (31.2 percent) of all manufacturing employment in the United States. The Manufacturing Southeast region employs 15.4 percent of all manufacturing workers in the country.31

The unionization rate differs greatly across regions (Figure Z). In the Other Pacific region, 18.4 percent of workers are either a member of a union or covered by a union contract at their place of work. Other regions with a relatively high share of union workers are California (17.2 percent), Great Lakes (16.5 percent), and Other Northeast (14.5 percent). The regions with the lowest unionization rates are Texas (5.6 percent), Manufacturing Southeast (6.2 percent), and Other Southeast (7.0 percent).

Figure Z

Conclusion

This report seeks to lay out key elements of the wage crisis facing the U.S. economy. The two most salient features of wage trends over the last four decades are the slow growth in real wages for the large majority of workers and the increasingly unequal nature of the wage structure across multiple dimensions, including class, race, gender, and geography. These trends have played out against a backdrop of big increases in the educational attainment and work experience of the U.S. workforce, particularly among women. Both of these increases in the “human capital” of the workforce should have worked to raise wages and narrow inequalities.

But, clearly, other forces are pushing in the opposite direction—toward slower and more unequal wage growth. A complete analysis of those forces lies beyond the scope of this report, but three factors identified here are clearly implicated. The first is the long-term decline in manufacturing employment, which has traditionally been the sector of the economy that pays relatively high wages to the non-college-educated workers who still make up a large majority of the U.S. workforce. The second countervailing force is the long-term decline in union membership, which has had a negative effect on the pay and benefits of both union and nonunion workers. The third factor is the persistent failure to run the economy at full employment, which has undermined the important leverage workers have in tight labor markets.32

Endnotes

1. For a comprehensive review of the historical data on wages and incomes, see Lawrence Mishel et al., The State of Working America (Ithaca, N.Y.: Cornell Univ. Press, 2012).

2. For salaried workers, we rely on estimates of hourly pay based on dividing an annual salary by the usual hours worked per year.

3. We use the CPI-U-RS, linked to the CPI-U-X1 and then the CPI-U, for years when the CPI-U-RS is not available.

4. Together, women and workers of color are already a large majority of the U.S. workforce. Workers of color will make up the majority of the U.S. working class (defined as those without a four-year college degree) by 2032 (see Valerie Wilson, People of Color Will Be a Majority of the American Working Class in 2032, Economic Policy Institute, June 2016).

5. Bureau of Economic Analysis (BEA), National Income and Product Accounts, Tables 2.1 and 6.9, accessed August 2018 via BEA’s interactive data application.

6. Bureau of Labor Statistics, Current Employment Statistics interactive data, average hourly earnings for production and nonsupervisory workers, series CES0500000008, accessed August 2018.

7. Social Security Administration, Wage Statistics, various years, accessed August 2018; and Kopczuk Wojciech, Emmanuel Saez, and Jae Song, “Earnings Inequality and Mobility in the United States: Evidence from Social Security Data Since 1937,” Quarterly Journal of Economics 125, no. 1 (February 2010): 91–128. https://doi.org/10.1162/qjec.2010.125.1.91.

8. Thomas Piketty and Emmanuel Saez, Table B3, in Excel files downloadable at https://eml.berkeley.edu//~saez/TabFig2015prel.xls (published 2016, accessed August 2018); files are 2015 data updates to the tables and figures in Piketty and Saez, “Income Inequality in the United States, 1913–1998,” Quarterly Journal of Economics 118, no. 1 (February 2003): 1–39. Mishel et al. provide evidence that the Piketty and Saez results reflect, in part, that households supplied more hours of work to the paid labor market beginning in 1979; as a result, household earnings growth was faster than individuals’ earnings growth; see Lawrence Mishel et al., The State of Working America (Ithaca, N.Y.: Cornell Univ. Press, 2012).

9. The main source of data presented here is the Current Population Survey (CPS)’s Outgoing Rotation Group drawn from either the Center for Economic and Policy Research extract of the CPS Outgoing Rotation Group or from the Economic Policy Institute’s State of Working America Data Library.

10. The increases in age and education hold for workers at the middle and the bottom of the pay scale. For more details, see Lawrence Mishel, Low-Wage Workers Have Far More Education Than They Did in 1968, Yet They Make Far Less, Economic Policy Institute, January 2014; and Cherrie Bucknor, Low-Wage Workers: Still Older, Smarter, and Underpaid, Center for Economic and Policy Research, May 2015.

11. See Josh Bivens and Ben Zipperer, The Importance of Locking in Full Employment for the Long Haul, Economic Policy Institute, August 2018; and Dean Baker and Jared Bernstein, Getting Back to Full Employment: A Better Bargain for Working People, Center for Economic and Policy Research, 2013.

12. See Table 2.

13. See Thomas Piketty and Emmanuel Saez, “Income Inequality in the United States, 1913–1998,” Quarterly Journal of Economics 118, no. 1 (February 2003): 1–39; see also the 2015 data updates to Piketty and Saez 2003, downloadable at https://eml.berkeley.edu/~saez/TabFig2015prel.xls.

14. For a detailed look at wage trends since 2000, see Elise Gould, The State of American Wages 2017: Wages Have Finally Recovered from the Blow of the Great Recession but Are Still Growing Too Slowly and Unequally, Economic Policy Institute, March 2018. For additional details on the closing of the gender pay gap, see Elise Gould, Jessica Schieder, and Kathleen Geier, What Is the Gender Pay Gap and Is It Real?: The Complete Guide to How Women Are Paid Less Than Men and Why It Can’t Be Explained Away, Economic Policy Institute, October 2016.

15. These race and ethnicity categories are defined consistently throughout. For example, unless otherwise specified, “white” is used to mean “white non-Hispanic.” In this example, the “Asian American/Pacific Islander” group is Asian Americans and Pacific Islanders from 1989 through 2017, spliced to the “Other” category for 1979 to 1989, which includes Asian Americans, Pacific Islanders, Native Americans, and other groups.

16. For more on the relationship between a sustained tight labor market and wage growth, see Josh Bivens and Ben Zipperer, The Importance of Locking in Full Employment for the Long Haul, Economic Policy Institute, August 2018.

17. See Elise Gould, The State of American Wages 2017: Wages Have Finally Recovered from the Blow of the Great Recession but Are Still Growing Too Slowly and Unequally, Economic Policy Institute, March 2018.

18. See Economic Policy Institute, State of Working America Data Library, “Productivity and Hourly Compensation” and “Wage Ratios” (accessed August 2018).

19. See Economic Policy Institute, State of Working America Data Library, “Health Insurance Coverage” (accessed August 2018); and Hye Jin Rho and John Schmitt, Health-Insurance Coverage Rates for US Workers, 1979–2008, Center for Economic and Policy Research, March 2010.

20. Lawrence Mishel et al., The State of Working America (Ithaca, N.Y.: Cornell Univ. Press, 2012), Figure 4J.

21. Economic Policy Institute, State of Working America Data Library, “Pension Coverage” (accessed August 2018).

22. Lawrence Mishel, Yes, Manufacturing Still Provides a Pay Advantage, but Staffing Firm Outsourcing Is Eroding It, Economic Policy Institute, March 2018.

23. For more on the benefits of collective bargaining, see Josh Bivens et al., How Today’s Unions Help Working People: Giving Workers the Power to Improve Their Jobs and Unrig the Economy, Economic Policy Institute, August 2017.

24. Lawrence Mishel et al., The State of Working America (Ithaca, N.Y.: Cornell Univ. Press, 2012), Table 4.33.

25. See Tom VanHeuvelen, “Moral Economies or Hidden Talents? A Longitudinal Analysis of Union Decline and Wage Inequality, 1973–2015,” Social Forces, May 2018, https://doi.org/10.1093/sf/soy045; and Jake Rosenfeld, Patrick Denice, and Jennifer Laird, Union Decline Lowers Wages of Nonunion Workers: The Overlooked Reason Why Wages Are Stuck and Inequality is Growing, Economic Policy Institute, August 2016.

26. See Hye Jin Rho et al., Diversity and Change: Asian American and Pacific Islander Workers, Center for Economic and Policy Research, July 2011.

27. Using data from the decennial census in 1980, the immigrant share of the total labor force—which includes the workforce plus the unemployed—was 6.7 percent; see Migration Policy Institute, “Immigrant Share of the U.S. Population and Civilian Labor Force, 1980–Present” (online chart), accessed August 2018.

28. For further details, see Josh Bivens et al., How Today’s Unions Help Working People: Giving Workers the Power to Improve Their Jobs and Unrig the Economy, Economic Policy Institute, August 2017; and John Schmitt and Kris Warner, “The Changing Face of U.S. Labor, 1983–2008,” WorkingUSA 13, no. 2 (June 2010): 263–279.

29. For maps defining regions and divisions, see the pages “Regions” and “Divisions” at the U.S. Census Bureau’s Geography Web Atlas (https://www.census.gov/geo/reference/webatlas/).

30. Andrew Stettner, Joel S. Yudken, and Michael McCormack, Why Manufacturing Jobs Are Worth Saving, The Century Foundation, June 2017.

31. Authors’ analysis of CPS microdata.

32. For further reading, see Josh Bivens and Ben Zipperer, The Importance of Locking in Full Employment for the Long Haul, Economic Policy Institute, August 2018; Elise Gould, The State of American Wages 2017: Wages Have Finally Recovered from the Blow of the Great Recession but Are Still Growing Too Slowly and Unequally, Economic Policy Institute, March 2018; Josh Bivens, Lawrence Mishel, and John Schmitt, It’s Not Just Monopoly and Monopsony: How Market Power Has Affected American Wages, Economic Policy Institute, April 2018; and Lawrence Mishel et al., The State of Working America (Ithaca, N.Y.: Cornell Univ. Press, 2012).

An extended period of low unemployment—sometimes called a “high-pressure labor market”—could deliver large benefits by boosting wages and healing the economic damage done by the Great Recession and the slow recovery following it. While such an extended period of high-pressure labor markets could be achieved without threatening the Fed’s ability to keep inflation expectations anchored at its preferred 2 percent, if the Fed raised its long-run inflation target (which Bivens [2017] notes would be a wise move), then the period of time they could tolerate extremely low unemployment would be even longer.

This report highlights how important extended periods of labor market tightness are for the wage growth of low- and middle-wage workers generally. It also highlights how extended periods of labor market tightness can have powerful effects in reducing disparities in labor market outcomes between workers of different races. As the Fed considers the proper path of interest rate increases going forward, it needs to internalize these substantial benefits for workers who have largely not enjoyed a proportionate share of economic growth in recent decades. Too often it is asserted that there is little the Fed can do to help push back against the problem of rising inequality (either by income class or race); this is not true. What is true is that pushing back against this inequality does require them to take on a small bit of risk—the risk that by keeping labor markets tight for an extended period of time they may find inflation temporarily surpassing their target. But this risk is small, and the costs of a short period of time running the economy “too hot” and spurring excess inflation is dwarfed by the potential benefits stemming from faster and more equal wage growth and from reducing race-based disparities in labor market outcomes.

Key findings of this report are:

• High unemployment is linked with rising inequality. Excessively high unemployment likely explains a large part of the growing gap between economywide productivity and hourly pay for the typical American worker—and the resulting rise in income inequality—that has emerged since the late 1970s.
• Equitable wage growth is linked with extended low unemployment. Since 1979, the only period of strong across-the-board wage growth occurred in the late 1990s and early 2000s, which was also the only period of extended low unemployment in recent decades. This coincidence of extended labor market tightness and healthy, equitable wage growth was not by chance.
• Low- and moderate-wage workers benefit the most from decreases in the unemployment rate. Using a panel data set of labor market outcomes by state each year since 1979, we reconfirm well-established findings that pay for low- and moderate-wage workers is more sensitive to changes in the unemployment rate than pay for higher-wage workers. Put simply, these workers need lower unemployment rates than their high-wage peers do in order to achieve decent wage growth.
• Increasingly lower unemployment rates have been required for workers to reap these benefits. National data confirm that pay for the median worker needed increasingly lower unemployment rates to see any growth at all between 1979 and 2007, before the Great Recession. This likely reflects the steady downward pressure on workers’ bargaining power over those years.
• During the Great Recession and its aftermath, the relationship between pay and unemployment has been weaker. It will likely become stronger again as the economy normalizes. The statistical link between unemployment and growth in pay post-2008 is weaker and less robust, both in state-level and national data. This means that the very large increases in the unemployment rate due to the recession resulted in virtually no decline in nominal wages. While it may seem hard to believe that workers’ pay has been better protected against excess unemployment since 2008 than before, this finding is consistent with the hypothesis of downward nominal wage rigidity, the phenomenon where even deep and long recessions rarely lead to cuts in nominal pay. The corollary has been true as well in the years since 2008: as unemployment has dropped, nominal wage growth has not accelerated as quickly as one might expect from historical trends. As the economy normalizes after the Great Recession and slow recovery, we expect historical statistical relationships between pay and unemployment to be re-established.
• Tight labor markets can narrow racial employment gaps (specifically, for EPOPs and average hours worked). Contrary to common assumptions, tight labor markets can narrow not just absolute racial employment gaps, but also relative racial employment gaps (or ratios). We find that tighter aggregate labor markets lead to disproportionate gains in African American employment (as measured by employment-to-population ratios, or EPOPs) and average hours worked for African American households relative to white workers. Ingrained pessimism on this point assumes that while absolute gaps in racial employment outcomes may be reduced through an improving labor market, relative gaps will remain essentially constant. Our results on employment are more hopeful, with even relative gaps closing as labor markets tighten. However, we do not find evidence that relative unemployment or labor force participation gaps narrow between African American and white workers when these aggregate unemployment or labor force participation rates improve.

Background

Between 2008 and the end of 2016, the broad consensus among macroeconomic observers was that the economy clearly suffered from too slow growth in aggregate demand (spending by households, businesses, and governments) and that the key task facing the Federal Reserve was to boost this demand growth. Since the end of 2016, however, there has been a growing belief that the U.S. economy is either at, or very near, full employment, and that the Fed can and should return to its pre−Great Recession task of guarding against outbreaks of above-target inflation rather than continuing to try to push unemployment lower. Essentially, this congealing conventional wisdom of the past 18 months constitutes a return to the pre-2008 view of how macroeconomic policy should be conducted.

From roughly the late 1970s to just before the Great Recession, the predominant view among economists was that the hard part of the Fed’s job was restraining aggregate demand so that it did not continually run ahead of growth in the economy’s productive capacity and thereby cause inflation to accelerate. It was acknowledged that recessions did occasionally happen, and that during these recessions the Fed should lower interest rates to spur demand growth. But generally it was thought that recessions ended quickly (either on their own or prodded by Fed actions) and that within a year or two of the beginning of a recession, the Fed should focus once again on making sure that aggregate demand did not race ahead of productive capacity.

However, this pre-2008 view was likely wrong then and is even more likely to be wrong now. For far too many of the years between 1979 and 2007, the Fed tolerated levels of unemployment that were higher than necessary to keep inflation in check. Figure A shows estimates of the natural rate of unemployment (sometimes referred to as the nonaccelerating inflation rate of unemployment, or NAIRU) from the Congressional Budget Office over those years. The NAIRU is an estimate of the lowest unemployment rate consistent with nonaccelerating inflation. One could certainly argue that many estimates of the NAIRU are too conservative in that lower unemployment rates might not in fact have spurred accelerating inflation, but even judged by official NAIRU yardsticks the Fed’s monetary policy has been too contractionary in recent decades. Between 1949 and 1979, the cumulative difference between the actual unemployment rate and estimates of the NAIRU was negative 15.3 percentage points—meaning that on average actual unemployment sat below the estimated NAIRU. In contrast, between 1979 and 2017 the cumulative difference was positive 35.7 percentage points, meaning that actual unemployment was persistently above the estimated NAIRU. This was not just driven by the crisis of the Great Recession. Between 1979 and 2007 the cumulative difference was positive 15.5 percentage points.

Figure A

This excess unemployment since 1979 is a key reason why those years were characterized by a growing divergence between growth in economywide productivity (the average amount of income generated in an hour of work in the economy) and the hourly pay of typical workers. Figure B shows hourly pay growth (including benefits) for production and nonsupervisory workers (a group consisting of roughly 80 percent of the private-sector workforce) and economywide net productivity over two periods, 1948–1979 and 1979–2016. In the 1948–1979 period, when unemployment was kept consistently below the NAIRU, pay and productivity rose essentially in lockstep. In the latter period, as unemployment was kept consistently above the NAIRU, pay for typical workers flattened out significantly and began lagging far behind productivity, even as the growth rate of productivity slowed.

Figure B

The fruits of full employment: Fast and equal wage growth in the late 1990s

While the Fed and other macroeconomic policymakers tolerated too high unemployment for most the time between 1979 and 2007, there was in those years a brief experiment of tolerating very low rates of unemployment for an extended period of time. The unemployment rate fell rapidly in the late 1990s, far below the contemporaneous estimates of the NAIRU (which hovered around 5.3 percent in 1995). During these years, the Fed kept the effective federal funds rate (their main policy tool) roughly stable even as unemployment fell beneath these NAIRU estimates. By the end of 2000, the unemployment rate averaged 4.1 percent over two full years and actually sat beneath 4 percent for five months in that span of time. The result was not inflation rapidly accelerating out of the Fed’s comfort zone; instead, it was the only period of strong and across-the-board wage growth in a generation. Figure C shows average annual wage growth for selected parts of the wage distribution between 1996 and 2001, compared with average annual wage growth for all other years between 1979 and 2017. Between 1996 and 2001, average annual wage growth for the 20th, 50th, and 95th percentiles of the wage distribution (corresponding to low-, middle- and high-wage workers) was 1.8, 1.7, and 2.0 percent, respectively. But in every other year between 1979 and 2017 aside from these high-pressure labor markets of 1996 to 2001, average annual wage growth for the 20th, 50th, and 95th percentiles was -0.2, zero, and 1.0 percent, respectively.

Figure C

One key fact about the late 1990s and early 2000s expansion is that it did not end because accelerating inflation forced the Federal Reserve to raise interest rates. This specter of tolerating too low unemployment and thereby letting the inflation genie out of the bottle to damaging effect is a prime reason why the Fed today is being urged by many to raise interest rates. But this last historical experience with persistently tight labor markets did not end because its own inflationary momentum needed to be stopped; instead, it ended when the stock market bubble that characterized these years burst.

Some have claimed that the tight labor markets of the late 1990s were inherently unsustainable because they were driven by a bubble in the stock market that led to rapid growth in aggregate demand. However, this argument confuses an unsustainable source of aggregate demand growth with an unsustainable pace of growth. Because the late 1990s expansion did not end in response to an outbreak of inflation, we know that there was nothing unsustainable about the low unemployment or the pace of aggregate demand growth that characterized this period.

It is true that the source of this demand growth (a stock market bubble) was unsustainable, but the lesson policymakers should glean from this is that future efforts to maintain tight labor markets need to rest on firmer foundations. Aggregate demand growth fueled by consistent and equitable wage increases, or public investments, or a reduction in the nation’s chronic trade deficit, would all provide a sustainable foundation for tight labor markets in the future if policymakers could achieve them.

Being a full-employment hawk means looking beyond the unemployment rate

By far the most persuasive bit of evidence that the U.S. economy is presently near full employment is that unemployment has averaged 4.0 percent so far in 2018 (as of July). This is quite low by historical standards. But “quite low by historical standards” is a thin analytical reed to base policy decisions on. How do we know, for example, that 4.0 percent, or even 3.0 percent, defines the lower limit of unemployment simply because it’s close to the lowest point unemployment has reached in recent decades? Couldn’t it instead be the case that recent decades have simply seen persistently too high unemployment?

This second interpretation gains plausibility even just from examining other measures of the quantity side of the labor market, which indicate that substantially more slack may remain. Most notably, the share of “prime-age” (25- to 54-year-old) adults with a job remains stubbornly below pre−Great Recession peaks and well below the peaks reached during the tight labor markets of the late 1990s and early 2000s, as shown in Figure D.

Figure D

But comparisons of quantity-side measures of the labor market are largely irrelevant to the central question of whether or not the economy is at genuine full employment. The definition of full employment is wage and price growth that are near the Fed’s target levels and are not sustainably accelerating. The entire rationale for making monetary policy more contractionary when the economy hits full employment is to slow economic growth and to soften labor markets (raise unemployment) in order to keep wage and price inflation from accelerating. But as long as price inflation is running below the Fed’s target rate, there is little rationale for raising interest rates to slow the economy. This is particularly true when inflation is below the Fed’s target and shows little to no upward trend, as is the case in mid-2018.

The inability to make strong inferences about just where full employment lies is not new. In past work, for example, Staiger, Stock, and Watson (1997) have shown that, depending on the inflation indicator used, the natural rate of unemployment (essentially the NAIRU) could be anywhere from 2.9 to 8.3 percent. Their estimates are a little more precise using the core personal consumption expenditure (PCE) deflator, but still range from 4.1 to 6.7 percent in a given year. In more recent work, the Council of Economic Advisers (2016) could not even rule out zero as the natural rate of unemployment.

Crucially, the type of inflation the Fed is hoping to restrain by reducing the growth rate of aggregate demand is inflation that stems from excess wage growth. Inflation that results from, for example, a supply-side shock like rising oil prices should not generally be met with a contractionary monetary policy response. This makes the pace of wage growth crucial for determining whether the economy is clearly at (or rapidly approaching) full employment.

Bivens (2015b) notes that the pace of nominal wage growth consistent with the Fed’s price inflation target is simply the sum of trend potential productivity growth and the Fed’s inflation target. This means that if trend growth in potential productivity is 1.5 percent and the Fed is looking to hit a 2.0 percent price inflation target, then nominal wage growth should be 3.5 percent.1

Importantly, this nominal wage target is also consistent with a stable labor share of overall income. Should we desire to raise the labor share from its depressed state following a recession, a period of above-trend nominal wage growth would be necessary. The fact that the labor share of income remains quite depressed in mid-2018 relative to its pre−Great Recession level bolsters our case that the Fed should tolerate a period of above-target price inflation. In fact, nominal wage growth of nearly 4.0 percent might be needed for a number of years in order for the labor share of income to return to pre-recession levels.2

Recent years should have driven home the lesson that quantity-side measures of labor market slack need to be supplemented with wage measures. This is because of the breakdown of forecasts stemming from wage Phillips curves over this time. A wage Phillips curve plots the percentage change in nominal wages against the level of unemployment. It is commonly accepted that at very high levels of unemployment the wage Phillips curve may be nearly flat, meaning that a movement from, say, 10.0 to 9.0 percent unemployment may not lead to any detectable change in wage growth at all. One possible explanation for the failure of wage growth to respond to falling unemployment even in the past few years is that the economy remains on the flat portion of the wage Phillips curve. If this is the case—if the wage Phillips curve is indeed quite flat even at 4.1 percent unemployment—then the rate of unemployment consistent with a price inflation rate of 2.0 percent (the Fed’s target) is much lower than conventional estimates of the natural rate of unemployment would indicate.

An often-cited rough estimate of the Phillips curve slope (once unemployment rates go low enough to move off of the flat portion of the curve) is 0.5 percent.3 The year-over-year change in the PCE deflator, a commonly watched measure of price inflation, was 1.5 percent in the first quarter of 2018. If the trend growth of potential productivity is 1.5 percent, then this implies that nominal wage growth consistent with stable price inflation would be 3.0 percent. Yet hourly wages for all employees had risen only 2.7 percent over the previous 12 months as of March 2018. This implies a cyclical drag on wages of roughly 0.3 percentage points. A 0.5 percent Phillips curve slope would imply that the rate of unemployment consistent with no cyclical drag on wage growth would be lower than 3.5 percent—far below any conventional estimate of the natural rate of unemployment.4

We can sum up this evidence as follows: It is not true that the U.S. economy has already unambiguously reached full employment. In fact, much data is consistent with the view that labor market slack remains. Further, this evidence indicates that even if the U.S. were at full employment, there is little to no evidence that the economy is “overheating,” that is, that we are seeing wage and price inflation accelerate above policymakers’ comfortable targets. Finally, full employment is not something that policymakers should see as needing to be stomped out by raising interest rates. Instead it should be seen as a target for where policymakers want the economy to remain, and it does not follow that hitting full employment automatically implies that macroeconomic policy should become substantially more contractionary.5 This is especially true given the large potential benefits that the economy may be able to realize with an extended period of labor market tightness. We turn to these potential benefits in the next section.

The benefits of extended labor market tightness

The above sections provide some analytical markers for when policymakers should start to worry about the potential costs of economic overheating leading to inflation. This section documents the benefits of avoiding premature contraction. Given how unequal economic outcomes in the United States are across racial and class lines, a crucial part of this story is how sharply progressive the benefits of extended tight labor markets are. In this section, we first look at benefits for growth across the distribution of hourly wages, and we then turn to measures of labor market disparities across racial groups.

Extended labor market tightness leads to faster hourly wage growth for low- and moderate-wage workers

We note above that the period between 1979 and 2017 saw unemployment consistently above estimates of the NAIRU and that it also saw a pronounced decoupling of growth between economywide productivity and the wages of nonsupervisory workers in the United States. In other work it has been noted that this excess of growth of productivity over wages for typical American workers is the root cause of the large increase in income inequality in recent decades.6 This excess productivity growth had to go somewhere if it was not showing up in the paychecks of the bottom 80 percent of American workers. Where it largely went was to the top 1 percent of American households, in the form of greater concentration of labor incomes (think of the rising ratio of CEO pay to the pay of typical workers7) and in the form of profits rising at the expense of wages. This pattern of evidence is consistent with a hypothesis that for wage growth of low- and middle-wage workers to approach the rates of productivity growth lower rates of unemployment than for equivalent wage growth for the top 10 (and 5 and 1) percent are required.

A range of authors have documented that hourly wages for low- and middle-income workers are more sensitive to changes in unemployment rates than are wages for higher-paid workers.8 In this section, we extend and update this work using a panel of state-level data from 1979 to 2016 to look at the effect of labor market tightness on hourly wage changes for various wage deciles. Because measures of labor market tightness have varied substantially more across states in the past four decades than they have varied over time at the national level, this panel data set should provide more reliable and precise estimates of the effect of unemployment on wage growth. The regression specification is in the appendix.

Key findings are summarized in Figure E, which shows the percentage-point change in the growth rate of inflation-adjusted wages following a 1-percentage-point increase in the state-specific unemployment rate. The blocks of bars show results for the 10th, 50th, and 90th percentiles of wages, corresponding to low-, middle-, and high-wage workers. Besides showing correlations between wage growth and unemployment, the figure also shows the wage changes associated with 1-percentage-point increases in the employment-to-population ratio (EPOP) for the population ages 16 and up and the EPOP for the population between the ages of 25 and 54 (the “prime-age” EPOP).

Figure E

The results indicate that a 1-percentage-point drop in unemployment results in annual wage growth for workers at the 10th percentile of the wage distribution that is 0.5 percentage points faster. For example, if annual real wage growth is at 1.0 percent, then a 1-percentage-point fall in unemployment would result in annual real wage growth rising to 1.5 percent. For workers near the median of the wage distribution, wage growth is faster by 0.4 percentage points following a 1-percentage-point decline in the unemployment rate. For workers at the 90th percentile of the wage distribution, wage growth is faster by 0.3 percentage points following a 1-percentage-point decline in the unemployment rate.

(Note that the estimated changes are symmetric: an equal and opposite change in wage growth occurs in response to a 1-percentage-point rise in unemployment: at the 10th percentile, workers see wage growth slow by 0.5 percentage points; wage growth slows by 0.4 percentage points at the median and by 0.3 percentage points at the 90th percentile.)

For a 1-percentage-point increase in the overall or prime-age EPOP, the results are similar (note that while a falling unemployment rate denotes a tightening labor market, a rising EPOP implies the same). Each percentage-point increase in the prime-age EPOP is associated with annual wage growth that is faster by about 0.2 percentage points for workers at the 10th and 50 percentiles, and wage growth that is about 0.1 percentage points faster for workers at the 90th percentile. In short, the well-established pattern seen in national time-series data—that low- and middle-wage workers see their wage growth respond more robustly than high-wage workers do to changes in the unemployment rate—is supported in our state-level panel data that examines a much wider range of unemployment experiences and with much more statistical power than the national data can provide. However, in our state-level data, the differences represented in this pattern are not uniformly statistically significant. That is, the coefficient representing (say) the effect of lower unemployment on wage growth for workers at the 10th percentile is higher than for workers at the 50th percentile, but the differences are not statistically significant. In the studies that use national data, these differences are often statistically significant.

To examine why the results from our state panel differ from prior studies focusing on national data, we look separately at the years before and after the Great Recession (using 2007 as our cutoff). The results of this split sample are shown in Figure F. There are large and significantly different values of the coefficient on unemployment for all three wage percentiles we examine. Essentially, the sensitivity of wage growth to unemployment was significantly larger (both economically and statistically) in the pre-2008 period for all percentiles, compared with the 2008–2016 period.

The sensitivity of wage growth to the prime-age EPOP is also larger in the pre-2008 period than in the 2008–2016 period, and these differences between periods are statistically different from zero at the 5 percent level for both the 10th percentile and the median.

Figure F

As seen in Figure F, a 1-percentage-point increase in unemployment during the 2008–2016 period led to smaller decrease in annual real wage growth relative to the pre-2008 period. This might strike some as a strange finding—was workers’ wage growth really better insulated against rising unemployment during the Great Recession than before?

Strangely, the answer might be (a heavily qualified) “yes.” Essentially, given the rise in unemployment during the Great Recession, pre-2008 statistical relationships between unemployment and wage growth would have predicted even worse wage growth in the face of high unemployment than what has actually been seen in the past 10 years. This actually jibes well with another key finding about labor markets during and after the Great Recession: they were characterized by clear evidence of downward nominal wage rigidity (DNWR). DNWR essentially means that workers strongly prefer not having their nominal pay cut, and (perhaps the real puzzle) employers tend to respect this preference, even during steep recessions.

For example, research has shown that both workers and employers prefer layoffs to nominal wage cuts as a strategy for absorbing reductions in labor demand.9 As unemployment rose sharply post-2008, statistical models were predicting that nominal wage growth would indeed become negative for most of the American workforce. But numerous authors have shown that instead of widespread nominal wage cuts, there was a large piling up of nominal wage changes at exactly zero year-to-year (which, combined with low inflation, explains the minimal losses to real wage growth).10 For example, Daly and Hobijin (2014) show that in 2006 just 12 percent of workers saw zero nominal wage growth, but in the recession-damaged labor markets of 2011, this share rose to 16 percent.

Further, there was a pronounced decline in economywide productivity following the onset of the Great Recession (and, indeed, perhaps this decline began a few years prior to the recession). In a limited number of years in the post-2008 period, precisely separating out the effect of falling productivity from rising unemployment on nominal wage growth is asking a lot of our model. As the economy and labor spends a number of years in more normal states, we would not be surprised if pre-2008 statistical relationships between wage growth and unemployment reasserted themselves, in part because we anticipate a rebound of productivity growth to more normal ranges.11

The unemployment rate consistent with zero real wage growth

Turning briefly to national data on median wage growth and unemployment allows us to examine a concept first identified by Katz and Krueger (1999)—the unemployment rate consistent with zero expected real wage growth (URCZWG). Essentially, this is the unemployment rate required for median-wage earners to see their nominal wage growth just keep up with price inflation. In their work, Katz and Krueger identify a small decline in this measure after 1988. One way to interpret this finding is that influences pushing down on wages besides unemployment intensified after 1988 in their data (or, alternatively, influences pushing up on wages got weaker).

Using national data, we examine the relationship between unemployment and wage growth using Katz and Krueger’s (1999) regression specification. Figure G shows the unemployment rate consistent with zero wage growth for four time periods: 1973−1978, 1979−1988, 1989−2000, and 2001−2007. Our results show a large decline in this measure in the 1989−2000 period and then again in the 2001−2007 period. Our interpretation of this finding is that median-wage workers needed lower and lower unemployment (that is, increasingly higher-pressure labor markets) just to not see declines in the purchasing power of their hourly wages.

Figure G

This finding is consistent with a theory that structural and institutional supports for workers’ bargaining power have been dismantled over time without seeing new supports put into place. Obvious examples include the decline of unions and collective bargaining and an erosion of the purchasing power of the federal minimum wage. This suggest two parallel policy tracks for those wanting to boost wage growth: (1) restoring these structural and institutional supports for workers’ bargaining power and (2) readjusting the targeted unemployment rate downward over time, as needed, so that it continues to be low enough to promote adequate wage growth.12

High-pressure labor markets disproportionately benefit African American workers

It has become conventional wisdom that the ratio of unemployment rates for African Americans to whites is 2:1, in good times or bad. This conventional wisdom, even if true, has often led to overly pessimistic policy conclusions. For example, the implicit claim has been made that this constant 2:1 ratio of unemployment rates means that there’s little the Federal Reserve (or other macroeconomic policymakers) can do about race-based disparities in the economy (Appelbaum 2015). But the opposite is actually true: a constant 2:1 ratio means that African Americans are hurt worse by aggregate economic slumps but benefit more from reductions in overall unemployment (Bivens 2015a). So any actions that improve unemployment overall will disproportionately benefit African American workers.

Given the inherent uncertainty about the overall unemployment rate that constitutes genuine full employment, and given the racist history of the United States, this disproportionate benefit to low unemployment accruing to African Americans should be weighed heavily indeed by policymakers. Remember, the central job of the Federal Reserve is to balance the benefits of lower unemployment against the risk that this lower unemployment will spark inflation as empowered workers demand and receive excessive wage gains. If policymakers knew with certainty what specific unemployment rate would spark accelerating inflation, their jobs would be easy. But recent decades have highlighted with great force that policymakers do not know this “natural” rate of unemployment ex ante. In fact, the uncertainty is so large that in 2016 the Economic Report of the President highlighted that, in statistical terms, a NAIRU of zero could not be ruled out (CEA 2016).

So policymakers are making decisions regarding this tradeoff between lower unemployment and potentially accelerating inflation in an environment of uncertainty. Their choice about whether to aggressively probe how low the unemployment rate can go depends in large part on their estimation of the benefits of low unemployment. The most obvious benefit is that more people are working when unemployment is lower. Another benefit is the faster wage growth that stems from lower unemployment, as we have documented above.

The disproportionate benefit to communities of color stemming from low overall unemployment should also be something that policymakers explicitly consider when evaluating the tradeoffs. Loury (2007) frames a series of questions around this precise subject:

Would it be legitimate to tolerate a somewhat greater chance of inflation while maintaining a strong demand for labor because doing so also manages to hold the unemployment rate of black youth at humane levels for the first time in a half-century? Can we reckon that this is a good policy because it contributes to overcoming racial stigma, draws blacks more fully into the mainstream of society, and permits them to earn the respect of their fellow citizens? (Here I mean to suggest that, but for this racial benefit, a different decision might be taken.) In other words, can we explicitly count as a benefit to society what we calculate to be the racially progressive consequences (reducing black economic marginality) of what is a race-blind action (electing to take a greater risk of inflation)?

It seems to us that the answer is clearly “yes”; in an environment of uncertainty, where potential benefits and costs are being weighed, ignoring potential benefits stemming from reducing race-based economic disparities seems wrong.

High-pressure labor markets narrow employment (EPOP) gaps, even while the black–white unemployment ratio remains stubbornly at 2:1

While the rationale for taking racial employment gaps into account when making decisions regarding monetary policy seems strong to us even in the context of an unchanging ratio of African American to white unemployment rates, it would unquestionably be even stronger if high-pressure labor markets managed to reduce these ratios.

The hope that high-pressure labor markets might reduce the ratio of African American to white unemployment rates does not seem ridiculous on its face. Tight labor markets have been shown to erode other unemployment gaps likely caused by employer discrimination. Consider discrimination against the long-term unemployed. Starting in 2010, the long-term unemployment rate became elevated relative to its historic relationship with other labor force indicators. Suggestive studies argue that part of this elevated long-term unemployment rate was the result of employers engaging in discrimination against workers who had been out of work for long stretches, with these employers using the fallacious reasoning that long-term unemployment status is a marker of low productivity. Overall job growth during this time was constrained by weak demand, so there were more workers than job openings. This weak overall growth, combined with employers’ discriminatory practices, too often meant that long-term unemployed workers were sent to the back of the job queue. However, as the labor market recovery continued into 2012 and beyond, the long-term unemployment rate actually began falling faster than other labor market indicators would have suggested. The need to hire began to override employers’ dysfunctional strategy of prioritizing whom to hire.

This claim that labor market tightness can stop employers from discriminating against the long-term unemployed has a good pedigree: Olivier Blanchard (former chief economist of the International Monetary Fund [IMF]) notes in a 1996 paper that models predict that “in tight labor markets, firms will not discriminate against the long term unemployed. But in more depressed labor markets, they will” (Blanchard 1996). This prediction is largely borne out in empirical work by Ghayad (2013) and Kroft, Lange, and Notowidigdo (2013).

Figure H probes the question of racial (un)employment gaps and labor market tightness using our state-by-year panel data. The first set of bars shows the change in group-specific unemployment accompanying a given change in the overall unemployment rate. The standard finding that the percentage change in unemployment for African American and white workers is very similar for a given change in overall unemployment is confirmed. For example, the results show that if the overall unemployment rate rises by 10 percent (say, from 5.0 to 5.5 percent), then white unemployment rises by 10 percent, while African American unemployment rises by 9.1 percent. Because each group-specific unemployment rate changes by nearly the same amount in response to overall changes, pre-existing gaps are not closed by tighter labor markets. In fact, if there is any difference at all, it seems that African American unemployment rates respond slightly less (though not to a statistically significant degree) to changing overall unemployment, implying that the racial unemployment gap should be expected to rise slightly during economic booms (as white unemployment drops at a slightly faster rate than black unemployment). We also find no change in labor force participation gaps when the overall labor force participation rate changes.

Figure H

However, when we turn to looking at the EPOPs, something interesting appears in the data. When we look at the EPOPs for ages 16 and up and at the prime-age EPOPs (for adults ages 25–54), we find that the African American rates respond much more strongly than the white rates (and to a statistically significant degree) to changes in the respective overall rates. For each 10-percent increase in the overall age-16+ EPOP (say, from 60 to 66 percent), the white age-16+ EPOP rises by 10.2 percent, but the African American age-16+ EPOP rises by 14.3 percent. For prime-age EPOPs, the changes in response to a 10-percent rise in the overall prime-age EPOP are 9.2 percent for the white prime-age EPOP and 16.2 percent for the African American prime-age EPOP. This means that for prime-age adults, the response of the African American EPOP is about 1.7 times as large as for whites. This implies that during tight labor markets as (say) the overall prime-age EPOP rises, the gap between the African American and white prime-age EPOPs should close rapidly.

Extended labor market tightness reduces racial hours gaps

The implications of labor market outcomes for living standards does not hinge solely on whether an individual works; they also hinge on how many hours that individual works over the course of a year. Similar to our employment-based results, we also find that tighter labor markets are associated with narrowing racial gaps in total hours worked per year by households.13

Figure I shows the effect of a 1-percentage-point change in an employment indicator (unemployment, employment, or labor force participation) on the average number of hours worked by households over the course of a year. The results are again presented by race, where we define the race of the household to be the race of the head of household. Here we find clear evidence that high-pressure labor markets boost average hours worked and that the responsiveness of African American households is greater than it is for white households.

Figure I

The results in the figure indicate that a 1-percentage-point decline in the overall unemployment rate is associated with a 1.1 percent increase in average hours worked by white workers and a 2.7 percent increase in hours worked by African American workers. For EPOPs, each 1-percentage-point increase in the overall age-16+ EPOP is associated with a 1.3 percent increase in hours worked by white workers and a 2.3 percent increase in hours worked by African American workers; each 1-percentage-point increase in the overall prime-age EPOP is associated with a 1.0 percent increase in hours worked by white workers and a 2.2 percent increase in hours worked by African American workers. For labor force participation, each 1-percentage-point increase in the overall rate is associated with a 1.4 percent increase in white hours worked and a 2.1 percent increase in African American hours worked. All of these increases are statistically different from zero at the 1 percent level.

The gray bars in Figure I show the percentage-point differences between changes in annual hours worked for African American households compared with changes for white households for each employment indicator. These differences are statistically significant at the 1 percent level for unemployment, the overall EPOP, and the prime-age EPOP. In the case of labor force participation, the difference is statistically significant only at the 10 percent level. As shown in detail in Appendix Table 4, these results are robust to additional controls for state-specific trends over the sample time period.

In Appendix Table 5, we show the results of running similar specifications for average weekly hours worked by individual workers (rather than average total household hours over an entire year), this time conditional on the individual workers having worked a positive number of hours. The differential responses across races for these outcomes are not always statistically different from each other. When controlling for state-specific linear trends, the differences are in a similar direction as above and are statistically significant at the 10 percent level. Because we cannot be positive that including linear state-specific trends is the best way to control for state-specific developments, we consider these results using average weekly hours worked to be suggestive rather than dispositive. One possible reason for the difference in results between total annual hours worked by households and weekly hours worked by individuals (conditional on those individuals having worked) is that much of the differential hours gains of tighter labor markets are attributable to people newly entering or re-entering the labor market. Further probing of these results is a potential future research endeavor.

Conclusion

The benefits of an extended period of high pressure in the labor market are large and progressive. The failure to aggressively pursue high-pressure labor markets in most years since the late 1970s seems highly implicated in the rise of inequality and the radical slowdown in wage growth for typical workers since then. The econometric evidence strongly suggests that low- and moderate-wage workers are more in need of tight labor markets to see any wage gains at all than their peers higher up the wage scale.

Further, economic observers have become too pessimistic about the benefits of high-pressure labor markets for reducing racial disparities in labor market performance. The relatively stubborn and consistently high ratio of African American to white unemployment has led too many to give up on tight labor markets as a plausible solution to easing racial disparities in the job market. In this report, we use a panel data set of U.S. states since 1979 to examine a broader set of measures—including employment-to-population ratios, labor force participation rates, and average hours worked. Many of these broader measures of labor market performance demonstrate that high-pressure labor markets can narrow gaps in some employment outcomes between African American and white workers.

The fact that the aggressive pursuit of full employment attacks both class-based inequalities and racial disparities in labor market outcomes means that it should be a high-priority agenda item for economic progressives. Further, it means that macroeconomic policymakers—particularly the Federal Reserve—should place a very high weight on the benefits of full employment when weighing the costs and benefits of policies aimed at aggressively pushing toward high-pressure labor markets.

Appendix

The tables in this appendix contain the full regression results from the figures in the text based on state-level, annual data from the Current Population Survey (CPS).

Appendix Table 1 is from a regression of log of a race-specific labor market outcome on the log of the same overall labor market outcome (for all races), for the 1979−2016 period. The labor market outcomes are the annual, state-level unemployment rate, employment-to-population ratio, prime-age employment-to-population ratio, and labor force participation rate. The regressions also include state and year fixed effects and are weighted by state-level overall annual population estimates from the CPS. Standard errors are clustered at the state level.

Appendix Table 1

Appendix Table 2 is a regression of the annual percent change of the 10th, 50th, or 90th real wage percentile on the annual mean of a labor market outcome, for the 1980–2016 period. Here, labor market outcomes are scaled so that a 1.0-unit change is a 1-percentage-point change. For example, a 1-percentage-point increase in unemployment is associated with a -0.49-percent decline in the annual rate of growth of the 10th-percentile real wage. The regressions also include state and year fixed effects, control for the annual percent change in the real minimum wage, and are weighted by state-level overall annual population estimates from the CPS. Standard errors are clustered at the state level.

Appendix Table 2

The coefficients in Appendix Table 3 are from the same regression as in Appendix Table 2 except that we interact the annual mean of the labor market outcome with an indicator for the 1980–2007 period. The table shows the coefficients from the interacted terms as well as the difference between these two coefficients.

Appendix Table 3

In Appendix Table 4, the “White” and “Black” columns contain the results of a regression of the log of the race-specific annual mean of annual household hours worked on the annual mean of a labor market outcome. Under the column “Difference,” the dependent variable is the log of the ratio of black-to-white hours worked. Here, the labor market outcomes are scaled so that a 1.0-unit change is a 1-percentage-point change and, additionally, the coefficients are multiplied by 100 to approximate percent changes. For example, a 1-percentage-point increase in unemployment is associated with about 1.1 to 1.2 percent increase in white weekly hours worked, depending on the specification. The top panel includes state and year fixed effects and the bottom panel additionally includes state-specific linear trends. The regressions are weighted by state-level overall annual population estimates from the CPS. Standard errors are clustered at the state level. Appendix Table 5 is the same as the previous table, except that the outcome is individual weekly hours worked, conditional on having worked positive hours.

Appendix Table 4

Appendix Table 5

Endnotes

1. While productivity growth has been below 1 percent in several recent years, there remains little reason to believe that the long-run trend growth for potential productivity has dropped much below 1.5 percent. Key forecasters—both private and public—continue to use 1.5 percent as their estimate of potential productivity growth. These forecasts, as well as evidence indicating that productivity would likely rebound if wage growth picked up, can be found in Bivens 2017.

2. Bivens 2015b provides a number of scenarios for how fast wage growth in excess of nominal wage targets would claw back the loss in labor’s share over the post-2009 recovery.

3. See Leduc and Wilson 2017 for an overview of estimates of the wage Phillips curve. They find that this 0.5 estimate describes the pre-2008 period well, but (consistent with our argument) that it may well have declined post-2008.

4. When this calculation was done in Bivens 2015b, the implied natural rate of unemployment was lower. However, since then, other potential margins of labor market slack—like depressed labor force participation—have firmed up. This means future labor market tightening is more likely to show up as reduced unemployment than as increased labor force participation.

5. Indeed, much of the current pressure being put on the Fed to raise rates stems explicitly from strong claims that the U.S. economy today is clearly operating at more than full employment, and that interest rate increases are needed to slow the economy and move it on a glide-path back to the slightly higher unemployment rates that constitute the sustainable natural rate of unemployment. Needless to say, if the evidence that today’s unemployment rate unambiguously represents full employment is lacking (as we think it is), evidence that today’s unemployment rate unambiguously represents overfull employment is even more lacking.

6. See Bivens et al. 2014 for documentation of this claim.

7. See Mishel and Schieder 2018.

8. See, for example, Katz and Krueger 1999, Mishel et al. 2012, and Wilson 2015.

9. See Bewley 2002 for research on employer and worker attitudes toward nominal wage cuts.

10. It is important to note that one reason why employers may respect this preference against nominal wage cuts is that there are other margins along which these employers can cut compensation. For example, reducing contributions to retirement benefits, or increasing co-pays or employee contributions for health insurance, are effectively compensation cuts, but allow workers to not face cuts in nominal wages.

11. See Bivens 2017 for the case that productivity growth is likely to rebound in coming years.

12. When we tested the inclusion of the post-2008 period, the estimated URCZWG actually increased. This could in theory imply that these structural pressures relented during and after the Great Recession. But the much more plausible theory is one related to our regression findings above: downward nominal wage rigidity likely led to a breakdown in the correlation between wage growth and unemployment as the latter reached historically high levels. It will be interesting to see what happens to the estimated URCZWG when the full business cycle following the Great Recession has run its course.

13. Because a tighter labor market reduces the actual gap in average hours worked between white and African American households, this means (by definition) that the ratio of these hours is also decreasing.

References

Appelbaum, Binyamin. 2015. “Black Jobless Rates Remain High, but Fed Can Only Do So Much to Help.” New York Times, March 3, 2015.

Bewley, Truman. 2002. Why Wages Don’t Fall during a Recession. Cambridge, Mass.: Harvard Univ. Press.

Bivens, Josh. 2015a. “The Fed’s ‘Hammer’ Can Be Used to Great Effect to Improve Prospects for Minority Workers.” Working Economics (Economic Policy Institute blog), March 4, 2015.

Bivens, Josh. 2015b. A Vital Dashboard Indicator for Monetary Policy: Nominal Wage Targets. Report for the Policy Futures Project at the Center for Budget and Policy Priorities, June 2015.

Bivens, Josh. 2017. A ‘High-Pressure’ Labor Market Can Help Boost Productivity and Provide Even More ‘Room to Run’ for the Recovery. Economic Policy Institute, March 2017.

Bivens, Josh, Lawrence Mishel, Elise Gould, and Heidi Shierholz. 2014. Raising America’s Pay: Why It’s Our Central Economic Challenge. Economic Policy Institute, June 2014.

Blanchard, Olivier. 1996. “The Plight of the Long-Term Unemployed.” Working paper, MIT Economics Department.

Bureau of Economic Analysis, National Income and Product Accounts (BEA-NIPA). Various years. National Income and Product Accounts Tables [data tables].

Bureau of Labor Statistics (BLS). 2018. “Series ID: LNS14000000. (Seas) Unemployment Rate.” Public data series aggregated from basic monthly CPS microdata. Series report accessed August 2018.

Bureau of Labor Statistics, Consumer Price Indexes program (BLS-CPI). Various years. All Urban Consumers: Chained Consumer Price Index (CPI) [database].

Bureau of Labor Statistics, Current Employment Statistics program (BLS-CES). Various years. Employment, Hours and Earnings–National [database].

Bureau of Labor Statistics, Current Population Survey (BLS-CPS). Various years. Public data series aggregated from basic monthly CPS microdata and accessed through the Labor Force Statistics database and series reports.

Bureau of Labor Statistics, Labor Productivity and Costs program (BLS-LPC). Various years. Major Sector Productivity and Costs and Industry Productivity and Costs [databases].

Bureau of Labor Statistics, National Compensation Survey–Employment Costs Trends (BLS-NCS). Various years. Employer Costs for Employee Compensation [series of economic news releases].

Congressional Budget Office (CBO). 2018. “Online Data on Potential Output and Its Underlying Inputs” [downloadable Excel file]. Supplemental data file accompanying the report The Budget and Economic Outlook: 2018–2018 (April 2018).

Council of Economic Advisers (CEA). 2016. “The Year in Review and the Years Ahead.” In Economic Report of the President, 51–117.

Daly, Mary, and Bart Hobijin. 2014. “Downward Nominal Wage Rigidities Bend the Phillips Curve.” Journal of Money, Credit and Banking 46, no. S2.

Economic Policy Institute (EPI). 2018. State of Working America Data Library.

Ghayad, Rand. 2013. “Why Is It So Hard for Job-Seekers with Valuable Experience to Find Work? Evidence from a Field Experiment.” Northeastern Working Paper.

Katz, Lawrence, and Alan Krueger. 1999. “The High-Pressure Labor Market of the 1990s.” Brookings Papers on Economic Activity 1 (1999).

Kroft, Kory, Fabian Lange, and Matthew J. Notowidigdo. 2013. “Duration Dependence and Labor Market Conditions: Evidence from a Field Experiment.” Quarterly Journal of Economics 128, no. 3: 1123−1167.

Leduc, Sylvain, and Daniel Wilson. 2017. “Has the Wage Phillips Curve Gone Dormant?” Federal Reserve Bank of San Francisco Economic Letter.

Loury, Glenn. 2007. “Trans-Generational Justice: Compensatory vs. Interpretative Approaches,” in Reparations: Interdisciplinary Inquiries, edited by Jon Miller and Rahul Kumar. New York: Oxford Univ. Press.

Mishel, Lawrence, Josh Bivens, Elise Gould, and Heidi Shierholz. 2012. The State of Working America, 12th Edition. An Economic Policy Institute book. Ithaca, N.Y.: Cornell Univ. Press.

Mishel, Lawrence, and Jessica Schieder. 2018. CEO Compensation Surged in 2017. Economic Policy Institute, August 2018.

Staiger, Douglas, James Stock, and Mark Watson. 1997. “The NAIRU, Unemployment and Monetary Policy.” Journal of Economic Perspectives 11, no. 1: 33–49.

U.S. Census Bureau, Current Population Survey Annual Social and Economic Supplement microdata (U.S. Census Bureau CPS-ASEC). Various years. Survey conducted by the Bureau of the Census for the Bureau of Labor Statistics [machine-readable microdata file].

U.S. Census Bureau, Current Population Survey basic monthly microdata (U.S. Census Bureau CPS basic). Various years. Survey conducted by the Bureau of the Census for the Bureau of Labor Statistics [machine-readable microdata file]. https://thedataweb.rm.census.gov/ftp/cps_ftp.html.

U.S. Census Bureau, Current Population Survey Outgoing Rotation Group microdata (U.S. Census Bureau CPI-ORG). Various years. Survey conducted by the Bureau of the Census for the Bureau of Labor Statistics [machine-readable microdata file]. https://thedataweb.rm.census.gov/ftp/cps_ftp.html.

Wilson, Valerie. 2015. The Impact of Full Employment on African American Employment and Wages. Report for the Full Employment Project at the Center on Budget and Policy Priorities, March 2015.

Two weeks ago, the Tax Policy Subcommittee of the Ways and Means Committee of the U.S. House of Representatives announced a series of hearings on the effects of the Tax Cuts and Jobs Act (TCJA) passed last December. The committee has invited submissions of written testimony from the public.

The authors of this testimony work for the Economic Policy Institute (EPI), the nation’s premier research institute studying the effect of policy on the living standards of low- and middle-income American families. Our view, supported below, is that the TCJA is deeply flawed policy. It will deliver the vast majority of its direct benefits to the richest households; claims that large benefits will trickle down through indirect channels to low- and middle-income families are clearly incorrect. The key points we make in this testimony are:

• There is no compelling evidence yet available to make firm claims about the effect of the TCJA on the American economy. Claims that evidence is already showing large positive effects are based on data cherry-picking and are either innumerate or dishonest.
• The clearest reason why evidence so far shows no compelling effect of the TCJA on the economy is that it’s simply too soon since its passage for a full set of data to be assembled. However, the data that have come in so far have shown no break in trend at all for most economic indicators. There is no basis to claim that data so far show any positive effect of the TCJA.
• The largest and only permanent cut in taxes stemming from the TCJA is a cut to corporate tax rates. Thus the best predictor of the likely effect of the TCJA is what happened after past episodes when corporate tax rates were cut.
• This evidence based on past experience with corporate rates cuts—either in the United States, in international peer countries, or in individual U.S. states—argues strongly that capital investment and pay for most American workers will not noticeably increase due to the TCJA.

Do data since the TCJA passed tell us anything about its effects?

All objective analysis indicates that the direct, first-round benefits of the TCJA will accrue overwhelmingly to richer households when the act is fully phased. The Tax Policy Center (TPC), for example, estimates that the top 1 percent of households will see 83 percent of the total benefits from the law by 2027.1 In the run-up to the passage of the TCJA, proponents often claimed that the law would have indirect effects as that would lead to higher capital investment which in turn would eventually filter through to the economy in the form of higher productivity and pay for most American workers.2 Since its passage, these proponents have not been shy about claiming it has already had beneficial effects for the U.S. economy. The two broad claims made most often are that the TCJA is responsible for an alleged tsunami of wage increases and bonus payments that appeared in its wake, and that it is responsible for a large increase in investments made by American firms in plants, equipment, and research and development. These two strong claims about what has already appeared in the data are clearly false.

With regard to claims about bonuses and wages, there are several points to make. First, there is no evidence that wage growth has materially picked up since the TCJA’s passage. Figure 1 shows nominal wage growth (measured relative to the same month last year) before and after the passage of the TCJA. There is no obvious effect at all.

Figure 1

Second, though there were plenty of anecdotal claims about bonuses being spurred by the TCJA, bonuses are not permanent wage increases, i.e., not the sort of boost that leads to permanently higher wages because productivity has been boosted and because workers have been given the bargaining clout to claim a fair share of this productivity growth. In a sense, focusing on bonuses over permanent wage increases already constitutes an admission that claims made during the run-up to the TCJA’s passage about permanent wage gains were excessively rosy.

Third nearly 40 percent of American workers get bonuses every year.3 So, in any given year tens of millions of stories could’ve been written about the bonuses workers received. No TCJA proponents have mobilized any systematic evidence that bonuses surged because of the tax cut.

Finally, there really was an incentive in the TCJA to provide bonuses in 2017.4 Because labor costs are an expense that can be written off on corporate taxes, these bonuses become more costly when tax rates are low. Because the TCJA slashed the headline corporate tax rate from 35 percent to 21 percent, it made granting bonuses much cheaper in 2017 than in 2018 and all other post-TCJA years. What this means is that even if some increase in bonuses occurred in 2017 because of the TCJA (this remains a big “if”), there is no reason to think such bonuses will recur in the future. In short, if you didn’t get your paycheck bump from the TCJA by New Year’s Eve 2017, you’re likely to not get it at all.

With regard to the corporate investment claim, there is no serious evidence that the TCJA spurred a notable pickup in business investment. In an embarrassingly over-the-top defense of the TCJA made in the Wall Street Journal last month, the director of the Council of Economic Advisers (CEA) for the Trump Administration, Kevin Hassett, made lots of hay over the fact that business investment in the last quarter of 2017 grew 6.3 percent, an improvement over very recent years’ growth. He argued that this represented a radical improvement over an investment-starved recovery during the Obama administration’s tenure. To explain away the inconvenient fact that the TCJA didn’t come into effect until 2018, and hence would have a hard time directly explaining any late-2017 investment surge, Hassett speculated (with no evidence) that firms could predict the future improvements to post-tax profitability that would accompany the law’s effect, and that “when profits go up, capital investment goes up, and wages follow.” 5

However, in recent decades as profits and corporate managers’ salaries rose rapidly, the pay for typical workers didn’t.6 But, setting this aside for one second, even the claim that capital investment responds strongly to higher profits is impossible to square with the level of investment during the Obama administration. (As Hassett scornfully noted, “For perspective, real private nonresidential fixed investment was anemic at the end of the Obama administration.”) It’s hard to square because during these same years when capital investment genuinely was weak, profitability of corporations was historically high.7 In short, it just cannot be the case that low after-tax profitability was a binding constraint on capital investment and economic growth in recent years because profitability was not low, it was high.

Further, it’s worth lingering a bit on the data point Hassett trumpets as proving the TCJA’s effectiveness in spurring investment. He points to the 6.3 percent year-over-year increase in real, private nonresidential investment in the last quarter of 2017. We plot this in Figure 2. Readers can decide for themselves if they think this second-to-last last data point (which is what Hassett is lauding) looks like the TCJA (or anything else) has been a game changer in driving investment. It doesn’t seem like it to us, given that the year-over-year increase in real, private nonresidential investment was substantially higher several times during the Obama recovery that Hassett claims was “anemic.”

Figure 2

The preliminary data for the first quarter of 2018 showed that nonresidential investment growth slowed in the first quarter of 2018 to 6.1 percent. The second round of revisions to this data (another round will come next month) boosted this growth to 6.8 percent—faster than the last quarter of 2017, but still below rates of growth from 2011 and 2014. In short, we do not yet have economy-wide data showing a rapid upsurge of investment due to the TCJA.

While this economy-wide data is quite damning of claims that the TCJA has spurred investment, the law’s proponents have taken some succor in a recent report citing evidence from nonpublic data that tracks investment by S&P 500 firms.8 This report claimed a large pickup in business investment that saw it grow faster in the first quarter of 2018 than at any point since 2011.

But this report should not lead one to think that the official, economy-wide data should be discounted. Besides being nonpublic and hence largely non-replicable, the firms surveyed from the S&P account for well under 10 percent of total private business investment in the U.S. economy.9 As such, even if faster investment occurred in these firms, this more-rapid investment apparently could not pull up the economy-wide numbers that are much more relevant for American living standards.

Further, much investment done by S&P companies is likely done overseas. This does not boost American productivity and does not show up (properly) in U.S. national accounts. It is by now well-known that the TCJA provided incentives for firms to make investments overseas rather than in the United States.10 A much-slower pace of overall American capital investment relative to what S&P firms undertook is consistent with the TCJA doing little-to-nothing to boost American investment, but nudging multinationals to invest offshore. In short, American workers should take no solace at all in S&P firms boosting investments while growth in overall national investments by businesses remains flat.

What past evidence on corporate tax cuts tells us

Again, no side in the debate over the effect of the TCJA should be making large claims about what the data tell us so far; there simply isn’t enough data yet to make a serious evaluation. One can, of course, look at past experience to assess the likely effect of lowering corporate tax rates on investment and wage growth. EPI did that in a number of reports and analyses released during the debate spurred by the introduction of the TCJA.

For example, the Trump administration’s Council of Economic Advisers released a paper last year arguing that cuts in the statutory corporate tax rate would lead to gains in business investment, productivity, and wages.11 We noted in report released shortly thereafter why this was unlikely to be true.12

The simplest reason that cutting corporate taxes will not boost American productivity or wages is that the past history corporate tax cuts in the United States shows no such relationship. Figure 3 shows the top corporate tax rate, productivity growth, and growth in typical workers’ hourly pay since the 1950s. It shows clearly that productivity and pay actually grew more rapidly when tax rates were higher.

Figure 3

The CEA report claimed that the wage-boosting effect of corporate tax cuts was “highly visible in the data” and presented as evidence a graph that showed faster unweighted wage growth over just two years in a set of “low-tax” countries relative to a set of “high-tax” countries. It is deeply puzzling just what this graph was supposed to prove. Most notably, there is no claim that corporate tax policy changed in the years analyzed by the CEA, and that these changes hence could be linked to differing wage outcomes. Even the CEA report itself implicitly confirms that corporate tax policy changes are the correct variable to assess. In a later section of its report, the CEA praises another study for assessing “long-run outcomes…[that should be] thought of as the recurring flow of income after the corporate tax changes have fully taken hold” (emphasis added). Showing a short-run graph that has results that are driven by corporate tax levels rather than changes completely fails to demonstrate that the benefits to wage growth of cutting corporate taxes are “clearly visible in the data.”

Using data from the Organization for Economic Cooperation and Development, we were able to essentially replicate the CEA results. The most striking finding is that the fast wage growth of the “low-tax countries” highlighted by the CEA in 2015 and 2016 was driven disproportionately by three small countries: Estonia (6.6 percent average wage growth in those years), Iceland (7.5 percent average wage growth), and Latvia (6.8 percent average wage growth). These three countries combined account for 30 percent of the unweighted “low-tax” sample but well over half of the wage growth in the low-tax group, yet their GDPs combined represent less than 0.4 percent of U.S. GDP.

Finally, we use this same OECD data to show changes in corporate tax rates and cumulative wage growth from 2000 to 2016 (see Figure 4). This assessment that is longer-run and examines the effect of corporate rate changes on wage growth is much more relevant for testing theoretical predictions about changing corporate rates and wages. This figure shows that there is no obvious correlation between corporate rate changes and wages; again the beneficial effect of cutting corporate taxes on wages is absolutely not “highly visible in the data.” In fact, the simple slope of the line through the scatterplot is positive, indicating that steeper cuts in corporate rates (the farther to the left of zero) were associated with slower wage growth (slightly and insignificantly, to be sure, as rate cuts just don’t affect wages much).

Figure 4

The key theory behind claims that corporate rate cuts will boost wages is the idea that these rate cuts will lead to substantially faster investment in productivity-enhancing plants and equipment, boosting the nation’s capital stock and making workers more productive.13 We can assess the first link in that chain of causation below, asking simply “are lower corporate tax rates associated with a larger capital stock”? Figure 5 shows a scatterplot of the relationship between the average statutory corporate tax rate between 2000 and 2014 and the capital-to-labor ratio in 2014. The hypothesis is that low-tax countries should have attracted more capital investment and hence should have accumulated a large stock of capital relative to their workforce by the end of the period. As the trend line through the scatter indicates, the relationship actually goes the wrong way—countries with higher corporate tax rates over this period had larger capital stocks by 2014.

Figure 5

This irrelevance of corporate tax rates for the wage growth of typical workers is confirmed by looking across U.S. states as well. These states are in many ways the best possible candidates for providing any evidence that lower corporate taxes actually show up as higher wages. In the jargon of economists, these individual states can be thought of as “small and open” economies—meaning that their wages, prices, and interest rates are highly driven by influences external to the state economy. This is important because economic models that suggest that corporate income tax rate cuts could translate into large wage gains essentially require economies be small and open. If the data show that even individual U.S. states see little correlation between corporate rate cuts and wage growth, it is almost impossible to credibly claim that a cut in the federal corporate rate for the entire—clearly not “small and open”—U.S. economy would deliver wage growth.

Figure 6 shows the change in state corporate income tax rates from 1980 to 2010 and the change in the inflation-adjusted state median wage in that period. There is no correlation at all visible in the data. This reveals a key truth policymakers should face: boosting wages will require a range of policies, and most of these useful wage-boosting policies will not involve taxes.14

Figure 6

Are we sure that nobody in these states will see any gain from cutting corporate taxes? No. Figure 7 looks at the same 1980 to 2010 period and shows the relationship between the change in state corporate income tax rates and the change in the share of total state income claimed by the richest 1 percent of households. Here the correlation (evidenced by the downward sloping trend line) is clearer: a reduction in the corporate rate is associated with an increase in the share of total income claimed by the top 1 percent.

Figure 7

Conclusion

The case that large, deficit-financed corporate tax cuts will boost capital investment, productivity, and wages in the United States is extraordinarily weak. Evidence from past changes in federal taxes, from cross-national comparisons, and from the experiences of individual U.S. states all argue strongly that wages for typical Americans will not benefit from the TCJA.

Claims that the meager data collected so far—data that show no trend break at all in capital investment or wage growth—already indicate a large, positive impact of the TCJA are simply not credible. All in all, the TCJA will serve to boost incomes for the already-rich while doing nothing to help the wages of typical American workers. Additionally, the TCJA is likely to increase incentives to offshore production and profits of American firms. We can certainly do better for America’s workers than the TCJA, and future Congresses should.

Endnotes

1. For this estimate, see Tax Policy Center, Distributional Analysis of the Conference Agreement for the Tax Cuts and Jobs Act, December 18, 2017.

2. These claims and an assessment of them can be found in Josh Bivens and Hunter Blair, Competitive Distractions: Cutting Corporate Tax Rates Will Not Create Jobs or Boost Incomes for the Vast Majority of American Families, Economic Policy Institute, May 9, 2017 and in Josh Bivens, Cutting Corporate Taxes Will Not Boost American Wages, Economic Policy Institute, October 25, 2017.

3. For this number, see “38 Percent of Private Industry Workers Had Access to Nonproduction Bonuses in 2017,” The Economics Daily (from the Bureau of Labor Statistics), December 21, 2017.

4. See Daniel Hemel and David Kamin, “Yes — the Tax Law Could Be Causing Corporations to Pay Bonuses. But It May Be a Tax Game That Won’t Last,” Medium, January 27, 2018.

5. See Kevin Hassett, “The Wages of Tax Reform are Going to America’s Workers,”Wall Street Journal, April 17, 2018.

6. For evidence on this disconnect between economy-wide growth and typical workers’ pay, see Josh Bivens and Lawrence Mishel, Understanding the Historic Divergence between Productivity and Typical Workers’ Pay: Why It Matters and Why It’s Real, Economic Policy Institute, September 2, 2015.

7. For evidence on this, see the references in footnote 2, particularly Figure D in Competitive Distractions: Cutting Corporate Tax Rates Will Not Create Jobs or Boost Incomes for the Vast Majority of American Families.

8. See Akane Otani, Ben Eisen, and Chelsey Dulaney, “Capital Spending Boom Is No Great Boost to Capital Markets,” Wall Street Journal, May 15, 2018.

9. This assumes that the $166 billion total capital investment number from Otani, Eisen, and Dulaney’s 2018 article in the Wall Street Journal (from the previous footnote) is correct. This number is then compared with total private nonresidential fixed investment from the Bureau of Economic Analysis.

10. For evidence on this, see Kimberly Clausing, “Profit Shifting and Offshoring, Then and Now,” and Rebecca Kysar, “Profit Shifting and Offshoring in the New International Regime,” presentations for “Will the Trump Tax Cuts Accelerate Offshoring by U.S. Multinational Corporations,” a conference hosted by the Economic Policy Institute, May 7, 2018.

11. See Council of Economic Advisers, Corporate Tax Reform and Wages: Theory and Evidence, October 2017.

12. See Josh Bivens, Cutting Corporate Taxes Will Not Boost American Wages, Economic Policy Institute, October 25, 2017.

13. Again, we would note a previous EPI report finding that productivity growth alone has not boosted typical workers’ wages in recent decades (see Josh Bivens and Lawrence Mishel, Understanding the Historic Divergence between Productivity and Typical Workers’ Pay: Why It Matters and Why It’s Real, Economic Policy Institute, September 2, 2015). Complementary policies that ensure this productivity growth actually translates into broad-based growth in labor compensation are also needed.

14. For evidence on policies that will work to boost wages, see Ross Eisenbrey and Lawrence Mishel, “How to Raise Wages: Policies That Work and Policies That Don’t,” Economic Policy Institute, March 19, 2015.

Economists have started to identify concentration in both labor and product markets as a potential threat to living standards and wages of typical American families. Concentration in product markets (a limited number of sellers) is generally labeled monopoly power while concentration in labor markets (a limited number of employers—or buyers of labor) is generally labeled as monopsony power. This focus on market power in the form of market concentration represents a welcome and overdue shift. For too long, many researchers tried to explain troubling trends in American workers’ wages with textbook models of perfectly competitive labor markets. Specifically, this long research effort claimed that rising wage inequality and slow wage growth for typical workers was the result of economic influences (such as new technologies) that “shift” demand and supply curves for labor in a competitive model. This approach has decisively failed.1 Given this, any new research effort that introduces market power is an important step in the right direction.

This paper highlights some empirical findings from the new literature on the effect of labor and product market concentration on wages. We address three questions about market concentration that have not always been placed front and center in this literature. The first question is, “Does concentration adversely affect wages at a point in time?” The second question is, “Has concentration grown over time?” The third question is, “Can growing concentration by itself explain a significant portion of the change in wage trends in recent decades?” We find there is evidence to answer “yes” to the first and second questions but not the third. To be clear, the failure to answer affirmatively to the third question is not a criticism of these studies. The studies are not claiming that rising concentration alone can explain wage stagnation or inequality. Yet too many readers have taken these studies’ findings to this conclusion.

Finally, this paper makes two broader points about market power. First, market concentration is not the only source of power—particularly employer power—in markets. Second, even unchanged employer power (like that conferred by market concentration) can play a role in growing wage suppression and inequality if it is accompanied by a collapse of workers’ market power. The new literature on market concentration tells us a lot about employer power, but further exploration of what has happened to workers’ market power remains a key research agenda.

This paper highlights the need to tackle sluggish wage growth and rising inequality with a broad menu of policy interventions that go beyond those provided by competitive models to focus on employer and worker power, and even beyond the antitrust agenda suggested by focusing exclusively on market concentration.

Following are our key conclusions:

Labor market concentration is negatively correlated with wages, but the scope of its downward pressure on wages is limited.

• New research shows that labor market concentration is negatively correlated with wages. However, the effect of labor market concentration is comparatively modest when scaled against what we consider the most significant wage trend in recent decades: the growing gap between typical (median) workers’ pay and productivity.
• The new literature on market concentration has not yet provided concrete empirical estimates of a key labor market trend of recent decades—rising compensation inequality. This should be a priority for this research agenda in the future.
• The new concentration literature does allow us to estimate the effect of market concentration on the share of overall income claimed by labor compensation. These estimates suggest that concentration has not risen enough, nor is its effect on labor’s share of income strong enough, to account by itself for an economically important share of the divergence between economywide productivity and the typical worker’s pay in recent decades.
• The new research on labor market concentration implies that this concentration reduced wage growth by roughly 0.03 percent annually between 1979 and 2014, a decline that would explain about 3.5 percent of the total divergence between the median worker’s pay and economywide productivity over the same period.
• One important study shows that the “average” labor market is “highly concentrated.” But differences between measures of concentration of the average labor market and the labor market experienced by the average worker have important implications for how to assess the impact of labor market concentration on long-term wage trends. In other words, many labor markets suffer from high degrees of concentration, but most people work in labor markets with only low-to-moderate degrees of concentration.
• Nonetheless, labor market concentration is a particular challenge for rural areas and small cities and towns. This is an important finding for those looking to provide economic help to residents of those areas.
• Research on labor market concentration within manufacturing shows a modest increase in labor market concentration between 1979 and 2009.

Product market concentration has increased for some sectors—but at varied rates—and the scope of its downward pressure on wages is also limited.

• Product market concentration rose steadily across six sectors from 1982 to 2012 (manufacturing, retail, wholesale, services, finance, and utilities and transportation), but the magnitude of this rise has varied substantially and it is unclear how much product market concentration has affected labor market trends.
• The new literature on product market concentration indicates that it may have reduced overall wages by roughly 0.08 percent annually from 1979 to 2015, or less than 10 percent of the total divergence between a typical worker’s pay and productivity over that period.

The focus on market power as a key driver behind American wage trends should focus as well on developments that have weakened workers’ power.

• Explaining the expanding pay–productivity gap and increasing inequality in America requires labor market models that allow for employer market power, but the conception of power must go beyond measurable market concentration. Instead, this analysis of power must focus on what has happened to the countervailing power American workers were once able to wield but which now seems radically reduced.
• Correspondingly, a policy response to rising employer power over wages must go well beyond antitrust reform to focus on every possible margin along which policy could strengthen workers’ leverage and bargaining power.

A quick economic background for the new literature on market concentration

Several recent studies have moved analyses of market power front and center in the discussion of American wage trends. This focus on market power is a welcome shift away from analyses claiming that the wage trends we see are simply perfectly competitive labor markets responding smoothly to exogenous “shocks” to labor supply or demand. These textbook models of labor markets have largely failed to explain the most important trends in American wages, such as the long-term stagnation in hourly wages for the typical worker and the closely related divergence between hourly pay for most workers and economywide productivity.

Most of the recent studies focusing on market power look at a relatively narrow slice of potential power: market concentration. Specifically, they look at firm concentration in product markets (or monopoly power) or employer concentration in labor markets (or monopsony power). In highly concentrated product markets, the combination of few sellers and a lack of credible competition from new entrants grants incumbent firms an ability to set prices, rather than simply having to accept the going price (i.e., they have the ability to set prices over marginal product and make economic profits).2 Monopoly power raises the prices paid by consumers and shifts national income away from workers and to corporate profits.

In concentrated labor markets, few buyers of labor and a lack of credible competition from new entrants gives employers the ability to set wages lower than they would be in a competitive market (i.e., lower than their workers’ marginal value product).3 As this description suggests, the impact of labor market concentration on the share of overall income claimed by labor is straightforward: growing concentration should reduce labor’s share of overall income. However, the impact of labor market concentration on rising wage (or compensation) inequality is indeterminate. For inequality to increase, higher-wage workers would need to be getting a greater share of overall labor income. If labor market concentration affects all workers equally, it will not increase wage or compensation inequality. But if labor market concentration is more pronounced in economic sectors that disproportionately employ one particular type of worker—say less-credentialed workers—then concentration could, in theory, contribute to rising wage inequality. As an example, assume that concentration occurred only in the retail sector, which employs many relatively low-wage, non-college-educated workers. This rising concentration would push down retail wages, leading to a growing pay gap between college-educated workers and their non-degreed peers.

The need to explain the productivity–pay divergence

Before assessing the role of any particular factor in driving wage trends, one needs to settle on the most important trend to explain. From our perspective, the key trend to explain is the divergence between the growth of wages and compensation for the typical American worker and the growth in economywide productivity for most of the post-1979 period.4

One of our goals is to assess how much of this growing productivity–pay gap can be explained by rising concentration in product and labor markets. Assessing the role of market concentration in the growing pay–productivity gap is equivalent to assessing how growing concentration affects increasing compensation inequality and the erosion of labor’s share of income, as these are the two channels through which increased productivity can bypass the pay of typical workers.

In fact, in earlier research (Bivens and Mishel 2015, Table 1), we found that the widening productivity–pay gap is the result of these two developments: the erosion of labor’s share of income and the growth of wage or compensation inequality. Table 1 presents our 2015 findings. It shows that the gap between net productivity (economywide productivity net of depreciation) and real median hourly compensation grew 0.8 percent per year between 1973 and 2014, and grew a larger 1.1 percent per year when just looking at the 2000 to 2014 period. It further shows that growing compensation inequality is responsible for the large majority (82.5 percent) of the growth of the productivity–pay gap over the entire period, 1973–2014, whereas the erosion of labor share of income explains about one-sixth of the gap (16.3 percent per year). However, the erosion of labor’s share became more important after 2000, explaining 46.3 percent of the gap in the 2000–2014 period and responsible for lowering real median hourly compensation by roughly 0.5 percent each year relative to net productivity.

Table 1

The data in Table 1 provide context for assessing the role of a factor, such as rising market concentration, in explaining the growing productivity–pay gap or the gap’s two components: the loss of labor’s share of income and rising compensation inequality. These data also demonstrate that analyses that cannot shed light on rising compensation inequality will have limited ability to explain the divergence between typical workers’ pay and productivity. The new papers on labor market concentration have used newly developed microdata to assess the scale of labor market concentration and its impact on overall wages. However, these papers do not examine how labor market concentration affects different groups of workers or the inequality of wages or compensation. This seems like an obviously promising vein of future research.

The impact of labor market concentration on wage trends in a particular period (e.g., 1979–2017, or 2000–2017) essentially depends on two key factors: first, how much labor market concentration has risen has over that period, and, second, the effect of a given rise in labor market concentration on wages. The effect of concentration on wage trends over the entire period is essentially just the first factor multiplied by the second. But both factors would need to be economically significant for concentration alone to explain wage or inequality trends.

As we show below, the new literature on market concentration is careful, valuable, and interesting. But it cannot, by itself, provide a compelling explanation for developments in the American labor market in recent decades, nor can it identify the key policy responses. Identifying market power as a factor in driving American wage trends is a huge advance. However, we think that before too long, conceptions of market power besides market concentration will have to be considered if we want to make serious headway in explaining wage trends and enabling broad-based wage growth.

Assessing the empirical findings of the new literature on market concentration

The literature on market concentration is still emerging, but a number of significant new contributions have been published in just the past few years. In this section we review the findings of a number of recent papers identifying market concentration as a force holding back American workers’ wage growth. Azar, Marinescu, and Steinbaum (2017); Benmelech, Bergman, and Kim (2018); and Naidu, Posner, and Weyl (2018) assess the effect of labor market concentration on wages, whereas Autor et al. (2017) and Struyven (2018) examine the rise in product market concentration and its potential effect on wages.5 Our aim is to translate these papers’ findings into estimated effects that can be compared with effects of other factors found to contribute to wage growth and inequality so that we can provide a rough assessment of how much market concentration alone can explain wage inequality trends.

On some level, it is surprising that this exercise is needed. The labor economics literature on wage inequality used to be full of entries examining particular sources of wage suppression (e.g., deunionization, immigration, trade) and presenting their estimated wage impact, including their estimated contribution to the growth of various measures of inequality (e.g., the wage premium that college-educated workers enjoy relative to high school–educated workers, and the various gaps between pay of workers in the 90th, 50th, and 10th percentiles of the wage distribution). We put some of the recent findings on concentration in labor and product markets into this broader context, something that the recent spate of papers has not done.

The effect of concentration in the labor market (monopsony)

Azar, Marinescu, and Steinbaum (2017) and Benmelech, Bergman, and Kim (2018) have greatly elevated our empirical knowledge of labor market concentration and wages, kicking off a likely wave of forthcoming research. Their papers represent a breakthrough in research on labor market concentration because they utilize data that can directly measure the degree of concentration rather than relying on inference, and thus are able to estimate the impact of labor market concentration on wages.

Azar, Marinescu, and Steinbaum 2017: Measuring labor market concentration across industries and commuting zones

In the conclusion to their paper, Azar, Marinescu, and Steinbaum (2017) summarize their contribution to the conversation about wages and market power:

[W]e contribute to this growing debate by calculating measures of market concentration in local labor markets for the most frequent occupations on the leading employment website CareerBuilder.com. We have shown that concentration is high, and increasing concentration is associated with lower wages. Our results suggest that the anti-competitive effects of concentration on the labor market could be important. The type of analysis we provide could be used to incorporate labor market concentration concerns as a factor in antitrust analysis. (19)

The strength of this paper is that it uses a unique data source to directly measure the degree of labor market concentration within 26 occupations in 681 commuting zones (local labor markets). An additional feature of the data is that they cover the entire private sector.6 Moreover, Azar, Marinescu, and Steinbaum do a careful job of estimating the impact of labor market concentration on wages offered by employers.

However, Azar, Marinescu, and Steinbaum’s 2017 research is motivated in large part by the question of whether antitrust enforcement actions pursued by the federal government should consider labor market concentration when scrutinizing mergers between firms. The authors conclude that it should. The basis for this conclusion is their empirical estimate that greater labor market concentration does indeed appear to lower wages within this relatively recent period. Specifically, they find that moving from a less concentrated to a more concentrated labor market is associated with a decline in wages. The application of this result to antitrust enforcement does seem like a solid policy recommendation stemming from careful research.

Some economic writers have characterized Azar, Marinescu, and Steinbaum’s 2017 paper as providing a new explanation (and perhaps even “the” explanation) for adverse trends in American labor markets in recent decades.7 This interpretation, however, is not consistent with the data Azar and his co-authors present, for a number of reasons. First, their measures of labor market concentration span only three years (2010–2013), so the research can’t speak to how labor market concentration has changed over any significant period of time. Second, their measure of labor market concentration is for the “average market,” in a sample of labor markets dominated by small, nonurban markets. The paper does not present data on how concentration affects the average worker, nor does it tell us how many workers face high, moderate, or low labor market concentration.

The paper features a map of labor markets measured on a Herfindahl-Hirschman Index (HHI) showing that concentration is high in rural areas and modest in the larger population centers.8 The metrics provided in Table 1 in the paper indicate that the “average” labor market is “highly concentrated,” while an average person lives in a population-weighted labor market characterized by “moderate concentration” (Table 1 and Figure 1 in the paper indicate that the precise measure of concentration faced by the “average person” is actually on the lower end of the moderate concentration range).

This difference between “average labor market” and “average person” is important. Given that the average worker lives in a labor market characterized by the low end of moderate concentration, it could well be the case that workers in urban areas (areas that Azar, Marinescu, and Steinbaum [2017] characterize as having lower rates of concentration) face “low concentration.” In their data, urban areas, in turn, account for more than half of the workforce.

This is directly confirmed by a 2018 paper, using a different data set, specifically, a data set from Burning Glass Technologies. In this paper, the same authors (Azar, Marinescu, and Steinbaum), joined by a fourth (Taska), estimate that 65 percent of commuting zones and labor markets are highly concentrated or moderately concentrated. Together, the highly and moderately concentrated labor markets account for only 23 percent of total employment, with 17 percent of employment in “high” and 6 percent in “moderately concentrated” labor markets (Azar et al. 2018, 13).

Therefore, characterizations of this work as showing “substantial labor market concentration in labor markets throughout the United States” (Naidu, Posner, and Weyl 2018, 20) may confuse some consumers of this research. We would say instead that substantial labor market concentration is a problem faced by a nontrivial share of the American workforce—a problem that is particularly challenging for workers in less densely populated areas.

This regional difference in the effect of labor market concentration is an important finding for policy discussions of ways to improve the economic fortunes of Americans in rural areas and small cities and towns. These regions have lagged the rest of the country in many important economic respects. So far, the policy agenda proposed for the economic challenges of rural areas has been weak tea, consisting generally of a call to speed up the build-out of broadband internet (a worthy cause, but hardly one that seems like a game-changer for these regions). Azar, Marinescu, and Steinbaum (2017) highlight a glaring region-specific challenge being faced by these areas: labor market concentration. Policy efforts to fight labor market concentration would have disproportionate effects on rural areas, and Azar, Marinescu, and Steinbaum’s 2017 study provides one of the most important policy insights on this topic that we have seen.

But what about the main topic of this review: the empirical impact of labor market concentration on wage trends? Azar, Marinescu, and Steinbaum (2017, 9) identify “the effect of concentration on wages” based on the “variation in concentration over time within a labor market” for a set of specific occupations. Hence, one can think of their estimate as, for example, tracing the impact of changes in concentration over a limited period of time (2010 to 2013) on offered wages for “Accountants and Auditors in the commuting zone around Kansas City.” Azar, Marinescu, and Steinbaum (2017, 13) report, “We find that higher labor market concentration is associated with significantly lower real wages.” They find that a one-log-point increase in the HHI is associated with a decline in wages of about 0.04 log points in the baseline ordinary least squares (OLS) results, or 0.13 log points in the baseline instrumental variables (IV) results. Moreover, they report that “going from the 25th percentile of market concentration to the 75th percentile of market concentration” lowers posted wages by 5 percent using their baseline OLS results and by 17 percent using their baseline IV results.

Another way to evaluate their results is to assess the effect of moving from “moderate” to “high” labor market concentration. Suppose one shifted from the midpoint of their gauge of moderate concentration (characterized by an HHI ranging from 1,500 to 2,500) to the midpoint of high concentration (characterized by an HHI ranging from 2,500 to 5,000), an increase of 1,750. This (sizable) increase would lower wages by 2.4 percent using the OLS results and by 8.0 percent using the IV results.9 One obtains a similar wage impact for concentration rising from the mean to one standard deviation above the mean (a common metric for measuring a substantial change in dependent variables).

If there had been a rise in labor market concentration in recent decades as large as this 1,750 increase in the HHI, then the market concentration identified in the Azar, Marinescu, and Steinbaum 2017 results could indeed be a significant driver of overall wage trends. However, their measure of labor market concentration does not span enough time to know if this is true or not (in fact, their report does not indicate whether average labor market concentration rose or fell in the three-year span their data encompass).

We can, however, do a thought experiment combining the Azar, Marinescu, and Steinbaum 2017 estimated coefficients with estimated trends in concentration over time taken from another study. In the next section, we discuss a study that does have some time-series variation to shed light on the question of trends in concentration: Benmelech, Bergman, and Kim (2018) report that labor market concentration in manufacturing grew by one-fifth of a standard deviation from 1977 to 2009. Here, we can undertake a thought experiment using the Benmelech, Bergman, and Kim 2018 measure of rising concentration over time and the Azar, Marinescu, and Steinbaum 2017 coefficient estimates of concentration’s effect on wages to see what the trend in concentration identified by Benmelech, Bergman, and Kim (2018) would imply for wages if the Azar, Marinescu, and Steinbaum 2017 estimated coefficients were correct.

Using Azar, Marinescu, and Steinbaum’s (2017) baseline results, a rise in the mean of labor market concentration of one-fifth of a standard deviation (what Benmelech, Bergman, and Kim [2018] find for rising concentration) would result in wage losses of from 0.5 percent to 1.7 percent.10 These are nontrivial results, but still leave a lot of the overall divergence between pay and productivity to be explained. For example, take the midpoint of this thought experiment, where rising concentration drives a 1.1 percent wage decline between 1979 and 2014.11 This translates into an average annual decline of 0.03 percent. Over this same period, the gap between typical workers’ pay and productivity grew by roughly 0.9 percent annually, so these estimates indicate that increasing labor market concentration alone could explain only about 3.5 percent of the divergence.

Benmelech, Bergman, and Kim 2018: Looking at increasing labor market concentration in manufacturing

Benmelech, Bergman, and Kim (2018, 23) note that they use direct measures of labor market concentration and wages from manufacturing plant-level data “to provide evidence that wages are significantly lower in local labor markets in which employers are more concentrated.” As they explain, they do so by using

the Longitudinal Business Database (LBD) to construct the Herfindahl-Hirschman Index (HHI) of firm employment at the county-industry-year level. These HHI measures of employer concentration are then related to measures of average wages and productivity at the establishment level constructed from the Census of Manufacturers (CMF) and the Annual Survey of Manufacturers (ASM).” (3)

Describing their baseline finding, Benmelech, Bergman, and Kim (2018, 3) note:

 [T]here is a negative relation between the local-level HHI measures of employer concentration and wages…[W]e show that these results continue to hold after controlling for a host of observables likely to affect wages, including establishment-level labor productivity and local labor market size, as well as firm-by-year fixed effects. Identification is thus achieved using within-firm-year variation in which (after controlling for such observables as productivity and market size), within a given year, two establishments belonging to the same firm but located in areas with varying levels of labor market concentration are compared.

The Benmelech, Bergman, and Kim 2018 research is notable for directly measuring labor market concentration from 1977 to 2009 and for estimating the wage impact of labor market concentration over this same period. Their results therefore allow us to examine how increases in labor market concentration or changes over time in the impact of concentration on wages (at a given level of concentration) may have affected wage trends over the long term. According to their data, labor market concentration has increased:

During 1977–2009, the standard deviation of local-level employer HHI (defined at the 4-digit SIC-level) was 0.334…the data show that local-level employer concentration has increased considerably over time, with the employment weighted mean four-digit county-level HHI increasing by 5.8%, from 0.698 during 1977–1981 to 0.756 during 2002–2009. (Benmelech, Bergman, and Kim 2018, 3)

They also provide various findings that indicate a statistically significant wage impact of labor market concentration:

• “[A] one standard deviation increase in the HHI measure of local employer concentration reduces wages by between 1% and 2%” (4).

• “[O]ur results indicate that the negative relation between employer concentration and wages doubles in magnitude over the sample period: during 1997–2001, a one standard deviation in local-level concentration is associated with a 1.37% wage reduction, whereas the equivalent effect in 1977–1981 is only 0.63%” (4).

• “[R]elative to other plants in nonperfectly concentrated local labor markets, plants in perfectly monopsonistic labor markets pay 1.7% lower wages, controlling for plant- and county-industry-level determinants of wages.… In Panel B, we compute employer concentration using four-digit SIC industries and find results that are consistent with those in Panel A—local labor markets with only one firm would pay wages that are 3.1% lower than otherwise similar plants within firms” (18–19).

What can these findings tell us about how much the rise in labor market concentration affected wage growth from 1977 to 2009? Labor market concentration did rise, but by a relatively small amount over this period—5.8 percent—a rise of one-fifth of one standard deviation. The wage impacts Benmelech, Bergman, and Kim (2018) report for a full standard deviation rise in labor market concentration are between 1 and 2 percent, indicating a small wage impact (about 0.2 percent to 0.4 percent) due to the actual labor market concentration increase since the 1970s.

Benmelech, Bergman, and Kim (2018) actually provide more detailed results that allow us to further assess the wage impact of rising concentration: they estimate a separate wage elasticity of concentration for each subperiod and provide the concentration levels for each subperiod. This enables one to infer the rise in concentration and its resulting effect on wages in each subperiod.

The relevant data are presented in Table 2, with column 1 providing the coefficient on HHI estimated for each subperiod, column 2 providing the HHI for each subperiod, and the last column showing the impact on wages of HHI in each subperiod (the result of multiplying column 1 by column 2). The table shows that labor market concentration reduced wages by 3.1 percent in 1977–1981, by 3.2 percent in 1997–2001, and by 3.6 percent in 2002–2009. The amount that wages were reduced by labor market concentration rose by 0.5 percent over the entire period from 1977–1981 to 2002–2009, but there is no consistent trend in between those years.

Thus, these results do not show that a rise in labor market concentration over roughly three decades greatly increased its adverse impact on wages, at least in manufacturing. Hence, rising labor market concentration does not seem to be a major factor that explains wage inequality or the slow growth in pay for the vast majority of American workers. However, it should be noted that the Benmelech, Bergman, and Kim 2018 results only apply to manufacturing. We don’t know how much labor market concentration increased in other sectors. It certainly could be the case that manufacturing has always been highly concentrated relative to other sectors, but that other sectors have seen substantially greater increases in concentration in recent decades.

Table 2

Naidu, Posner, and Weyl 2018: Looking at labor market power, though not necessarily labor market concentration

Naidu, Posner, and Weyl (2018) provide a wide-ranging treatment of how market power—particularly in the labor market—can lead to both inefficiency and inequality. They survey a broad literature assessing the potential strength of labor market power and use the estimated parameters to put bounds on how large the efficiency and distributional effects of labor market power might be. They find it could be quite large: in one calibration they find that labor market power might reduce overall gross domestic product by almost 13 percent and overall wages by almost 25 percent. They use these results to highlight how important it is for regulatory authorities to carefully scrutinize how mergers of firms would affect labor market competition.

There are two important things to note about their extraordinary findings. First, their calibrations do not explicitly factor in the source of labor market power, and there is no direct connection between their results and labor market concentration. So, whether their findings are driven by concentration or by some other source of labor market power remains an open question. In short, their findings highlight how potentially important it is to confront employer power of all kinds, not just that driven by labor market concentration (we identify other sources of employer power in a later section). Second, they present no direct evidence that any one particular type of labor market power has increased over time. Our prior belief is that relative employer power in the labor market has indeed increased substantially in recent decades, but the source of this growth is not predominantly explained by growing labor market concentration—it is instead explained by an intentional policy assault on the market power of American workers.12

The effect of concentration in the product market (monopoly)

To gauge the impact of product market concentration on wages, this section examines the findings of a recent highly influential paper (Autor et al. 2017) and an analysis by Goldman Sachs (Struyven 2018), both of which examine the determinants of labor’s share of income.

Autor et al. 2017: Finding growing market power lowers labor’s share of national income

Autor et al. (2017) break new ground by using firm-level data to examine the impact of product market concentration on labor’s share of income in six different large sectors. Autor et al. are largely concerned with the question of whether the lower labor share is a widespread phenomenon across all firms, or is driven by a small number of large “superstar” firms (think Facebook, Apple, Amazon, and Google, among others). Use of firm-level data allows the study to test these contrasting theories, as they respectively imply “…heterogeneous vs. homogeneous changes in the labor share across firms in an industry” (2).

Specifically, Autor et al. (2017) analyze data from the Economic Census from 1982 to 2012 for manufacturing, retail trade, wholesale trade, services, finance, and utilities and transportation. They summarize their findings:

We establish the following facts that are broadly consistent with our model’s predictions for how the rise of superstar firms can lead to a fall of labor’s share: (i) there has been a rise in sales concentration within four-digit industries across the vast bulk of the U.S. private sector; (ii) industries with larger increases in product market concentration have experienced larger declines in the labor share; (iii) the fall in the labor share is largely due to the reallocation of sales between firms rather than a general fall in the labor share within incumbent firms; (iv) the reallocation-driven fall in the labor share is most pronounced in precisely the industries [that] had the largest increase in sales concentration; and (v) these patterns are also present in firm- and industry-level datasets from other OECD countries. (3)

How much did the rise in concentration affect labor’s share of income? Because of data limitations, Autor et al. (2017) are only able to examine changes in labor’s share of income in manufacturing. In other sectors they examine changes in labor’s share of total sales or revenue. They encapsulate their results:13

Observed measures of concentration can account for some of the fall of the labor share, but not the majority. In services for example, the labor share of sales fell from 37% to 34.5% (2.5 percentage points). We predict that this fall would counterfactually have been to 35.2% in the absence of the rise in concentration, i.e., a 1.7 percentage point decline, implying that about a third of the reduction in labor share is proximately explained by rising concentration—a non-trivial fraction. Similarly, rising concentration accounts for 10% of the decline in the labor share in manufacturing, 25% percent in utilities and transportation, and more than 100% in retail trade. While the labor share actually rose in both wholesale trade and finance, our regressions imply that it would have risen by an additional 50% and 150% in these two sectors, respectively, had concentration not increased.

Although the magnitude of the effects is modest when looking over the entire period, it is striking that the importance of concentration has risen over time. For example, if we restrict attention to the second half of the sample (1997–2012), where the relationship between concentration and labor share strengthened and where the rise in concentration was more dramatic, we calculate that rising CR20 concentration in manufacturing accounts for a third of the fall in the labor share [in that sector]. (14–15)

The authors do not translate their results into a specific wage impact. However, a careful look at these results indicate that concentration alone can account for only a minor portion of the changes in labor’s share over recent decades. And, earlier in our paper, we established that changes in labor’s share of income can explain only a small share of the overall divergence of typical workers’ pay and economywide productivity since 1979. This means that the Autor et al. 2017 results are necessarily limited because they only address a portion of changes in labor’s share of income (with labor’s share of income already constituting a minority player in driving the overall divergence between pay and productivity) and do not address changes in the more important issue of changes in overall compensation inequality.

To illustrate, say that that rising monopoly power could explain a third of the fall in labor’s share. What is the corresponding impact on wages or compensation? According to the Bureau of Labor Statistics, labor’s share of income in the private business sector fell from 68.2 percent in 1979 to 66.2 percent in 1997 and then to 62.8 percent in 2015.14 The long-term fall was thus 5.4 percentage points, which, if restored, would lift labor compensation by 8.6 percent.15 That’s a sizable compensation loss due to a fall in labor’s share. If rising monopoly power could explain a third of the loss in labor’s share (a generous assumption), then this is equivalent to causing a 2.9 percent compensation loss.16

A 2.9 percent loss in compensation since 1979 is not trivial, but the annual loss of 0.08 percent since 1979 would explain less than 10 percent of the annual 0.9 percent divergence between pay and productivity that occurred over this time.

Finally, while there is no mechanical one-to-one correlation between rising monopoly power and rising monopsony power, the idea that rising product market concentration could lead to rising labor market concentration has been floated by researchers (CEA 2016). It certainly makes intuitive sense—if rising product market concentration leads to fewer firms, this would seem to imply that the labor market (where firms are buyers rather than sellers) may become more concentrated as well.

The Autor et al. 2017 results are hence useful complements to Azar, Marinescu, and Steinbaum 2017 and Benmelech, Bergman, and Kim 2018 in that Autor et al. show concentration (albeit, product market concentration rather than labor market concentration) over time for a large range of industries. While Benmelech, Bergman, and Kim (2018) show rising labor market concentration within manufacturing, the Autor et al. 2017 results show that manufacturing is no outlier when it comes to rising product market concentration. Specifically, Autor et al. (2017) show that concentration in manufacturing, measured by the share of the market claimed by the 20 largest firms, rose by just under a fifth of a standard deviation between 1982 and 2012. This increase is larger than the increase for retail trade but smaller than the increases for wholesale trade, services, finance, and utilities and transportation (which saw increases ranging from a quarter to 40 percent of a standard deviation).

Struyven 2018: Finding increases in industry concentration are associated with falling labor shares of income

The Goldman Sachs economics research team analyzed the impact of product market concentration on wage growth as part of an overall analysis of wage growth. Their findings are described below:

We first assess the impact of product market concentration on labor shares in an industry panel dataset using revenue concentration, payrolls, and revenues data from the Economic Census. We find evidence shown in Exhibit 6 that increases in industry concentration are associated with falling labor shares. Combining the average increase in the share of industry revenue of the top-20 firms in 2002–2012 and our coefficient estimates, we estimate that the rise in product market concentration accounts for a 1.5% hit to the level of wages since 2002, or a 0.15 pp drag to annual wage growth. (Struyven 2018, 6)

This wage loss is due to a rise in the top-20-firm concentration ratio across 14 industries from 20.9 percent to 22.6 percent from 2002 to 2012, a seemingly small rise that led to a still-low concentration of sales.17 We would have more confidence in this result if there were accompanying results showing that this modest rise in concentration also produced higher profit margins consistent with these wage effects.

Struyven’s finding that increased monopoly power lowered wages by 1.5 percent over a 10-year period (or 0.15 percent per year) indicates a nontrivial impact. However, this would represent a small share of the 5.5 percentage-point decline in labor’s share of business-sector income over that period, a decline that would require a roughly 9 percent wage boost to reverse.18 This implies that rising monopoly power explained about 17 percent of the fall in labor’s share of income (1.5 percent wage decline divided by the 9 percent wage boost needed to reverse the fall in labor share in the business sector). Alternatively, the 0.15 percent annual wage decline caused by increased concentration explains 14 percent of the 1.1 percent annual divergence between pay and productivity from 2000 to 2014 period.

The still-missing link: How changes in market power unrelated to concentration might explain wage trends

The new papers on market concentration are useful and creative, and provide important new lenses for various forms of policy evaluation. For example, Naidu, Posner, and Weyl (2018) note that the potential wage-suppressing effect of corporate mergers should be a criterion considered by the regulators who approve these mergers. Yet, as we note above, the findings from these empirical analyses of market concentration also make clear that growing concentration alone is unlikely to emerge as a plausible primary driver of the adverse labor market trends affecting the vast majority of American workers in recent decades. But these new papers, combined with other insights on how power affects markets (particularly labor markets), can indeed provide valuable pieces of the puzzle of why wage growth for most American workers has been so sluggish since the late 1970s.

For example, another strand of economic models that invokes employer power as a determinant of wages is “dynamic monopsony.” Despite the presence of the word “monopsony,” these models do not necessarily have anything to do with labor market concentration.19 Instead, these models posit that labor market “frictions” can make markets behave as if there were concentration in the labor market and restricted options for workers seeking better pay than what their current employer provides. The clearest sign of this sort of monopsony power is that firms face an upward-sloping supply curve for labor.

The source of some of the frictions that drive dynamic monopsony is just the normal functioning of markets in the real world as opposed to the perfectly competitive models described in textbooks. For example, workers may have incomplete information about potentially better jobs available at other employers. Or transportation costs might make it difficult to seek higher wages at employers further from workers’ homes and hence might constrain searches for better jobs. Care responsibilities could limit workers to certain schedules and likewise reduce their potential pool of alternative employers. Policy responses to reduce the damage these frictions can do to workers’ earning power could include investments in public transit and in high-quality, affordable child care, both of which could expand the range of jobs workers could consider.

What if relative employer power has grown only because workers’ power has been hamstrung by policy?

Given that frictions can reduce workers’ ability to find alternative employment, it is no surprise that some employers strive to create such frictions. For example, many firms require that new employees sign noncompete agreements as a condition of employment. If workers believe that these noncompete agreements are enforceable, then their ability to search for better jobs can be restricted, giving firms some degree of monopsony power. Or employers may collude to refrain from poaching one another’s employees. The baldest real-world example of this type of employer collusion surfacing in recent years was the cartel of Silicon Valley employers that agreed to not hire one another’s employees.20

A common assumption uniting analyses of labor market concentration and studies documenting frictions that generate dynamic monopsony power is that growing employer power might lie behind American wage trends. It seems clear to us that employers do wield more relative power vis-à-vis their workers and that this plays a large role in driving wage trends. But this rise in the relative market power of employers might owe less to growing market concentration or labor market frictions and more to the collapse of policies and institutions that buttressed the relative market power of workers.

It may have always been the case that American labor markets are concentrated, and that this concentration—all else equal—puts downward pressure on wages. It may also have always been the case that labor markets (particularly low-wage labor markets) are riven with frictions that—all else equal—give employers the power to set wages. But in previous decades, these always-and-everywhere sources of employer market power were likely neutralized by institutions and policies that provided countervailing power to workers. In more recent decades, several of these institutions and policies have been eroded or rolled back, with nothing to replace them as sources of countervailing power.

For example, since 1979, macroeconomic policy (particularly monetary policy) has prioritized steady and very low inflation over low unemployment. Even by too-conservative standards set by official estimates of the natural rate of unemployment, macroeconomic policy has failed to secure full employment for the large majority of these years. This has led to labor markets with too much slack to allow low- and moderate-wage workers to demand and achieve consistent wage gains. The evidence is quite clear that low- and moderate-wage workers need lower rates of unemployment to post wage gains than do their higher-wage peers. It is no coincidence, in our view, that the only period of strong, across-the-board wage growth since 1979 was during the late 1990s and early 2000s, when unemployment was allowed to fall far below levels that had previously been thought to lead to accelerating inflation. In those years, while wages grew across the board, inflation did not accelerate.

Besides labor markets tight enough to allow robust wage growth for most workers, other bulwarks of market power for typical workers (labor standards, broadly defined) have also eroded in recent decades. The most prominent example is the federal minimum wage, which in inflation-adjusted terms is now roughly 25 percent lower than it was at its height in 1968, even though productivity has nearly doubled and low-wage workers have become far more educated in the intervening years (Cooper 2017). Policymakers have failed to enact sufficient increases in the federal minimum wage despite growing economic evidence that most minimum wage increases since 1990 (at the federal or state level) have not caused measurable employment loss, contrary to predictions of competitive labor market models (Cooper, Mishel, and Zipperer 2018). This finding is consistent with low-wage labor markets that are characterized by dynamic monopsony power held by employers. In models of dynamic monopsony, legislated wage increases can lead to higher wages and greater, not lessened, employment.

For middle-wage workers, the key labor standard that has eroded is collective bargaining. Research demonstrates that this erosion has had a substantial impact on middle-wage workers, both union and nonunion (Rosenfeld, Denice, and Laird 2016). The view that labor market concentration and other specific sources of employer power have always been present but were tamed in previous decades by countervailing worker power is consistent with the empirical findings by Benmelech, Bergman, and Kim (2018), which, as noted earlier, seem to indicate that the growth of labor market concentration in and of itself cannot explain a dominant portion of rising wage inequality or the divergence between typical workers’ pay and economywide productivity. Though not mentioned earlier, Benmelech, Bergman, and Kim (2018) also provide empirical results clearly showing that the wage-suppressing effect of labor market concentration is lessened when union coverage is strong. So, if labor market concentration has been relatively constant, but the countervailing force imposed by unionization has eroded, this combination could well have led to significant wage losses. If this is the case, then concentration is clearly an important ingredient in the story, even if eroding employee power is the real lever.

Benmelech, Bergman, and Kim (2018) also show that the wage-suppressing effect of labor market concentration is increased when imports (particularly from lower-wage nations) are high as a share of the local economy. This finding, which is consistent with well-publicized findings by Autor, Dorn, and Hanson (2013), also highlights that trade flows can place downward pressure on wages through channels besides their effect on relative demand for various types of labor. The shorthand description for the effects of globalization on wages running through labor market power is that trade flows impose a threat effect that can dampen wages. Again, if labor market concentration has been an ongoing fact of life that was “trying” to suppress wages for decades, but growing trade flows from lower-wage nations led to a collapse of worker-side market power and the combination led to wage losses, then concentration is a key ingredient in this story.

While economists have been slow to wrestle with the labor market fallout of policy efforts to shift relative market power from workers to employers (at least until the recent spate of literature), employers and their representatives in the policy world certainly seem to think these policies are important. Besides the policy changes listed above, employers have pursued an aggressive host of practices meant to limit workers’ bargaining position, and policymakers, particularly through the blessing of case law in the courts, have often ratified these practices. Examples of these employer practices include mandatory forced arbitration agreements, noncompete agreements, and nonpoaching agreements.21

Further, because of the thin policy framework surrounding mandatory benefits and labor standards, the provision of nonwage compensation—such as health insurance coverage, retirement contributions, paid sick and family leave, vacation time, and the availability of consistent and predictable scheduling—can differ radically across employers. This makes it harder for employees to seamlessly compare jobs with full information and makes it hard for them to unambiguously identify better outside options for employment. All of these factors have systematically undercut workers’ individual and collective leverage relative to employers and led to slower wage growth, especially for low- and middle-wage workers.

Conclusion

Exciting new economic research—particularly studies published over the past year—highlights the role of market power in determining wages. This is a welcome new vein to mine for those wanting to convincingly explain wage trends for American workers over the past generation of economic life. Explaining these trends with the toolbox of competitive labor market models has been a failed effort.

Much of the emerging research so far has examined the effect of market concentration in either labor markets or product markets. There is good reason for this focus. Concentration clearly seems prevalent and by many measures seems to be increasing over time. Concentration is also—importantly—measurable, thanks primarily to data only recently available from public and private sources. The findings of the research on market concentration are compelling and are important grist for a range of policy debates. However, based on the data and estimates available so far, we do not think trends in market concentration have been a dominant driver of the most significant trends in American wages in recent years. Key among these significant trends are the rise in wage inequality and the divergence between economywide productivity growth and hourly pay growth of typical American workers. Instead, we think other models and concepts of power in labor markets will need to be analyzed and assessed to explain these larger trends and to propose effective policy solutions.

That said, because it is relevant to ongoing policy debates, and because it offers warnings about what could further threaten potential wage growth in the future, the new research on market concentration is welcome and important.

About the authors

Josh Bivens joined the Economic Policy Institute in 2002 and is currently EPI’s director of research. His primary areas of research include mac­roeconomics, social insurance, and globalization. He has authored or co-authored three books (including The State of Working America, 12th Edition) while working at EPI, edited another, and has written numerous research papers, including for academic journals. He appears often in media outlets to offer eco­nomic commentary and has testified several times before the U.S. Congress. He earned his Ph.D. from The New School for Social Research.

Lawrence Mishel is a distinguished fellow at the Economic Policy Institute after serving as president from 2002–2017. Mishel first joined EPI in 1987 as research director. He is the co-author of all 12 editions of The State of Working America. He holds a Ph.D. in economics from the University of Wisconsin at Madison, and his articles have appeared in a variety of academic and nonacademic journals. His areas of research are labor economics, wage and income distribution, industrial relations, productivity growth, and the economics of education.

John Schmitt became EPI’s vice president on January 1, 2018, returning to where he started his career as an economist from 1995 to 2001. Following his earlier tenure at EPI, he spent 10 years as a senior economist at the Center for Economic and Policy Research (CEPR) and, most recently, was the research director at the Washington Center for Equitable Growth. Schmitt has a Ph.D. and M.Sc. in economics from the London School of Economics.

Endnotes

1. Key studies highlighting the weakness of perfectly competitive labor market models in explaining key labor market trends include Card and DiNardo 2002 and Schmitt, Shierholz, and Mishel 2013.

2. In competitive models, economic profits are driven to zero by the threat of new entrants into product markets. Economic profits are profits earned over and above the opportunity cost of employers deploying their capital in some other endeavor.

3. The studies we discuss here are primarily interested in labor market outcomes. Some measure concentration using labor market concentration measures; others measure concentration using analogous product market measures.

4. Bivens and Mishel (2015) highlight the divergence between typical workers’ pay and economywide productivity. Mishel and various co-authors have also explored this growing gap in successive editions of EPI’s The State of Working America, including the 2012 edition (Mishel et al. 2012).

5. Other significant contributions not discussed in this paper include Barkai 2017; De Loecker and Eeckhout 2017; and Traina 2018.

6. The sample is made up of firms that chose to advertise openings using an online service and then specifically chose to use the CareerBuilder.com site. Furthermore, even for firms that used CareerBuilder.com, the firms likely did not use the site for all types of job openings.

7. Key examples include Weissmann 2018 and Covert 2018.

8. Page 7 in the report describes their measure this way: “Market power in a labor market is the Herfindahl-Hirschman Index (HHI) calculated based on the share of vacancies of all the firms that post vacancies in that market.”

9. This is calculated by first taking the difference in the log of 3,750 and 2,000 and then multiplying by the coefficient on HHI in the OLS baseline, 0.04, or the IV baseline, 0.13.

10. Movement from the HHI mean, 3,157, to one standard deviation above the mean, 6,081, is a log change of 0.656. Evaluated at the “baseline” coefficients of 0.0387 and 0.127, this change of HHI yields a wage impact of 2.5 percent and 8.7 percent, so one-fifth of a standard deviation impact is 0.5 percent and 1.7 percent.

11. The Benmelech, Bergman, and Kim 2018 findings on concentration span 1977–2009, not 1979–2014. But if we assume this latter period was characterized by the same change in concentration as the former period, it lets us undertake a straightforward thought experiment to scale concentration’s effect against wage and productivity trends between business cycle peaks.

12. See Bivens et al. 2014 for the fuller argument.

13. Autor et al. (2017) describe their methodology in footnote 25: “The fraction of the overall decline in the labor share that is explained by rising concentration comes from a simple back of the envelope calculation. From 1997–2012, the CR20 in manufacturing went up by around 6 percentage points and the labor share fell by around 6 percentage points. From Figure 7, the average coefficient relating the change in concentration to the change in labor share in manufacturing over this period was 0.345, implying that concentration explained (-.345*6)/6 X100 = 34.5% of the fall in the labor share in manufacturing over this period.”

14. Data from the BLS Multifactor Productivity series are obtained at https://www.bls.gov/mfp/mprdload.htm#Historical.

15. This is calculated as follows: 68.2/62.8 = 1.086 = 8.6 percent higher.

16. This is simply one-third of the 8.6 percent loss due to the fall in labor’s share.

17. Based on private email communication from Daan Struyven.

18. In the business sector, a 5.5 percentage-point decline from 67.6 percent in 2002 to 62.1 percent in 2012 requires an 8.9 percent increase (5.5/62.1) to restore labor’s share.

19. Manning 2003 is the definitive statement on the “dynamic monopsony” approach.

20. Ames 2014 is the definitive overview of the employer cartel in Silicon Valley.

21. Krueger and Posner (2018) point to such agreements as a key reason why low-wage workers in particular have had reduced bargaining power in recent decades.

22. The canonical models of this are Bowles 1985 and Shapiro and Stiglitz 1984. Dube, Giuliano, and Leonard (2015) empirically document the importance of fairness considerations in spurring behavioral responses from workers (specifically, “quits”).

23. The canonical fairness model is Akerlof 1982; see also Akerlof and Yellen 1985. This model also informs the findings in Ball and Moffitt 2001 regarding wage “aspirations.” See Krueger’s (2013) examination of fairness in wage determination.

References

Akerlof, George. 1982. “Labor Contracts as Partial Gift Exchange.” Quarterly Journal of Economics 94: 543–569.

Akerlof, George, and Janet Yellen. 1985. “The Fair Wage-Effort Hypothesis and Unemployment.” Quarterly Journal of Economics 105, no. 2: 255–283.

Ames, Mark. 2014. “Revealed: Apple and Google’s Wage-Fixing Cartel Involved Dozens More Companies, over One Million Employees.” Pando, March 22, 2014.

Autor, David H., David Dorn, and Gordon H. Hanson. 2013. “The China Syndrome: Local Labor Market Effects of Import Competition in the United States.” American Economic Review 103, no. 6: 2121–2168.

David Autor, David Dorn, Lawrence F. Katz, Christina Patterson, and John Van Reenen. 2017. “The Fall of the Labor Share and the Rise of Superstar Firms.” Massachusetts Institute of Technology Working Paper, May 2017.

Azar, José, Ioana Elena Marinescu, and Marshall Steinbaum. 2017. “Labor Market Concentration.” National Bureau of Economic Research Working Paper no. 24147. https://doi.org/10.3386/w24147.

Azar, José, Ioana Elena Marinescu, Marshall Steinbaum, and Bledi Taska. 2018. “Concentration in US Labor Markets: Evidence from Online Vacancy Data.” Working paper, March 2018. http://dx.doi.org/10.2139/ssrn.3133344.

Ball, Laurence, and Robert Moffitt. 2001. “Productivity Growth and the Phillips Curve.” National Bureau of Economic Research Working Paper, August 2001. https://doi.org/10.3386/w8421.

Barkai, Simcha. 2017. “Declining Labor and Capital Shares.” London Business School, Working paper.

Benmelech, Efraim, Nittai Bergman, and Hyunseob Kim. 2018. “Strong Employers and Weak Employees: How Does Employer Concentration Affect Wages?” Working paper, January 2018.

Bivens, Josh, Elise Gould, Lawrence Mishel, and Heidi Shierholz. 2014. Raising America’s Pay: Why It’s Our Central Economic Policy Challenge. Economic Policy Institute, June 2014.

Bivens, Josh, and Lawrence Mishel. 2015. Understanding the Historic Divergence between Productivity and Typical Workers’ Pay: Why It Matters and Why It’s Real. Economic Policy Institute, September 2015.

Bowles, Samuel. 1985. “The Production Process in a Competitive Economy: Walrasian, Neo-Hobbesian and Marxian Models.” American Economic Review 75: 16–36.

Card, David, and John DiNardo. 2002. “Skill-Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles.” Journal of Labor Economics 20, no. 4: 733–783.

Cooper, David. 2017. Raising the Federal Minimum Wage to $15 by 2024 Would Lift Wages for 41 Million Workers. Economic Policy Institute, April 2017.

Cooper, David, Lawrence Mishel, and Ben Zipperer. 2018. Bold Increases in the Minimum Wage Should Be Evaluated for the Benefits of Raising Low-Wage Worker’s Total Earnings: Critics Who Cite Claims of Job Loss Are Using a Distorted Frame. Economic Policy Institute, April 2018.

Council of Economic Advisers (CEA). 2016. Labor Market Monopsony: Trends, Consequences, and Policy Responses. October 2016.

Covert, Bryce. 2018. “Does Monopoly Power Explain Workers’ Stagnant Wages?” The Nation, February 15, 2018.

De Loecker, Jan, and Jan Eeckhout. 2017. “The Rise of Market Power and the Macroeconomic Implications.” Working paper, August 2017.

Dube, Arindrajit, Laura Giuliano, and Jonathan Leonard. 2015. “Fairness and Frictions: The Impact of Unequal Raises on Quit Behavior.” IZA Discussion Paper 9149, Institute for the Study of Labor, June 2015.

Krueger, Alan. 2013. “Fairness as an Economic Force.” Remarks addressed to “Learning and Labor Economics” Conference at Oberlin College, Oberlin, Ohio, April 26, 2013.

Krueger, Alan, and Eric Posner. 2018. A Proposal for Protecting Low-Income Workers from Monopsony and Collusion. The Hamilton Project, February 2018.

Manning, Alan. 2003. Monopsony in Motion: Imperfect Competition in Labor Markets. Princeton, N.J.: Princeton Univ. Press.

Mishel, Lawrence, Josh Bivens, Elise Gould, and Heidi Shierholz. 2012. The State of Working America, 12th Edition. An Economic Policy Institute book. Ithaca, N.Y.: Cornell Univ. Press.

Naidu, Suresh, Eric Posner, and Glen Weyl. 2018. “Antitrust Remedies for Labor Market Power.” Working paper, February 2018.

Rosenfeld, Jake, Patrick Denice, and Jennifer Laird. 2016. Union Decline Lowers Wages of Nonunion Workers: The Overlooked Reason Why Wages Are Stuck and Inequality Is Growing. Economic Policy Institute, August 2016.

Schmitt, John, Heidi Shierholz, and Lawrence Mishel. 2013. “Don’t Blame the Robots: Assessing the Job Polarization Explanation of Growing Wage Inequality.” EPI-CEPR Working Paper. Economic Policy Institute and Center for Economic and Policy Research, November 2013.

Shapiro, Carl, and Joseph Stiglitz. 1984. “Equilibrium Unemployment as a Worker Discipline Device.” American Economic Review 74, no. 3: 433–444.

Struyven, Daan. 2018. “Superstar Firms, Super Profits, and Shrinking Wage Shares.” US Economics Analyst, Goldman Sachs Economics Research Note, February 4, 2018 (not publicly available).

Traina, James. 2018. “Is Aggregate Market Power Increasing? Production Trends Using Financial Statements.” Stigler Center for the Study of the Economy and the State University of Chicago Booth School of Business, New Working Paper Series no. 17, February 2018.

Weissman, Jordan. 2018. “Why Is It So Hard for Americans to Get a Decent Raise?” Slate, January 16, 2018.

Rising wage inequality has been a defining feature of the American economy for nearly four decades. In 2017, with an improving economy, all deciles in the overall wage distribution have improved, meaning most workers finally have higher hourly wages now than in 2007, the labor market peak before the Great Recession hit. However, large gaps by gender, race, and wage level remain, and some of these gaps are increasing.

Rising inequality means that although we are seeing broad-based wage growth, ordinary workers are just making up lost ground rather than getting ahead. The bottom seven deciles have seen annual growth of hourly wages of 0.5 percent or less since 2000. The way rising inequality has directly affected most Americans is through sluggish hourly wage growth in recent decades, despite an expanding and increasingly productive economy. For example, had all workers’ wages risen in line with productivity, as they did in the three decades following World War II, an American earning around $40,000 today would instead be making close to $61,000 (EPI 2018e).

The latest data on hourly wages shows that the gap between those at the top and those at the middle and bottom has continued to increase through much of the 2000s. This report analyzes data from the Current Population Survey (CPS) and details the most up-to-date hourly wage trends through 2017 across the wage distribution and education categories, highlighting important differences by race and gender. By looking at real (i.e., inflation-adjusted) hourly wages by percentile, we can compare what is happening over time for the lowest-wage workers (those at the 10th and 20th percentiles) with wage trends for the highest-wage workers (those at the 90th and 95th percentiles). What stands out in this last year of data is that, while there have been welcome improvements, wage growth continues to be slower than would be expected in a stronger economy. Given this slow wage growth, policymakers should not presume that the economy has already achieved full employment. In short, papering over the damage done by the Great Recession does not constitute “mission accomplished” on wages. There remains much more work to be done to reduce wage disparities by gender and race and to reverse the damage done to wages by decades-long trends of rising inequality and wage stagnation.

Key findings

Comparisons between CPS and CES data. We use data from the CPS for our analysis because it allows us to make wage comparisons by race, gender, and education. However, we also compare CPS data with Current Employment Statistics (CES) data because looking at the similarities and differences in trends across the data helps us to get a more complete picture of the state of the labor market.1 We find that:

• Nominal hourly wage growth in the CPS and the CES were fairly consistent from 2016 to 2017, and from 2000 to 2017 the two surveys exhibit similar trends when the CPS is smoothed out using a three-year moving average. However, the CPS exhibits more year-to-year volatility than the CES. This means that one-year changes in wages by decile in the CPS—while providing new and valuable information—should be taken with a grain of salt.
• Strong wage growth in the CPS between 2015 and 2016 is partially attributable to statistical volatility, as is the slower wage growth between 2016 and 2017.
• CPS and CES wage trends both show nominal wage growth that is still below levels consistent with the Federal Reserve’s inflation target and with long-run trend productivity growth—a sign that the economy still has considerable slack.

Wage inequality. From 2000 to 2017, wage growth was strongest for the highest-wage workers, continuing the trend in rising wage inequality over the last four decades.

• Since 2007, the labor market peak before the Great Recession, the strongest wage growth has continued to be within the top 10 percent of the wage distribution.
• From 2016 to 2017, strong growth continued at the top (1.5 percent at the 95th percentile), but the 10th percentile saw the strongest growth at 3.7 percent. Median wages grew only 0.2 percent.

Wage inequality by gender. While wage inequality has generally been on the rise for both men and women, wage inequality is higher and growing more among men than among women.

• From 2016 to 2017, men saw wage declines at the top and bottom of their wage distribution. Modest wage gains at the median finally lifted men’s median hourly wages above their 2007 and 2000 levels.
• Women experienced a far more equal wage distribution, and their wage growth from 2016 to 2017 was relatively more broadly shared, with the strongest growth in the bottom 70 percent of the wage distribution.

Gender wage gap. The gender wage gap at the median has narrowed since 2000, with a typical woman now paid 84 cents on the male dollar, although significant gender wage gaps remain across the wage distribution.

• The gender wage gap at the bottom has also narrowed since 2000.
• The gender wage gap at the top narrowed over the last year but remains wider than it was in 2000.
• The regression-adjusted average gender wage gap narrowed slightly from 2016 to 2017 and is currently at 22.0 percent.

Wage growth in states with minimum wage increases. From 2016 to 2017, wages of the lowest-wage workers grew more in states that had increased their minimum wage.

• In states without minimum wage increases in 2017, the 10th-percentile wage rose 1.7 percent; in states with minimum wage increases in 2017, the 10th-percentile wage rose by 2.1 percent.
• The differential is larger when looking across recent years with many minimum wage increases: between 2013 and 2017, the 10th-percentile wage grew more than twice as fast in states with at least one minimum wage increase in that period versus states without.
• In both comparison periods, both men and women at the 10th percentile saw greater wage growth in states with minimum wage changes versus those without.

Wage growth by race/ethnicity. At every decile and at the 95th percentile, wage growth since 2000 was faster for white and Hispanic workers than for black workers.

• After suffering declines in the aftermath of the Great Recession, the 20th through 70th percentile of the black wage distribution is below or within only $0.03 of its 2000 level.
• From 2016 to 2017, the strongest wage growth among white workers was at the 95th percentile, while the median and the 10th-percentile wages both fell.
• From 2016 to 2017, Hispanic workers experienced more broadly based wage growth than black or white workers, with wages increasing across their wage distribution and growing more strongly at the median and the bottom than at the top.
• Except for the 10th and 30th percentiles, black wages fell across the entire wage distribution between 2016 and 2017.

Racial/ethnic wage gaps. Throughout the wage distribution, black–white wage gaps are larger today than in 2000; conversely, Hispanic workers have been slowly closing the gap with white workers at the bottom 70 percent of the wage distribution.

• In 2000, the regression-adjusted Hispanic–white wage gap was larger than the regression-adjusted black–white wage gap. By 2017, the reverse was true.
• The regression-adjusted black–white and Hispanic–white wage gaps (controlling for education, age, race, and region) have become larger over the last year. The Hispanic–white wage gap has narrowed slightly over the last 17 years, while the black–white gap remains significantly larger today than it was in 2000, up 6.0 percentage points.

Wage growth by education. From 2000 to 2017, the strongest wage growth occurred among those with an advanced degree, a college degree, and less than a high school diploma.

• Over the last year, average wages of those with some college, college degrees, and advanced degrees fell, a reversal in trend for the more educated workers from the previous couple of years.
• Workers with some college still have lower wages today than in 2007 or 2000.

Wage growth by education and gender. Since 2000, wage growth for those with a college degree was faster for men than for women, while wage growth for those with a high school diploma or some college was faster (or less negative) for women than for men.

• In general, the women’s wage distribution by educational attainment is more compressed, that is, the wage differences between workers of different levels of education are not as large for women as they are for men.
• While there has been a slow narrowing of gender wage gaps since 2000 for those with high school diplomas and those with some college, gender wage gaps were wider than in 2000 among those with less than high school, college degrees, or advanced degrees. At every education level, women are paid consistently less than their male counterparts.

Wage growth by education and race/ethnicity. From 2000 to 2017, wage growth for white and black workers tended to be faster (or less negative) for those with more education than those with less education.

• Average wages grew faster among white and Hispanic workers than among black workers for all education groups from 2000 to 2017.
• Among black workers, only college degree holders had higher wages than in 2000, but their wage growth was considerably slower than white or Hispanic workers with college degrees.
• From 2016 to 2017, wage growth was weak (or fell outright) for those with college or advanced degrees in all racial/ethnic groups, while wage growth was strongest for both black and white workers with less than a high school diploma.
• Black–white wage gaps by education were larger in 2017 than in 2000 for all education groups, while Hispanic–white wage gaps were narrower for workers with less than high school and high school diploma levels of education. At every education level, workers of color were paid consistently less than their white counterparts.

Wage inequality and the college premium. Over 2000–2017, the boost to wages that comes from earning a college degree increased, but nowhere near fast enough to explain the total rise in wage inequality over that time.

• Despite weak wage growth in the past year for workers with four-year college degrees, over the longer term (since 2000 and 2007) these workers have seen stronger wage growth than those with high school diplomas.
• While those with college degrees saw wage growth of 6.5 percent from 2000 to 2017, educational attainment has not been sufficient to return many workers to where they were before the recessions of the 2000s: the bottom 50 percent of workers with a college degree still have lower wages than they did in 2000 or 2007.
• The regression-adjusted college wage premium fell from 2016 to 2017, but it is higher than where it was in 2000. The college premium is the percent by which hourly wages of four-year college graduates exceed those of otherwise equivalent high school graduates. The rise in the overall college premium has been driven by increases for men.
• The pulling away at the top of the wage distribution cannot be explained by the rising college wage premium; the increase in the college wage premium slowed considerably in the 2000s and is much smaller in magnitude than the rise in the 95/50 wage gap (the gap between the top and the middle).

Wage survey (CPS and CES) comparisons

Comparisons between the two main wage surveys—CPS and CES—show year-to-year volatility but fairly consistent trends over time.

The Bureau of Labor Statistics releases two surveys every month as part of their Employment Situation report: the Current Population Survey (CPS) and the Current Employment Statistics survey (CES). The CPS collects employment and demographic information from households, while the CES collects information from employers’ payroll records.

We use CPS data because they allows us to examine changes in wages by demographic characteristics such as gender, race and ethnicity, and education. However, the CPS has some weaknesses; in particular, it draws data from a much smaller sample than the CES does. Therefore, we compare CPS wage trends with CES wage trends in order to get a more complete picture of the strength of wage growth in the economy today.

When we do so, we find that the CPS exhibits significantly more volatility than the CES, due primarily to its smaller sample size (see Figure A). However, when the CPS is smoothed using a three-year moving average, the CPS and CES show similar wage trends (see Figure B). A more detailed analysis of wage data in the CPS and the CES follows.

From 2016 to 2017, the U.S. economy experienced nominal hourly median and average wage growth of 2.4 percent and 2.6 percent, respectively, according to data from the Current Population Survey Outgoing Rotation Group (CPS-ORG). Similarly, the CES—the series that provides wage data as part of the Bureau of Labor Statistics’ monthly jobs report—illustrated similar growth in average wages for all private-sector workers and for production/nonsupervisory workers of 2.5 percent and 2.3 percent, respectively. It is unsurprising that median wage growth would be a bit weaker than average in the CPS, as such a difference is indicative of rising inequality. It is also unsurprising that wage growth for production/nonsupervisory workers from the CES would be lower than for all private-sector workers because the former category (constituting roughly 82 percent of private payroll employment) excludes managers and supervisors, who are likely to be more highly paid on average (EPI 2018d).

However, trends in wage growth have not always been consistent between the two surveys in a given year. From 2015 to 2016, the CPS-ORG showed relatively strong growth in nominal median hourly wages of 4.4 percent (Gould 2017). At the same time, the CES showed relatively weaker wage growth for private-sector workers and production/nonsupervisory workers over the same year, averaging 2.6 percent and 2.5 percent, respectively. This lower level of growth registered in the CES data is notably below levels consistent with Federal Reserve targets for inflation and long-run trend productivity growth (EPI 2018c). The 4.4 percent growth in the CPS-ORG was particularly striking in a year in which inflation rose only 1.3 percent. And this level of growth would be highly suggestive of a stronger economy than is indicated by other labor market statistics.

Every month, policymakers, analysts, and journalists look to the monthly jobs report to assess the health of the labor market. Along with payroll employment growth and the unemployment rate, nominal wage growth is a key indicator of the tightness of the labor market, a measure of workers’ ability to secure pay increases from their employers. As workers become scarcer, employers have to pay more to attract and retain the workers they want. So making an accurate assessment of the state of wage growth is essential to a complete understanding of labor market dynamics and to determining how close the U.S. economy may indeed be to full employment.

While trends in the last year are relatively more consistent between the two surveys than in the past, it is still worthwhile to take a deeper look at prior trends and why they may be different across the two surveys. Here are the weeds of this argument. The CPS and the CES provide the main monthly statistics on the labor market. The CPS is a sample survey of about 60,000 households conducted by the U.S. Census Bureau for the Bureau of Labor Statistics (BLS). Its sample, based primarily on the U.S. Census, is designed to reflect the entire civilian noninstitutionalized population. On the other hand, the CES is collected from employers’ payroll records of about 651,000 individual worksites. This information is gathered by the BLS from a sample based on unemployment insurance tax records. Both the household survey and payroll survey data are collected for the week of each month (or pay period) containing the 12th of that month (BLS 2018). Given the larger sample size and the benchmarking of CES employment to unemployment insurance tax records, it has been well established that the CES is the better survey for assessing overall employment growth (Gould 2003).

The CPS samples respondents for eight months total—respondents are “in” for four months in a row, out for eight, and in for four months again. Data on wages from the CPS come from the subsample surveyed in the fourth and eighth months of their time in the survey, that is, questions about wages and earnings are asked only in the last month of each four-month period that a respondent is in the survey. The surveyed group is referred to as the “outgoing rotation group” (ORG) because they are in the last month of their survey rotation for that year. Because respondents are surveyed about wages and earnings only one month out of four, the sample size is only one-fourth of the original 60,000 households surveyed in any given month. Furthermore, the sample may be additionally reduced because wage data is only available for the share of those surveyed who are workers. Therefore, if the CES survey is better on measuring employment changes because of limited data for the CPS, then the problem with comparable wages is likely to be worse. Figure A illustrates year-over-year percent changes in nominal wage growth for all four series in question: the CPS median wage and the CPS average wage (solid lines) and the CES private-sector wage and the CES production/nonsupervisory wage (dashed lines). For EPI’s methodology and sample restrictions in the calculations of means and median, see EPI 2018b.

Figure A

It is clear from Figure A that both CPS-ORG series exhibit far more volatility than the CES wage series do. This is not surprising given the differences in survey features, notably the significantly smaller sample size in the CPS. Looking only at last year’s CPS and CES numbers, one might be tempted to report a slowdown in wage growth from 2016 to 2017, but I would suggest attributing much of this “slowdown” to a reversion to the mean in the jumpier survey. The reversion to a very close matching in wage growth of the CES all private-sector employee series in the case of the CPS average and the CES production/nonsupervisory employee series in the case of the CPS median is merely a matter of odds; however, it allows us to tell a rather conveniently consistent story of wage growth. Using my preferred metric, typical (median) nominal hourly wages, we see that from 2016 to 2017, wages grew 2.4 percent. Given the Federal Reserve’s 2 percent inflation target and 1.5 percent long-run trend productivity growth, wages should be growing at least 3.5 percent for workers to reap the benefits of economic growth (EPI 2018c); 2.4 percent clearly falls short of that. This relatively slow rate of nominal wage growth provides a strong indication that the economy still has a ways to go before reaching full employment. Given that workers have limited leverage to bid up their wages, the economy is clearly exhibiting a fair amount of slack.

Figure B smooths out the CPS median and average series, creating a simple three-year moving average (smoothing medians without constructing medians using the pooled series) to compare with the one-year averages in the CES. It is striking how similar the trends now appear, providing further evidence that larger changes are driven by data volatility. While wage growth was stronger in the CPS from 2016 to 2017, much of that growth can be attributed to statistical volatility as opposed to genuine wage growth for workers. Figure B shows wages moving in the right direction but decidedly not exceeding target inflation plus productivity growth for the median worker.

Figure B

While the CPS’s weaknesses are clear, it remains the best series for measuring wages and wage growth by demographic characteristics as well as across the wage distribution. I suggest taking swings in year-to-year differences with a large grain of salt and paying more attention to longer-term trends. However, I do report cross-cutting differences from the CPS for the most recent year; a look at the most current available data remains valuable to understanding how today’s economy is serving U.S. workers across the labor market.

Wage inequality across the wage distribution

Wage growth from 2000 to 2017 continues long-run trends in rising inequality.

Since 1979, hourly pay for the vast majority of American workers has diverged from economy-wide productivity, and this divergence is at the root of numerous American economic challenges. After tracking rather closely in the three decades following World War II, growing productivity and typical worker compensation diverged (shown in Appendix Figure A). From 1979 to 2016, productivity grew 64.2 percent, while hourly compensation of production and nonsupervisory workers grew just 11.2 percent. Productivity thus grew nearly six times faster than typical worker compensation.

A natural question that arises from this story is just where did the “excess” productivity go? A significant portion of it went to higher corporate profits and increased income accruing to capital and business owners (Bivens et al. 2014). But much of it went to those at the very top of the wage distribution, as shown in Appendix Figure B. The top 1 percent of earners saw cumulative gains in annual wages of 148.6 percent between 1979 and 2016—far in excess of economy-wide productivity growth and nearly four times faster than average wage growth.

While the CPS-ORG—the primary data set used in this paper—does not allow disaggregation within the top 5 percent of the earnings distribution, it is still instructive for measuring the growth in wage inequality over the last 40-odd years. Appendix Figure C illustrates that for all but the highest earners, hourly wage growth has been weak. If it hadn’t been for the period of strong across-the-board wage growth in the late 1990s, wages for most would have fallen outright. Median hourly wages rose 9.5 percent between 1979 and 2017, compared with an increase of 4.4 percent for the 10th-percentile worker (i.e., the worker who earns more than only 10 percent of workers). Over the same period, the 95th-percentile worker saw growth of 51.7 percent.

Wage growth since the Great Recession has continued to follow this trend: slower growth for most compared with faster growth for those at the top. Table 1 shows hourly wages by wage decile (and at the 95th percentile) and includes data from 2000 (the previous business cycle peak), 2007 (the most recent business cycle peak), and the two most recent years of data (2016 and 2017). For a full discussion of EPI’s use of the CPS-ORG data, see EPI’s methodology for measuring wages and benefits (EPI 2018b). In the full business cycle from 2000 to 2007, growth was relatively slow overall and relatively unequal; the gains at the 90th and 95th percentiles were higher than at the middle or bottom of the wage distribution. After growing at practically the same rate from 2000 to 2007, the bottom grew about twice as fast as the middle from 2007 to 2017, narrowing slightly the ratio of wages at the 50th and 10th percentiles of the wage distribution (i.e., the 50/10 wage gap, or the gap between the middle and the bottom). However, because of the large and disproportionate gains at the top, both the 95/50 gap (the gap between top and the middle) and the 95/10 gap (the gap between the top and the bottom) grew substantially from 2007 to 2017.

Table 1

With the caveat that, as discussed above, we need to be careful to not assign too much meaning to one-year changes given concerns about data volatility, we note the following trends over the past year: The one-year change in the median wage from 2016 to 2017 was a paltry 0.2 percent, compared with 1.5 percent at the 95th percentile and 3.7 percent at the 10th percentile. With the recent bump at the 20th percentile, 2017 marks the first year that every wage decile shown has finally exceeded its 2007 and 2000 levels. The continued relatively strong growth at the 10th percentile may have been boosted by state-level minimum wage increases, as discussed below.

Figure C illustrates the trends in wages for select deciles (and the 95th percentile), showing the cumulative percent change in real hourly wages from 2000 to 2017. The overall story of inequality is clear. The lines demonstrate that those with the highest wages have had the fastest wage growth in recent years. From 2000 to 2017, the 95th-percentile wage grew about four times faster than the wages at the median. By 2017, the 95/10 ratio had grown to 6.1 from 5.8 in 2007 and 5.4 in 2000 (see Table 1). This means that on an hourly basis, the 95th-percentile wage earner was paid 6.1 times what the 10th-percentile wage earner was paid. Similar trends are found in the 95/50 wage ratio, with those at the top pulling away from those at the middle. In 2017, the 95th-percentile wage earner was paid 3.3 times more than the median worker compared with 3.0 times more in 2007 and 2.8 times more in 2000.

Figure C

Wages by gender

Men are paid more than women; wage inequality is higher and growing more among men than among women.

Analyzing wages at different points in the wage distribution over time can mask different outcomes for men compared with women. Table 2 replicates the analysis of wage deciles for men and women separately, with a comparison of gender wage disparities over 2000–2017. Figures D and E accompany this table, illustrating the cumulative percent change over 2000–2017 in real hourly wages of men and women at select wage percentiles.

Table 2

Long-term trends suggest that low- and middle-wage men have fared comparatively poorly and that wage gaps between the top and the middle (the 95/50 ratio) and the top and the bottom (the 95/10 ratio) have increased more for men than for women. Male wages at the 95th percentile grew 28.9 percent from 2000 to 2017, twice as fast as at the 90th percentile (14.1 percent), while the median male wage barely budged, rising only 0.6 percent over the entire 17-year period. From 2016 to 2017, men saw their wages fall at the top and bottom of the wage distribution: a 0.9 percent drop at the 95th percentile and 0.8 percent and 0.6 percent decline at the 10th and 20th percentiles, respectively. In the last year, the median male wage grew a more respectable 1.2 percent, bringing the median wage to just above its 2007 and 2000 levels.

Figure D

Figure E

Women also experienced a growth in wage inequality from 2000 to 2017, with the 95th percentile continuing to pull away from the middle and bottom of the wage distribution. However, wage inequality among women in 2017 was not as high as it was among men; the 95th-percentile woman was paid 5.5 times more than the 10th-percentile woman, while the 95/10 ratio among men was 7.0. While inequality has grown modestly among women, the growth in women’s wages is more broadly shared across the wage distribution than men’s, with stronger growth among the bottom 70 percent than among the top earners from 2016 to 2017. In addition, women’s wages at all deciles increased from 2016 to 2017, and women at all deciles had higher wages in 2017 than in 2007 or 2000.

While significant gender wage gaps remain across the wage distribution, the gender wage gap at the median saw some continued improvement, with the typical woman now earning 84 cents on the male dollar. If we can stem the tide of rising inequality and claw back the disproportionate gains going to those at the top of the overall wage distribution, it would be economically feasible to see both men’s and women’s wages rise while simultaneously closing the gender wage gap (EPI 2018a). After widening from 2015 to 2016, the gender wage gap at the top of the wage distribution narrowed somewhat from 2016 to 2017, though it remains wider than it was in 2000. Over the last year, the gender wage gap at the bottom of the distribution narrowed and remains the narrowest across the distribution, likely because of the wage floor.

The regression-adjusted average gender wage gap (controlling for education, age, race, and region) showed a small narrowing to 22.0 percent and remains relatively low by historical standards: in 1979, it was 38.7 percent.2

Wage growth and the minimum wage

Wage growth at the bottom was faster in states that increased their minimum wage in 2017.

In 2017, the minimum wage was increased in 14 states and the District of Columbia through legislation and in seven states because the minimum wage is indexed to inflation in those states. Most of these increases occurred at the start of the year, though some occurred later in the year. Figure F displays in green the states with legislated minimum wage increases in 2017; states in blue had automatic increases resulting from indexing the minimum wage to inflation. Workers in states that increased their minimum wage in 2017 account for about 50 percent of the U.S. workforce. Comparing the average minimum wage in 2016 with the average in 2017, the amounts of the nominal minimum wage increases, legislated or indexed, ranged from $0.05 (or 0.5 to 0.6 percent) in Alaska, Florida, Missouri, and Ohio to $1.95 (or 24.2 percent) in Arizona.

Figure F

A comparison of 10th-percentile wage growth among states grouped by whether they had a minimum wage increase or not yields highly suggestive results. As shown in Figure G, when looking at 10th-percentile wages, growth in states without minimum wage increases was slower (1.7 percent) than in states with any kind of minimum wage increase (2.1 percent). While admittedly a very small differential when disaggregated, this result does hold true for both men and women at the 10th percentile. The 10th-percentile men’s wage grew 2.1 percent in states with minimum wage increases, compared with 1.9 percent growth in states without any minimum wage increase, while women’s 10th-percentile wage grew 1.3 percent in states with minimum wage increases and 1.2 percent in states without.

Figure G

It is not surprising that these differences are smaller than what has been seen in earlier years because as the economy gets closer to full employment, we would expect the 10th-percentile wage to increase across all states regardless of changes in the minimum wage (Gould 2017). Furthermore, 2017 changes in the minimum wage came on the heels of other recent changes to the minimum wage in many of the same states over the previous couple of years. In fact, when we compare states with any minimum wage change since 2013 with those without any, as shown in Figure H, the pattern is even more pronounced. Wage growth at the 10th percentile in states with at least one minimum wage increase from 2013 to 2017 was more than twice as fast as in states without any minimum wage increases (5.2 percent vs. 2.2 percent). As expected given women’s lower wages in general, this result is even stronger for women (5.1 percent vs. 0.8 percent).

Figure H

Wages by race/ethnicity

From 2000 to 2017, within-group wage inequality grew for white, black, and Hispanic workers.

Table 3 examines wage deciles (and the 95th-percentile wage) for white non-Hispanic, black non-Hispanic, and Hispanic workers from 2000 to 2017. From 2000 to 2017, the strongest growth among white, black, and Hispanic workers occurred at the top of the wage distribution, a sign of growing within-group wage inequality. At every decile and at the 95th percentile, wage growth since 2000 has been faster for white and Hispanic workers than for black workers. After suffering declines in the aftermath of the Great Recession, the 20th through 70th percentiles of the black wage distribution are below or within $0.03 of their 2000 levels.

Table 3

From 2016 to 2017, the strongest wage growth among white workers was at the 95th percentile (3.5 percent), while the median and the 10th-percentile wage fell 0.5 and 0.7 percent, respectively. Hispanic workers experienced more broadly based wage growth with wages increasing across their wage distribution, with stronger growth at the median (3.6 percent) and the bottom (2.6 percent) than at the top (0.5 percent). Black wages fell across nearly the entire wage distribution from 2016 to 2017. (Again, when looking at all of these numbers, we need to keep in mind that the CPS data is subject to a certain amount of volatility from year to year; for data on black wages, that volatility is likely to be even more pronounced because of the smaller data sample represented by the black population.) The only notable diversions from these losses were increases at the 10th percentile (1.9 percent) and the 30th percentile (2.7 percent). It’s not surprising that the 10th-percentile wages grew faster among black and Hispanic workers as their 10th-percentile wage is lower and more likely to be impacted by the minimum wage changes discussed above.

The bottom section of Table 3 displays wage disparities, showing black and Hispanic wages as a share of white wages at each decile of their respective wage distributions. Compared with white workers, black workers have been losing ground since 2000, with larger black–white wage gaps across the entire distribution. In 2000, black wages at the median were 79.2 percent of white wages. By 2017, they were only 74.6 percent of white wages. Conversely, Hispanic workers have been slowly closing the gap with white workers at the bottom 70 percent of the wage distribution. In 2000, median Hispanic wages were 68.5 percent of white wages, and, by 2017, they were 74.3 percent. The 95th-percentile Hispanic–white wage gap still remains wider than its 2000 level.

The regression-adjusted black–white and Hispanic–white wage gaps (controlling for education, age, race, and region) have become larger over the last year (EPI 2018d). While the Hispanic–white wage gap has narrowed slightly over the last 17 years, the black–white gap remains significantly larger today (16.2 percent) than it was in 2000 (10.2 percent). In 2000, the Hispanic–white wage gap was larger than the black–white wage gap. In 2017, the reverse was true. Further, between 2000 and 2017 the regression-adjusted black–white wage gap widened significantly for both men (+5.3 percentage points) and women (+6.3 percentage points), while the Hispanic–white wage gap narrowed for men (−2.1 percentage points) and grew slightly for women (+1.7 percentage points).

Wages by education level

Wage growth has generally been faster among the more educated, particularly among men, since 2000.

Table 4 presents the most recent data on average hourly wages by education for all workers and by gender, and Figure I displays the cumulative percent change in real average hourly wages by education. (The discussion throughout identifies each group as mutually exclusive such that those identified as having a college degree have no more than a bachelor’s degree. Those identified as having “some college” may have an associate degree or have completed part of a four-year college degree.)

From 2000 to 2017, the strongest wage growth occurred among those with advanced degrees (7.1 percent), those with college degrees (6.5 percent), and those with less than a high school diploma (7.0 percent). The gains among those with less than a high school diploma were particularly striking in the last couple of years and grew the most of any group from 2016 to 2017 (2.8 percent). Given that these are often the lowest-wage workers in general, it is likely that some of these gains can be attributed to state-level increases in the minimum wage. Over the last year, average wages of those with some college, college degrees, and advanced degrees actually fell, a reversal in trend for the more educated workers from the previous couple of years (EPI 2018d). Workers with some college still have lower wages today than in 2007 or 2000.

Table 4

Figure I

Figures J and K display the cumulative percent change in real hourly wages by education for men and women, respectively. Since 2000, wage growth for those with a college or advanced degree was faster for men than for women, while wage growth for those with a high school diploma or some college was faster (or less negative) for women than for men. In general, the women’s wage distribution by educational attainment is more compressed, that is, the wage differences between workers of different levels of education are not as large for women as they are for men.

For both men and women, the largest gains since 2000 were among those with a college or advanced degree. Wages of men with a high school diploma and with some college remain lower than their 2000 levels. Among women, all groups except for those with some college have now exceeded their 2000 wage levels.

Figure J

Figure K

While there has been a slow narrowing of gender wage gaps for those with high school and some college since 2000, gender wage gaps are wider among those with less than high school or a college or advanced degree. As Figure L illustrates, women are paid consistently less than their male counterparts at every education level.

Figure L

Wage growth for white, black, and Hispanic workers tended to be faster for those with more education than those with less from 2000 to 2017 (Table 5). Average wages grew faster among white and Hispanic workers than black workers for all education groups (which is not surprising given that the same was true at all deciles of the wage distribution). Black workers with less than a college degree have lower wages today than in 2007 or 2000. Consistent with our findings on the relationship between education and earnings for all workers (see Table 4), wage growth was weakest (or fell outright) for those with a college or advanced degree for all groups over the last year, while wage growth was strongest for both black and white workers with less than a high school diploma.

Table 5

Black–white wage gaps by education were larger in 2017 than in 2000 for all education groups, while Hispanic–white wage gaps were narrower for workers with less than high school and high school diploma levels of education. At every education level, black and Hispanic workers were consistently paid less than their white counterparts in 2017, while Hispanic workers were consistently paid more than black workers (Figure M).

Figure M

Wage inequality and the college wage premium

The college wage premium increased from 2000 to 2017, but not fast enough to explain growing wage inequality.

As discussed in the previous section, wage growth among those with an advanced degree or college degree rose 7.1 percent and 6.5 percent, respectively, from 2000 to 2017, while the wages of those with a high school diploma were only 2.1 percent higher than in 2000 (see Figure I). Because of the disproportionate gains for those with more credentials, it’s not surprising that the college wage premium—the regression-adjusted log-wage difference between the wages of college-educated and high school–educated workers—grew from 46.8 percent to 49.5 percent from 2000 to 2017. This rise in the college premium is primarily driven by increases for men, but their college premium actually fell from 2016 to 2017 with the decline in the average college wage (EPI 2018d).

A prevalent story explains wage inequality as a simple consequence of growing employer demand for skills and education—often thought to be driven by advances in technology. According to this explanation, because there is a shortage of skilled or college-educated workers, the wage gap between workers with and without college degrees is widening. This is sometimes referred to as a “skill-biased technological change” explanation of wage inequality. However, despite its great popularity and intuitive appeal, this story about recent wage trends being driven more and more by a race between education and technology does not fit the facts well, especially since the mid-1990s (Mishel, Shierholz, and Schmitt 2013). Furthermore, changes in relative demand for college-educated versus high school–educated workers can have a direct effect on the college wage premium from either side of the equation. Often, these changes—e.g., globalization, deunionization, lowering of the real minimum wage—serve to lower the high school graduate’s wage and thus raise the relative wage of college graduates. That’s not what we’re seeing happening here.

Even among college graduates, there has been a significant pulling away at the very top of the wage distribution. The bottom 50 percent of those with just a college degree still have lower wages than they did in 2000 or 2007. The 50th-percentile wage among those with bachelor’s degrees was 2.1 percent lower in 2017 than it was in 2000, while the 95th-percentile wage of those with bachelor’s degrees was 44.3 percent higher (not shown). The more salient story is not one of a growing differential of wages between college and high school graduates, but increasingly one of growing wage inequality overall and within various education groups.

Figure N shows that from 1979 to 2000, the log 95/50 wage ratio grew at roughly the same pace as the wage gap between college-educated workers and high school–educated workers. While this correspondence shouldn’t be overinterpreted as education driving the 95/50 wage gap, it is true that they both grew at similar rates. The regression-adjusted college wage premium continued to grow in the 2000s and 2010s, though at a slower rate than in the 1980s and 1990s. In fact, it had slowed considerably by the mid-1990s (Bivens et al. 2014). When we compare the relative size of the changes in each gap from 2000 to 2017, it is clear that gains in the college wage premium have not been large enough to drive the continued steady growth of the 95/50 wage gap.

Figure N

Conclusion

From 2016 to 2017, real hourly wages increased for many workers across the wage distribution though not for all genders and races or ethnicities nor for workers of all levels of educational attainment. A particularly bright spot in the data continues to be solid wage growth at the 10th percentile, particularly in states that have increased their minimum wage. In general, though, the years since 2000 have been associated with a continued pulling apart of the wage distribution with disproportionate gains at the top. Wages for those with additional schooling remain higher than wages for workers with less education, though modest increases in the college wage premium cannot explain the more extreme pulling away of the top earners.

Rising wages over the last few years have happened during a period of falling unemployment, with unemployment rates dropping near to (or even below) pre-Great Recession lows. This is no coincidence. If the unemployment rate is allowed to continue to fall, eventually low unemployment should boost workers’ leverage enough to see steady and large wage gains. However, there is no sign that we’ve reached the limits of how much we can sustainably boost wage growth with lower unemployment—wage growth remains weaker than we should expect in a fully healthy economy. This means that confident proclamations that we’ve achieved full employment should not be made and that the Federal Reserve should hold off on any further interest rate increases and allow the economy to continue to grow.

Full employment is one way that workers gain enough bargaining power to increase their wages; employers have to pay more to attract and retain the workers they need when idle workers are scarce. The “lever” for higher wages that comes from full employment is most important for workers at the bottom of the wage distribution: for a given fall in the unemployment rate, wage growth rises more for low-wage workers, and in the absence of stronger labor standards, it is often only in the tightest of labor markets that low-wage workers see stronger wage growth (Gould, Davis, and Kimball 2015).

Beyond seeking to keep labor markets tight, policymakers could take other steps to foster strong broad-based wage growth, such as raising the federal minimum wage, expanding eligibility for overtime pay, addressing gender and racial pay disparities, and protecting and strengthening workers’ rights to bargain collectively for higher wages and benefits. For more policies that will raise wages, see EPI’s Agenda to Raise America’s Pay (EPI 2016).

About the author

Elise Gould, senior economist, joined EPI in 2003. Her research areas include wages, poverty, economic mobility, and health care. She is a co-author of The State of Working America, 12th Edition. In the past, she has authored a chapter on health in The State of Working America 2008/09; co-authored a book on health insurance coverage in retirement; published in venues such as The Chronicle of Higher Education, Challenge Magazine, and Tax Notes; and written for academic journals including Health Economics, Health Affairs, Journal of Aging and Social Policy, Risk Management & Insurance Review, Environmental Health Perspectives, and International Journal of Health Services. She holds a master’s in public affairs from the University of Texas at Austin and a Ph.D. in economics from the University of Wisconsin at Madison.

Acknowledgments

The author thanks EPI research assistant Julia Wolfe and EPI data programmer Jin Dai for their valuable contributions to this study.

Appendix

Appendix Figure A

Appendix Figure B

Appendix Figure C

Endnotes

1. For more information about the CPS and CES employment measures, see BLS 2018.

2. Regression-adjusted figures are not shown in the tables in this report but are available in the State of Working America Data Library (EPI 2018d).

References

Bivens, Josh, Elise Gould, Lawrence Mishel, and Heidi Shierholz. 2014. Raising America’s Pay: Why It’s Our Central Economic Policy Challenge. Economic Policy Institute Briefing Paper no. 378.

Bureau of Labor Statistics (BLS). 2018. “Comparing Employment from the BLS Household and Payroll Surveys” (webpage). Last updated February 2, 2018.

Current Population Survey Outgoing Rotation Group microdata. Various years. Survey con­ducted by the Bureau of the Census for the Bureau of Labor Statistics [machine-readable microdata file]. Washington, D.C.: U.S. Census Bureau.

Current Population Survey public data series. Various years. Aggregate data from basic monthly CPS microdata are available from the Bureau of Labor Statistics through three pri­mary channels: as Historical ‘A’ Tables released with the BLS Employment Situation Summary, through the Labor Force Statistics Including the National Unemployment Rate database, and through series re­ports.

Economic Policy Institute (EPI). 2016. The Agenda to Raise America’s Pay. Last updated December 6, 2016.

Economic Policy Institute (EPI). 2018a. Gender Pay Gap Calculator. Last updated March 1, 2018.

Economic Policy Institute (EPI). 2018b. Methodology for Measuring Wages and Benefits. Last updated March 1, 2018.

Economic Policy Institute (EPI). 2018c. Nominal Wage Tracker. Last updated February 2, 2018.

Economic Policy Institute (EPI). 2018d. State of Working America Data Library.

Economic Policy Institute (EPI). 2018e. Wage Calculator. Last updated March 1, 2018.

Gould, Elise. 2003. Measuring Employment Since the Recovery: A Comparison of the Household and Payroll Surveys. Economic Policy Institute Briefing Paper no. 148.

Gould, Elise. 2017. The State of American Wages 2016: Lower Unemployment Finally Helps Working People Make Up Some Lost Ground on Wages. Economic Policy Institute.

Gould, Elise, Alyssa Davis, and Will Kimball. 2015. Broad-Based Wage Growth Is a Key Tool in the Fight against Poverty. Economic Policy Institute Briefing Paper no. 339.

Kopczuk, Wojciech, Emmanuel Saez, and Jae Song. 2010. “Earnings Inequality and Mobility in the United States: Evidence from Social Security Data Since 1937.” Quarterly Journal of Economics vol. 125, no. 1, 91–128.

Mishel, Lawrence, Heidi Shierholz, and John Schmitt. 2013. Don’t Blame the Robots: Assessing the Job Polarization Explanation of Growing Wage Inequality. Economic Policy Institute, Center for Economic and Policy Research Working Paper.

Social Security Administration. Various years. Wage Statistics [database].

Why it matters: Given that wage theft disproportionately affects workers from low-income households—who are already struggling to make ends meet—the loss of wages can be devastating. And these recovery numbers likely dramatically underrepresent the pervasiveness of wage theft—it has been estimated that low-wage workers lose more than $50 billion annually to wage theft. Regardless of what share of actual wage theft the recovery numbers represent, these data are one more reminder that wage theft is not isolated to a few bad employers, but affects workers much more broadly.

What can be done about it: Implement legislation to improve pay transparency; increase penalties for wage theft violations; support strong government enforcement of wage and hour laws; protect workers from retaliation when they report violations; and protect worker rights to collective and class action.

Introduction

The last four decades have been marked by rising wage inequality, with the vast majority of American workers experiencing wage stagnation while those at the top rung of the economic ladder reap the benefits of growth in productivity. These dynamics mean that many workers struggle to make ends meet; in 2016 one in five families in which at least one person worked were living below 200 percent of the federal poverty line (U.S. Census Bureau 2017).1 This situation is deeply exacerbated by wage theft, which continues to rob workers of billions of dollars in earned pay each year, with low-income workers being disproportionately affected (Bernhardt et al. 2009).

Wage theft occurs when employers fail to pay workers the full wages to which they are entitled for their labor. This includes, for example, refusing to pay workers the total amount of promised wages, not paying for time spent preparing a workstation at the start of a shift or closing up at the end of a shift, and not paying overtime premiums to workers who work more than 40 hours a week. Consider a full-time minimum wage worker earning the federal minimum wage of $7.25 an hour, around $15,000 per year. If this worker’s employer asks her to work 15 minutes “off the clock” before and after her 8-hour shift each day, that extra half hour of unpaid work each day represents a loss to the worker (and a gain to the employer) of around $1,400 per year, including the overtime premiums she should have been paid. This constitutes theft of nearly 10 percent of a minimum wage employee’s annual earnings—which can mean the difference between paying the rent and utilities or risking eviction or the loss of gas, water, or electric service.

Background and prior studies on wage theft

Wage theft costs workers billions of dollars in earned wages each year. In 2008, Bernhardt et al. surveyed front-line workers2 in low-wage industries in the cities of Chicago, Los Angeles, and New York and found that two-thirds (68 percent) of these workers experienced at least one pay-related violation in any given week (Bernhardt et al. 2009). The researchers estimated that the average cost to these workers over a year was $2,634 out of a total earnings of $17,616—15.0 percent of their wages.3 This adds up to a total of nearly $3 billion annually stolen across all forms of wage theft among these workers in 2008.4 If we generalize the 2008 results nationwide and then update the numbers to reflect nominal wage growth and employment growth from 2008 to 2016, we can estimate that low-wage workers in the U.S. lost more than $50 billion to all forms of wage theft in 2016.5

A recent EPI study (Cooper and Kroeger 2017) found that one particular form of wage theft, paying workers less than the minimum wage, impacts around 17 percent of low-wage workers in the 10 most populous states (California, Florida, Georgia, Illinois, Michigan, New York, North Carolina, Ohio, Pennsylvania, and Texas) and costs workers in those states around $8 billion per year. Should these findings hold true nationally, it is reasonable to assume that minimum wage violations alone are costing workers at least $15 billion per year. This means that stolen wages represent far more than FBI estimates of the total annual value of all robberies, burglaries, larceny, and motor vehicle theft in the United States, at $12.7 billion (Cooper and Kroeger 2017).

If the overall cost of wage theft is staggering, so are the costs to individual workers. Cooper and Kroeger (2017) find that workers suffering minimum wage violations are cheated out of $64 a week—$3,300 annually for year-round workers. These workers lose almost one-quarter of their earnings, receiving on average only $10,500 in annual wages instead of the $13,800 they should have received. Furthermore, Cooper and Kroeger find that while minimum wage violations affect workers broadly across demographic categories, disproportionate shares of wage theft victims are people of color, women, immigrants, young people, workers from modest-income households, nonunionized workers, and workers who have do not have a bachelor’s degree.

Most workers whose wages are stolen do not report the violation. When workers do report wage theft, state departments of labor and attorneys general often lack adequate enforcement resources. Fourteen states, most of which use the federal minimum wage, either lack the capacity to investigate wage theft claims or lack the ability to file lawsuits on behalf of victims (Galvin 2016).6 These states effectively defer to the federal government for enforcement. In 2016, the U.S. Department of Labor Wage and Hour Division, responsible for enforcing wage and hour laws, had around 1,000 investigators responsible for 7.3 million workplaces (U.S. DOL 2017). Given this, the probability of any specific workplace being investigated for wage theft in a given year is minuscule.

Findings of this study

In this study, we seek to contribute to the knowledge base on wage theft by collecting and aggregating available data on recovery of stolen wages in the United States. In order to get as comprehensive a picture as possible of wage recovery across the United States, we reviewed current U.S. Department of Labor wage enforcement data, surveyed state labor departments and attorneys general, and mined data from a report on private civil litigation class action settlements.

Wage recovery by the U.S. Department of Labor

U.S. DOL recovered $246.8 million from wage and hour violators in 2015 and $266.6 million in 2016, for a total of $513.3 million over both years (WHD 2017).7 See Appendix Table A1.

Wage recovery by state departments of labor and attorneys general

State departments of labor and attorneys general in 39 states recovered $170.0 million in 2015 and $147.5 million in 2016, for a total of $317.5 million over both years.8 See Appendix Table A2 for a breakdown by state.

Wage recovery through class action settlements

The top 10 wage and hour class action settlements in 2015 totaled $463.6 million; in 2016, they totaled $695.5 million, for a total of $1.2 billion over both years (Seyfarth Shaw LLP 2017).

The value of the top 10 wage and hour class action settlements has risen dramatically in recent years (see Figure A); the 2016 wage and hour class action settlements figure is the largest in the last decade (Seyfarth Shaw LLP 2017, 9). The growth in private litigation of these cases is indicative both of the extent of the wage theft problem and of the inadequacy of enforcement resources at the government level.

Figure A

Total wage recovery from these sources

Altogether, amounts recovered by U.S. DOL, by states, and through class action settlements totaled $880.3 million in 2015 and $1.1 billion in 2016, for a total of $2 billion over both years.

It is important to note that these data do not represent the amount of wages stolen from workers each year. In fact, U.S. DOL and state recovery figures likely dramatically underrepresent the problem of wage theft, since probably only a fraction of victims actually file complaints with the government. Similarly, only a fraction of victims are involved in class action settlements.

Conclusion: What can be done about wage theft

These data make it clear that wage theft is a widespread epidemic across our economy and not merely the practice of a few unscrupulous employers. This systemic violation of our nation’s most basic principle of labor and employment policy—that working people should be paid what they are owed for their labor—requires immediate action. Lawmakers must institute policies that combat wage theft. These reforms are not complicated initiatives; rather, they are commonsense measures that advocates have advanced for decades. These policies include:

• Raising the cost to employers for violating the law. Research demonstrates that meaningful penalties can have a deterrent effect on wage theft. Lawmakers should enact triple-damages laws that require an employer who has violated the law to pay a worker three times the amount of wages owed.
• Improving transparency. Many workers do not receive information from their employers about the basic terms and conditions of their employment. This lack of transparency makes it difficult for workers to establish that they are not being paid fairly. Lawmakers should ensure that every worker gets a statement of pay that shows rate of pay, hours worked, and deductions from pay.
• Supporting strong government enforcement. The U.S. Department of Labor and state agencies tasked with enforcing wage and hour protections must be given adequate resources to enforce the law.
• Protecting workers from retaliation. A national survey found that 43 percent of workers who complained to their employer about their wages or working conditions experienced retaliation (Bernhardt et al. 2009). This has a chilling effect on the entire workplace, which leaves all workers more vulnerable. Meaningful penalties for retaliation would be an effective way to deter employers from retaliating and to compensate workers who experience retaliation. Enabling enforcement agencies to receive anonymous worker complains or permitting third parties (such as unions or worker centers) to file complaints on behalf of workers also limits employer retaliation.
• Protecting workers’ right to class action. Workers depend on collective and class actions to enforce many workplace rights. Employment class actions are fundamental to the enforcement of wage and hour standards and have helped to combat race and sex discrimination. Without the ability to aggregate claims, it would be very difficult if not impossible for workers, particularly low-wage workers, to find legal representation in these matters. It is critical that workers not be forced to sign away their right to class action as a condition of employment.9

Wage theft is devastating to the workers whose wages are stolen and to their families. Victims of wage theft largely represent low-income households—who need every dollar they earn to pay for basics like rent, utilities, and groceries. To protect these workers’ rights to be paid the wages they have worked hard for, lawmakers must act to ensure that employers are held responsible for abiding by wage and hour laws and paying workers the wages they are owed.

Appendix of tables

Table A1

Table A2

Endnotes

1. Two hundred percent of the federal poverty line is commonly used as the threshold below which families are considered “low-income.”

2. Bernhardt et al. define “front-line worker” as “not a manager, professional or technical worker” (Bernhardt et al. 2009, 12). The workers surveyed were also specified to be “age 18 or older.”

3. Total earnings of $17,616 are the earnings these workers should have received. They actually received $17,616 – $2634 = $14,982.

4. Bernhardt et al. (2009) estimate that, in 2008, “front-line workers in low-wage industries in Chicago, Los Angeles and New York City los[t] more than $56.4 million per week [equivalent to nearly $3 billion annually] as a result of employment and labor law violations.”

5. Generalizing these results from 2008 to the 2016 nationwide workforce in low-wage industries requires a few adjustments. First, the three-city low-wage workforce in Bernhardt et al.’s (2009) study represented 15.1 percent of all workers in those cities, so scaling up to the national level requires multiplying by 14.3, the ratio of the size of 15.1 percent of the national workforce relative to the size of 15.1 percent of the three-city workforce. (For the size of the national workforce, we use counts of all wage earners from the Social Security Administration [SSA 2017] as the measure of employment.) Second, wage theft will have increased because nominal wages have grown (16.6 percent at the 10th percentile) and overall employment has grown (5.2 percent) between 2008 and 2016. (Low-wage workers in Bernhardt et al.’s study had a median wage of roughly $8.00, which corresponds to the 10th percentile wage in 2008; the nominal increase in the 10th percentile wage from 2008 to 2016 was 16.6 percent [EPI 2017]. According to the SSA employment counts [SSA 2017], overall employment grew 5.2 percent.) These adjustments yield an estimate of $52 billion a year that workers lost to all forms of wage theft in 2016.

6. Data are provided in an Excel spreadsheet attached to Galvin 2016, which is downloadable online.

7. Totals are computed using unrounded numbers. Data may not sum to totals due to rounding.

8. State-level data for this survey were collected by contacting state departments of labor and attorneys general and requesting data on recovered wages. We collected data from 39 states; for 11 other states, we either did not receive a response to our request or the states do not track the requested data. These data are available upon request.

9. The Supreme Court is currently considering whether workers can be required to waive their right to pursue work-related claims on a class action basis. The National Labor Relations Board has found that forced arbitration agreements that interfere with workers’ right to act together to improve their wages and working conditions violate the National Labor Relations Act. If the Court reaches a different conclusion, workers may lose the ability to pursue workplace disputes as class actions. Workers depend on class actions to enforce workplace rights and the class action settlement data makes it clear that class actions are fundamental to the enforcement of wage and hour standards. See McNicholas 2017.

References

Bernhardt, Annette, et al. 2009. Broken Laws, Unprotected Workers: Violations of Employment and Labor Laws in America’s Cities, 2009. Center for Urban Economic Development, National Employment Law Project, and UCLA Institute for Research on Labor and Employment.

Bureau of Labor Statistics (BLS). 2017. “Table 3. Employees on Nonfarm Payrolls by State and Selected Industry Sector, Seasonally Adjusted.” Local Area Unemployment Statistics (LAUS). Last modified November 17, 2017. Accessed December 8, 2017, at https://www.bls.gov/news.release/laus.t03.htm.

Cooper, Dave, and Teresa Kroeger. 2017. Employers Steal Billions from Workers’ Paychecks Each Year: Survey Data Show Millions of Workers Are Paid Less Than the Minimum Wage, at Significant Cost to Taxpayers and State Economies. Economic Policy Institute, May 10.

Economic Policy Institute (EPI). 2017. “Wages by Percentile” [data set]. In State of Working America Data Library. Accessed December 8, 2017, at www.epi.org/data/.

Galvin, Daniel. 2016. Supplementary materials for “Deterring Wage Theft: Alt-Labor, State Politics, and the Policy Determinants of Minimum Wage Compliance.” Perspectives on Politics vol. 14, no. 2.

McNicholas, Celine. 2017. “Supreme Court Should Uphold Working People’s Fundamental Rights in Murphy Oil.” Working Economics (Economic Policy Institute blog), September 27.

Seyfarth Shaw LLP. 2017. “Chapter I. Overview of the Year in Workplace Class Action Litigation.” In 13th Annual Workplace Class Action Litigation Report.

Social Security Association (SSA). 2017. “Wage Statistics for 2008” and “Wage Statistics for 2016.” Data tables accessed December 7, 2017, at www.ssa.gov/cgi-bin/netcomp.cgi.

U.S. Census Bureau. 2017. “2017 CPS Poverty Table: POV06. Families by Number of Working Family Members and Family Structure. Poverty Status in 2016. Below 200 Percent of Poverty. All Races” [XLS file]. Current Population Survey, 2017 Annual Social and Economic Supplement.

U.S. Department of Labor (U.S. DOL). 2017. Working for You: U.S. Department of Labor, 2009–2016.

Wage and Hour Division (WHD). 2017. “WHD: All Acts” [data table]. Fiscal Year Data for WHD. U.S. Department of Labor. Accessed December 5, 2017.

Why it matters: For decades, the drag on demand growth stemming from rising inequality has been compensated for by other economic and policy developments—notably a long-running decline in interest rates. Going forward, however, these compensating mechanisms are likely to fail, which means that the inequality-induced drag on demand would translate directly into slower economic growth overall.

What we can do about it: In the near term, we need more expansionary macroeconomic policies—lower interest rates and larger budget deficits—to counter the downward pressure on demand. In the longer run, we need to stop or reverse rising inequality by enacting policies that spur faster wage growth for low- and middle-wage workers. Raising these workers’ wages would not only raise living standards for American families, it would also ensure robust economic growth.

Executive summary

Income inequality in the United States has risen dramatically since the late 1970s, and the issue has drawn heightened attention in recent years. In the past decade, economic observers have also become increasingly worried about “secular stagnation”—or a chronic shortfall of aggregate demand, fearing that this shortfall will constrain American economic growth in coming years. These two phenomena—rising inequality and chronic weakness of demand—are related. Specifically, rising inequality transfers income from low-saving households in the bottom and middle of the income distribution to higher-saving households at the top. All else equal, this redistribution away from low- to high-saving households reduces consumption spending, which drags on demand growth.

This paper argues that a key lever for solving the problem of secular stagnation is halting, or even reversing, the root cause of rising inequality: the growing wedge between productivity and pay for typical American workers. Following are our key findings:

• A stunningly large upward redistribution of income has characterized the American economy in recent decades. In 1979, the bottom 90 percent of American households claimed roughly 70 percent of total income in the U.S. economy. By 2016, this share had fallen to around 60 percent.

• By far the most important driver of this upward redistribution is the growing wedge between economy-wide productivity growth (a measure of income generated in an average hour of work in the United States) and hourly pay of typical American workers since the mid-1970s. Had these two measures grown together the way they did in earlier eras, there would have been no possibility of upward income redistribution.

• A strong and growing body of macroeconomic evidence shows that the U.S. economy needs lower and lower interest rates simply to provide the same growth of aggregate demand over time. In short, something is pushing down the growth rate of aggregate demand, and macroeconomic policies need to become more and more expansionary in each successive year simply to hold demand constant. This development has sometimes been labeled “secular stagnation.”

The rise in inequality has contributed significantly to the downward pressure on demand growth that is labeled secular stagnation. Inequality has transferred income from low- and middle-income households with relatively low savings rates towards higher-income households with higher savings rates. All else equal, this transfer drags on demand growth as consumption grows more slowly. This transfer will likely slow growth in aggregate demand by an estimated 2 to 4 percentage points of gross domestic product (GDP) every year going forward from today. 1

• The demand drag imposed each year by rising inequality is equivalent to the peak boost to economic demand provided (in 2010) by the American Recovery and Reinvestment Act (ARRA, the stimulus package passed in the first months of the Obama administration). Essentially to offset the hit to demand posed by rising inequality, we’d need to enact a policy each and every year that delivers a boost of the rough magnitude of the peak ARRA boost.

• One puzzle that arises in the causal chain linking rising inequality to slower demand growth is that the personal savings rate measured in the National Income and Product Accounts (NIPA) has actually declined, not risen over recent decades. However, a closer analysis of the data and economics behind savings behavior shows that the declining savings rate measured conventionally in the NIPAs does not capture many ways in which savings of high-income households have increased. The most important fact is that unrealized capital gains spurred by corporate stock buybacks are not captured in the NIPA personal savings measure. However, these unrealized gains do constitute large increases in wealth (which is a form of savings) for shareholders.

Introduction

The problem of anemic wage growth—recognized for decades by American workers wishing for higher paychecks—has finally reached the front-burner of American politics. Angst over the stagnant pay of low-wage workers has for example, sparked recent movements to raise minimum federal, state, and local minimum wages far above levels that have characterized the recent past, and often even to levels that would constitute historical highs.2

This new attention to the crisis of American pay is totally proper. The failure of wages of the vast majority of Americans to benefit from economy-wide growth in productivity (or income generated in an average hour of work) has been the root cause of the stratospheric rise in inequality and the concentration of economic growth at the very top of the income distribution. Had this upward redistribution not happened, incomes for the bottom 90 percent of Americans would be roughly 20 percent higher today.3 In short, the rise in inequality driven by anemic wage growth has imposed an “inequality tax” on American households that has robbed them of a fifth of their potential income.

There would be huge benefits to American well-being from blocking or reversing this upward redistribution. This welfare gain stemming from blocking upward redistribution is the primary reason to champion policy measures to boost wage growth and lead to a more equal distribution of income gains. Put simply, a dollar is worth more to a family living paycheck to paycheck than it is to families comfortably in the top 1 percent of the income distribution.

Proponents of increases in the minimum wage and other measures to boost American wages have often argued that there are benefits to these policies besides the welfare gains stemming from pure redistribution. These proponents have often argued that boosting wages would even benefit aggregate economic outcomes, like growth in gross domestic product (GDP) or employment.

Recent evidence about developments in the American and global economies strongly indicate that these arguments are correct: boosting wages of the bottom 90 percent would not just raise these households’ incomes and welfare (a more-than-sufficient reason to do so), it would also boost overall growth. For the past decade (and maybe even longer), the primary constraint on American economic growth has been too-slow spending by households, businesses, and governments. In economists’ jargon, the constraint has been growth in aggregate demand lagging behind growth in the economy’s productive capacity (including growth of the labor force and the stock of productive capital, such as plants and equipment). Much research indicates that this shortfall of demand could become a chronic problem in the future, constantly pulling down growth unless macroeconomic policy changes dramatically.

Our rising inequality is being driven by the slowdown in wage growth for the bottom 90 percent

It is now well-known that incomes in America grew much less equally after 1979. Probably the most important fact about this growing inequality is that it has overwhelmingly been driven by trends in market-based income rather than in the taxes and transfers component of income. Table 1 shows the sources of income growth for the top 1 percent of households in the three decades before the Great Recession. It uses Congressional Budget Office (CBO 2016) data on comprehensive household income, which includes noncash market-based income such as employer contributions to health insurance premiums as well as non–market-based income such as government transfers. The CBO data show that inequality is increasing (the share of all income that is going to the top is rising) because the top 1 percent are getting a greater share of each type of market income and because the types of market income that are most concentrated at the top (particularly capital gains and business income) constitute a growing share of all income, whereas income from less-concentrated sources (particularly labor compensation) is falling as a share of overall income. The data in the table also indicate that the direct, arithmetic influence of taxes and transfers has been minimal, with rising inequality of market incomes explaining more than 100 percent of the rise in the after-tax income share of the top 1 percent.4

Table 1

The first block of columns simply shows the top 1 percent share of overall household income and of various income types as identified in CBO (2016). A clear finding is that the top 1 percent share of every source of income except government transfers rose significantly between 1979 and 2007. The share of overall income held by the top 1 percent more than doubles (rising from 8.9 to 18.7 percent of total income) between 1979 and 2007. And even with the enormous blow to top 1 percent incomes dealt by the 40 percent loss in the stock market from 2007 to 2010, the top 1 percent share in 2012 of 17.3 percent was almost double its 1979 level. Particularly salient to this analysis is the rough doubling of both labor and total capital shares claimed by the top 1 percent from 1979 to 2007 and 2012.

The next block of columns shows each income category’s share of overall household income. The most striking finding here is the large decline in the labor compensation share of total income, falling from 70.6 percent in 1979 to 61.0 percent in 2007 and 2012. Correspondingly, the share of total capital and business income (driven by capital gains and business income) rose substantially, from 17.5 percent in 1979 to 22.1 percent in 2007.5 Due to the stock market crash in 2007 and the hangover from that crash through 2010, capital income shares (and thus total capital and business income) remained lower in 2012 than in 2007, but still above the 1979 levels. Finally, pension payments and transfer incomes have risen steadily over time as shares of total income.

The third block of columns calculates how much growing concentration within each income category contributed to the increasing top 1 percent share of income from 1979 to 2007 and from 1979 to 2012. The growing concentration of particular income types in the top 1 percent of households contributed 7.2 percentage points to the 9.8 percentage-point increase in the top 1 percent’s income share from 1979 to 2007, accounting for essentially three quarters of the rise. The vast majority of this concentration within income sources is accounted for by labor and capital incomes. The last block of columns summarizes how much the shift from less-concentrated (labor) income to more-concentrated (capital) incomes boosted the top 1 percent share of overall household income. The sum of these shifts contributed 2.6 percentage points to the growth of the top 1 percent share from 1979 to 2007, and 0.4 percentage points from 2007 to 2012.

One way to summarize what these data tell us is that the vast majority of households (those outside the top 1 percent) are losing out in claiming their proportionate share of total income growth in two significant ways. First, workers as a group are losing out to capital owners, with the shift from labor to capital income explaining a significant portion of the rise of the top 1 percent. Second, the bottom 99 percent of income earners in America are able to claim only an ever-shrinking portion of the overall wage bill, with the highest-paid workers in the top 1 percent more than doubling their share of labor income over the last three and a half decades.

In our view, these are simply two sides of the same coin: a pronounced reduction in the collective and individual bargaining power of ordinary American workers that led to pay growth lagging productivity so badly in recent decades. If wages of the bottom 99 percent had kept pace with productivity growth for most of the past generation (the way that typical workers’ wages did in the post-WWII generation), then most of the increase in income inequality we have seen simply would not have had space to develop, as concentration within labor incomes would not have grown and the share of total output available to be claimed by capital owners would have been significantly smaller.6

But wages for the vast majority of workers stopped keeping pace with economy-wide productivity growth in the late 1970s, and the cumulative wedge between productivity and typical workers’ pay has risen ever since, as shown below in Figure A. This figure shows growth in economy-wide productivity, defined as the amount of income and output generated in an average hour of work in the economy. While the pace of productivity growth slowed down in the late 1970s, productivity still grew steadily in the following decades. The figure also shows a measure of hourly pay (including both wages and benefits) for production and nonsupervisory workers in the U.S. economy. This nonmanagerial group includes roughly 80 percent of the private-sector workforce. After growing right in line with productivity for decades following World War II, hourly pay for these workers all but stagnated after 1979. Because productivity kept growing but pay for 80 percent of the private-sector workforce stagnated, this means that the economy continued to generate growing incomes on average each year, but pay for typical workers slowed radically. In short, the growing wedge between these lines represents the disproportionate share of economic growth claimed by those at the top after 1979.

Figure A

Table 1 and Figure A together tell a clear story about the rise in American inequality: it has been made possible by the suppression of wage growth for the vast majority of American workers. Until this wage suppression ends and hourly pay for the vast majority of workers begins rising in lockstep with economy-wide productivity, there is very little reason to hope that rising inequality can be arrested. This makes focusing policy attention on boosting wage growth absolutely crucial.

“Secular stagnation,” or, the chronic shortage of aggregate demand constraining economic growth

A useful (if admittedly too-simple) way to think about an economy’s growth is as an interplay between the economy’s productive capacity and the level of aggregate demand. The economy’s productive capacity is a measure of potential that includes three major “inputs” of production: the labor force, the capital stock, and the state of technology. However, for these potential inputs to be fully utilized, aggregate demand—or spending by households, businesses, and governments—must be strong enough to mobilize them. Take the example of a hotel’s economic fortunes from 2007 to 2010. In 2007, the building and physical plant existed, the systems for taking reservations existed, and there were plenty of workers, both actual employees and potential workers willing to take jobs at the right wages. Also in that year, there were customers; rooms were likely booked to capacity and the owners may have even considered adding rooms. In 2010, this hotel still had a physical plant and reservation systems, and while their own staff was likely much smaller because of layoffs in the wake of the Great Recession, there was a huge increase in potential workers looking for jobs that could have been hired. But what kept the hotel’s hiring constrained and profits low in 2010 was lack of customers, not slow growth in the economy’s potential (or productive capacity).

Recently, a number of economists have noted that evidence over recent decades indicates that growth has been constrained more by slow growth in aggregate demand than by slow growth in the economy’s productive capacity. For example, the full business cycle between the peaks of 2001 and 2007 saw the slowest economic growth then on record. The result of this slow growth was that the unemployment rate never returned to prerecession levels, and the prime-age employment-to-population (EPOP) ratio never approached prerecession levels. (See Bivens and Irons 2008 for a full accounting of this business cycle’s place in historical comparisons.) All of this indicates that the slow growth that took hold even before the Great Recession hit was likely a function of too-slow growth in aggregate demand—or spending by households, businesses, and governments.

Before the Great Recession, most macroeconomists would have rejected the idea that economic growth could be constrained for long periods of time by too-slow demand growth relative to the economy’s productive capacity. The typical view was that growth in productive capacity was driven by long-run trends that did not change very fast, such as the aging of the population (which determines the pace of potential labor force growth), the accumulation of plants, equipment, and buildings that is the result of decades of past investment, and accelerations and decelerations of technology that were largely exogenous (unrelated to the state of the business cycle). In this view, ensuring that growth in productive capacity (or growth in potential GDP) is fully realized essentially means ensuring that aggregate demand grows quickly enough to keep resources (labor and capital) fully employed.

In past decades, policymakers considered it relatively easy to keep aggregate demand growing fast enough high enough to fully utilize the economy’s productive capacity. In fact, macroeconomic policymakers thought their most difficult task was restraining, not boosting, growth in aggregate demand. When aggregate demand for economic output outstrips the economy’s productive capacity to meet that demand, the result is inflation. So policymakers focused on controlling inflation—or ensuring that aggregate demand did not run chronically too fast. Of course, the U.S. economy underwent recessions during which demand growth lagged behind potential GDP growth, but it was thought that the demand shortfalls could be easily solved by the Federal Reserve reducing short-term interest rates to spur more spending. Because aggregate demand was thought to need policy restraint, not stimulus, this implies that overall growth was constrained by how fast the economy’s productive capacity could grow. Any worry that persistently slow growth (say lasting more than one year) in aggregate demand could be a primary constraint on economic growth over a meaningfully long time period was largely dismissed. We now know that this dismissal was premature, and that sluggish demand growth can pull down economic growth for long periods of time.

The data show we are in such a period, and likely have been for over a decade. The extraordinarily weak GDP growth between 2001 and 2007 was accompanied by decelerating wage growth, and low inflation and interest rates. These trends are strong indicators that demand was lagging growth in productive capacity. This weakness in demand was especially striking given that aggregate demand (or spending by households, businesses, and governments) was buoyed in those years initially by near-zero interest rates (set by the Federal Reserve in the early 2000s) and then by an enormous asset bubble in residential real estate that increased household wealth in the mid-2000s. The housing bubble burst, ushering in the Great Recession. The recovery from that recession was even slower than the recovery from the 2001 recession, despite extraordinarily expansionary monetary policy in the wake of the Great Recession.

Figure B shows the ratio of actual GDP to potential GDP since 1995. When this ratio is below 1, there is prima facie evidence that aggregate demand is constraining economic growth. As the figure shows, this ratio has been well below 1 for most of the past two decades, even as monetary policy has been historically expansionary.

Figure B

Weakness in aggregate demand since 2008 has likely degraded growth in the economy’s productive capacity

Aggregate demand growth has lagged growth in the economy’s productive capacity over much of the last two decades even as growth in this productive capacity itself slowed sharply. Crucially, there is ample evidence that the degraded growth in potential output is itself another casualty of too-slack demand. It is now well-known that changes in productive capacity (i.e., the supply side of the economy) are likely affected by changes in the demand-side of the economy. The most obvious example concerns capital investment. When demand is weak, customers disappear and workers’ wages don’t grow as fast, or grow at all, as rising unemployment crushes workers’ bargaining power. A shortage of customers and weak wage growth blunts the incentive of firms to invest in plants or equipment to expand capacity or save on labor costs. This in turn slows the growth of the economy’s capital stock, a key input in its productive capacity. Short recessions will leave only a small scar on an economy’s productive capacity, but there is now ample evidence that longer and steeper recessions can do serious damage to even the economy’s potential, let alone actual, growth. We have clearly seen this dynamic over the past decade.7 Growth in potential GDP has been so much slower than growth that was projected in 2008 for the next decade that forecasters by 2017 have essentially erased $2 trillion annually in potential output from the American economy, as shown in Figure C.

Figure C

If even one quarter of this slowing growth in potential GDP is a function of sluggish demand growth affecting the course of the economy (in economists’ terms, imposing “hysteresis” effects), then the failure to boost demand to grow in tandem with potential GDP has been an economic calamity, costing $500 billion annually.8 Given these stakes, a number of researchers have begun looking into this chronic shortage of demand. Some have labeled this “secular stagnation,” which has unfortunately confused many people, including many economists, by obscuring that this is an argument about aggregate demand, not potential GDP. (Summers 2016 and Krugman 2013 offer some of the clearest statements about the problem of “secular stagnation.”) Recently, the same phenomenon has been recast as a fall in the economy’s “natural” or “neutral” rate of interest. This has also been referred to as the fall in “R-star” or R*, referencing how this natural or neutral rate of interest is often notated in economics papers.9

Lowering interest rates has been our coping mechanism for slow demand growth

The relationship between a chronic shortage of aggregate demand (secular stagnation) and the fall in the economy’s “neutral” interest rate is straightforward. A decline in aggregate demand growth that is exogenous to interest rates will generally spur the Federal Reserve to reduce interest rates in order to stimulate consumption and investment and keep the economy at full employment. Hence, the interest rate consistent with the economy reaching full employment (the neutral rate) will fall over time if this exogenous drag on demand growth continues or worsens.

The evidence that this neutral rate has declined over time is compelling. Figure D shows the federal funds rate (FFR) as well as showing its average value over recent business cycles. The evidence that the neutral FFR rate has been steadily falling since the 1980s seems quite clear.

Figure D

The relationship between secular stagnation and inequality

There is a relatively straightforward causal link between the rise of inequality and the chronic shortfall of aggregate demand: higher-income households have much higher savings rates than low- and middle-income households. So, as a dollar is transferred from the bottom and middle to the top of the income distribution, less of it is spent.

When the economy is functioning well, reduced consumer spending is offset by increased investment in plants and equipment, and demand is hence unaffected. Here is how it could work. The reduced consumption stemming from this redistribution translates mechanically into higher savings. Higher savings in turn puts downward pressure on interest rates, and these lower interest rates induce more business investment in plants and equipment. This interest rate adjustment hence ensures that the reduced consumption spending that follows the upward redistribution of income is matched by an increase in investment spending, and hence does not constitute a drag on growth in aggregate demand.

The zero lower bound on interest rates means future drags on demand will translate into slowdowns in growth

Keeping demand growth constant in the face of upward redistribution of income requires ever-lower interest rates, consistent with the data highlighted above. But if this neutral rate falls far enough, the economy’s ability to seamlessly translate lower consumption (or higher savings) into higher investment spending will be potentially blocked by the zero lower bound (ZLB) on interest rates. Interest rates cannot be moved below zero (or at least not too much below zero) for extended periods of time because negative interest rates will just induce households to hold their wealth in cash rather than interest-bearing (or interest-subtracting!) bonds. With the economy’s ability to translate additional savings into higher investment blocked at the ZLB, further increases in savings will instead show up as unused capacity and output losses rather than interest rates reductions. Figure D showed that the U.S. economy has spent most of the post-2001 period hovering perilously close to the ZLB, so it is valid to fear that increases in inequality have dragged on growth and will continue to drag on growth.

How much is inequality sapping demand?

The previous section noted that rising inequality will, all else equal, slow demand growth, as income is transferred to higher-savings households at the top of the income distribution. The empirical question is just how much rising inequality has contributed to the decline in aggregate demand growth. Again, this decline in demand growth will show up in data as either a slowdown in overall growth, or a pronounced decline in the neutral interest rate. Rachel and Smith (2015), using a variety of techniques, estimate that the rise in inequality in recent decades will likely depress global interest rates by up to 0.6 percent in coming decades, making its importance on par with any other single influence they survey. They also note that while the global interest rate is important, country-specific rates can diverge from global rates due to country-specific factors. Given that the U.S. economy has seen a larger concentration of income at the top of the distribution than have other advanced countries, the effect of inequality on the pace of aggregate demand growth (and hence interest rates) is likely larger as well.

Cynamon and Fazzari (2015) assemble a range of indicators estimating the potential of both rising inequality and rising household debt to act as a brake on demand growth. They highlight the role of rising household debt as a buffer that kept consumption for the bottom 95 percent of households from falling in the face of relatively slow income growth in the pre-Great Recession period. This buffer was annihilated by the wealth lost when the housing bubble burst and ushered in the Great Recession in 2008. Cynamon and Fazzari (2015) argue that the decaying effect on demand of the transfer to the top, combined with the removal of the debt-consumption buffer, can largely explain the slow recovery from the Great Recession. These findings are largely buttressed by Alichi, Kantenga, and Sole (2016), who use a regression-based framework to estimate that changes in the American income distribution since 1998 had led to a demand-drag of more than 1 percent of total U.S. GDP by 2013.

A key component of an analysis of demand drag caused by rising inequality is the savings rate of rich households. Figure E is constructed in the spirit of Cynamon and Fazzari (2015), but makes some slightly different choices about income definitions and presentation of percentiles. It shows average savings rates, from 1989 to 2013, of the bottom four-fifths of the income distribution, as well as savings rates of households between the 80th and 90th percentiles, between the 90th and 95th percentiles, between the 95th and 99th percentiles, and of households in the top 1 percent. It replicates a procedure first identified by Maki and Palumbo (2001) to identify savings rates of the richest households. This procedure, and our adaption of it, uses the Survey of Consumer Finances (SCF) to calculate each income percentile’s share of various asset holdings. It then merges this SCF data on distribution with macroeconomic data from the Financial Accounts of the United States (FAUS), which shows the net acquisition of various types of assets. The assumption used is that if a given income group held, say, half of all Treasury bonds in a given year, then this group was also responsible for half of the aggregate net acquisition of those bonds in a given year. Multiplying each income group’s share of assets by the net acquisition of those assets and summing across all asset types provides an estimate of total assets acquired by each income group in a given year. This gives us a measure of total savings by income group in each year.10

Figure E

For each group’s income, we again use data on comprehensive household income from the Congressional Budget Office (2016). Unlike Cynamon and Fazzari (2015), we include government transfers in our estimate of household income. This income provides purchasing power to households, and this purchasing power absolutely informs households’ consumption decisions and savings rates. Excluding these transfers from household income gives an estimate of how market-based income trends by themselves would have affected aggregate demand growth. Including the transfers provides an estimate of how much shifting inequality overall (even including the generally equalizing effect of taxes and transfers) actually slowed aggregate demand growth in recent decades.

The point of Figure E is simply that savings rates vary enormously across the income distribution. The economy-wide savings rate averaged 11.6 percent from 1989 to 2013, while the savings rate for the top 1 percent averaged 47.4 percent. This large difference in savings rates gives rising inequality a very long lever with which to influence trends in aggregate demand growth. A straightforward back-of-the-envelope estimate of how much the redistribution of income toward the top stifled consumption growth involves multiplying each group’s share of income in earlier years by its savings rate and sum across groups, repeating this procedure for later years, and then simply subtracting the later aggregates from the earlier aggregates. Table 2 shows the results of this procedure.

Table 2

The top panel of Table 2 shows the average savings rate for each of the income groups featured in Figure E.11 In the middle panel of the table are estimates of the change in income share of each income group for three overlapping periods: 1979–2007, 1989–2007, and 1979–2012. Between 1989 and 2007, the share of income claimed by the 95th to 99th percentile of households and the top 1 percent of households rose by 0.6 percent and 6.5 percent, respectively. Multiplying each group’s change in income share over this period by their average savings rate implies that aggregate consumption spending fell by 3.1 percentage points from 1989 to 2007 due to the redistribution of income upward, holding all other influences constant. This figure is found in the bottom panel, which shows the demand drag potentially caused by rising inequality for the three periods. While data restrictions keep us from getting average savings rates going back to 1979 (as the SCF does not have reliable data before 1989), there was a large redistribution between 1979 and 1989 as well. The implied demand drag of the 1979–2007 and 1979–2012 periods assumes that average savings rates across income percentiles from 1989 to 2013 characterize these periods as well.

By 2007, the implied inequality-induced drag on aggregate demand that began in 1979 amounted to more than 4 percentage points of GDP every year. Even if we measure from 1989, and we take as given the large (but almost surely temporary) decline in top 1 percent income shares from 2007 to 2012, by 2012 inequality was imposing a drag of over 2 percentage points on aggregate demand growth. It is worth restating that this hit to the level of aggregate demand generated by rising inequality is cumulative: this demand drag is occurring each year by 2007 or 2012. This means that other macroeconomic influences must continuously ratchet up to keep demand growth from flatlining. We of course know one of these macroeconomic demand resuscitators—the sharp fall in the neutral interest rate.

It would take an ARRA every year to compensate for inequality’s drag on the economy

To get a sense of just how large this inequality-induced drag on aggregate demand had become before the Great Recession hit, it’s useful to compare it with well-known policy interventions. Perhaps the best-known policy effort to boost aggregate demand growth in recent decades has been the American Recovery and Reinvestment Act (ARRA) of 2009. Passed to help stem the downward spiral of the Great Recession and financial crisis, ARRA provided the largest discretionary fiscal stimulus ever provided to the American economy. Its year of peak effectiveness was 2010, when, the Congressional Budget Office (CBO 2015) estimated, it boosted aggregate demand growth (and hence GDP growth) by between 0.7 and 4.1 percentage points of GDP. This upper bound is roughly in line with the estimates above of the demand drag placed on the U.S. economy by rising inequality by 2007. This means that to fully offset the demand drag stemming from inequality using available policy tools, policymakers would need to pass the equivalent of a new ARRA each year. Of course, in the years between 1979 and 2007, other influences boosted demand as inequality sapped it. The most obvious influences were asset market bubbles in the stock and housing markets. But absent transitory (and damaging) influences like large asset market bubbles, the scale of policy intervention needed to keep aggregate demand growth constant in the face of rising inequality is absolutely huge.

Falling NIPA personal savings rate does not invalidate the link between inequality and slow demand growth

The assertion that a large upward redistribution of income over the past generation has slowed growth in aggregate demand implies an increase in economy-wide savings (because more of overall income is going to households that save more). That is why some skeptics could point to the most commonly referenced measure of economy-wide savings—the personal savings rate estimated each month by the Bureau of Economic Analysis (BEA) in the National Income and Product Accounts (NIPA)—to question our conclusion. Over the past generation, this rate has fallen sharply: from 9.8 percent in 1979 to just 3.0 percent by 2007. (This rate spiked upward during and after the Great Recession, as households responded to huge wealth losses by cutting back on spending to rebuild their net worth.)

This decline in the NIPA savings rate needs to be wrestled with. But both theory and evidence indicate that a falling NIPA personal savings rate can be reconciled with the story of inequality tamping down demand growth. Of all the evidence, the strongest support comes from data that suggest that the NIPA personal savings rate is falling at least in part because it does not include a huge source of savings for the wealthy—unrealized capital gains.

In regards to theory, interest rates are set by the interplay of desired savings and investment in the economy, not actual savings. It is theoretically possible that because the redistribution of income to the top of the income distribution slowed demand growth by increasing the American households’ sector desired savings rate, this slow demand growth constricted overall economic growth and thereby slowed the growth of actual savings (since savings are a function of income growth).12

In regards to data, while a cross-sectional redistribution of income from the bottom 95 percent to the top 5 percent will unambiguously reduce savings all else being equal (holding everything else constant), in the real world it is not likely that everything else holds constant. For example, as income was shifting to the top 1 percent from 1989 to 2012, there were often periods when these households were reducing their own savings rates over time. (As Cynamon and Fazzari 2015 document and the appendix of this paper shows, savings rates of top income households are quite volatile.) If rich households perceived the shift in income toward themselves over this period as permanent, a reduction in savings would have indeed been the textbook economic response. This time-series behavior of rich households with regard to their savings rates does not change the fact that a shift of income growth between income classes has potentially large effects on demand growth. Importantly, even when looking at the lowest savings rate recorded in a single year by top 1 percent households in the Survey of Consumer Finances data between 1989 and 2016, the top 1 percent savings rate is still roughly 10 times higher than the savings rate of the bottom 80 percent.

Finally, and most importantly, it is likely that much of the rise in savings from our decades-long upward redistribution of income has actually materialized as unrealized capital gains, which are not captured by the NIPA personal savings rates. Essentially, there are two ways that households accrue net wealth: they consume less than the full amount of income they earn and save the remainder, or, their stock of accumulated past savings gains in value. This gain in value of the stock of accumulated savings is a capital gain. If households sell the asset then they have “realized” this capital gain. Realized gains are captured in some income sources (such as the CBO income data we used in Table 1). But even if a household does not sell the asset as the asset’s price increases, the household has still seen an increase in wealth due to a rise in unrealized capital gains. The NIPA personal savings rate only measures savings out of current income flows—the difference between income and consumption spending in the U.S. household sector. Figure F shows the change in household net worth as a share of GDP. A broader definition of savings sometimes used by economists, this measure includes not only current income flows that are not consumed, but also the changes to wealth occurring from rising or falling asset prices, i.e., unrealized capital gains. This measure shows no obvious downward trend, although it clearly has become more volatile in recent years.

Figure F

As can be inferred from this data showing no downward trend in net worth, even in the face of falling flows of savings out of current income (shown in NIPA data), capital gains have been large (larger than flows of household savings) and growing in recent decades. For example, as can be seen in Figure G, unrealized capital gains on financial assets constituted more than three quarters of the annual rise in household net worth on average between 1979 and 2016. The remaining quarter was contributed by household savings out of current income.

Further, the rise in capital gains is likely driven by a shift in corporate strategy that has redistributed more profits to shareholding households in the form of stock repurchases and less in dividend payments (as shown in Figure H). Dividends and stock repurchases are just two different methods by which corporations can return the benefits of profits to their owners (shareholders). If firms decide to repurchase their own stock, this bids up the firm’s stock price and causes a capital gain. If the firm instead decides to take money that was being used to repurchase stock and use it instead to just pay dividends to shareholders, these dividends would show up in the NIPA measure of personal income and would be captured in the personal savings rate. Capital gains, again, are not measured as personal income and hence the change in corporate strategy to emphasize share repurchases over dividend payments affects measured savings rates.

In short, much savings among high-income households in recent decades has likely shown up more on corporate balance sheets than on household balance sheets. Further, the specific corporate actions—most notably the use of profits to repurchase stocks rather than pay dividends—has kept most measures of national savings from registering the pronounced increase in wealth deriving from the upward distribution of income. This helps explain why some measures of household savings show strong declines in recent decades even as total income claimed by high-income, high-savings households has increased dramatically: the extra savings resulting from that upward redistribution may be showing up in places besides household balance sheets.

Figure G

Figure H

Conclusion

Recent work has highlighted the possibility that rising inequality constitutes an exogenous shock to aggregate demand growth in the American economy. For years, this negative shock could largely be ameliorated by declining interest rates set by the Federal Reserve. But since 2000, the American economy has often found itself with a shortfall of aggregate demand even with short-term interest rates essentially at zero. This means that further increases in inequality will be damaging indeed to prospects of economic growth over the short and medium term unless some other lever of policy fills in the demand shortfall caused by the upward redistribution of income to high-saving households. Further, there is growing evidence that prolonged periods of too-low aggregate demand can damage the economy’s productive capacity.

Policymakers need to get much more serious about avoiding this vicious spiral of chronic demand shortages caused in part by rising inequality degrading productive capacity. Getting serious would mean adopting a more expansionary monetary and fiscal policy portfolio (public investments and expansions to social insurance programs) than has been pursued in recent decades. But, as Taylor et al. (2015) highlight, the scale of upward redistribution of income in recent years would require historically unprecedented changes in taxes and transfers to reverse. They also note that to move the dial on aggregate demand, policy efforts to spur wage increases will have to be much more ambitious than the adjustments to the federal minimum wage in recent decades. We need to enact a much larger raise in the minimum wage and advance policies to boost wage growth for workers making substantially more than the minimum wage.

This makes the EPI’s Raising America’s Pay agenda so vital. It proposes a series of policies that, together, could raise wages for American workers. Pay increases for the bottom 80 percent of households would not just raise the welfare and living standards of these families. Pay increases would also substantially loosen a binding constraint on economic growth: the chronic shortfall in aggregate demand. In short, boosting pay for America’s workers will indeed not only be good for their living standards, it would create a healthier economy overall.

About the author

Josh Bivens joined the Economic Policy Institute in 2002 and is currently the director of research. His primary areas of research include macroeconomics, social insurance, and globalization. He has authored or co-authored three books (including The State of Working America, 12th Edition) while working at EPI, edited another, and has written numerous research papers, including for academic journals. He often appears in media outlets to offer economic commentary and has testified several times before the U.S. Congress. He earned his Ph.D. from The New School for Social Research.

Appendix

The construction of savings rates by income percentile draws heavily on methods pioneered by Maki and Palumbo (2001). Measuring savings rates at the top of the income distribution with the most-used data on consumption, income, and savings—the Consumer Expenditures Survey (CEX)—is difficult because incomes are “top-coded” to preserve confidentiality. This means that reporting on incomes above a certain threshold is suppressed, and instead a single value—the “top-code”—is given to all incomes above the threshold. Further, much recent evidence suggests that the CEX misses a large amount of consumption spending by the rich (see Aguiar and Bils 2015).

Maki and Palumbo (2001) turn to the Survey of Consumer Finances to obtain better measures of savings behavior among high-income households. Besides oversampling wealthy households, the SCF also provides fine-grained income percentile rankings, allowing researchers to identify the top 1 percent of households by income. The measure of saving used by Maki and Palumbo (2001) and this paper is the net acquisition of assets. This is calculated by using the SCF to obtain the share of a particular asset that is held by a given income class. Forty percent of equities, for example, were held by the top 1 percent of income owners in 2013, while 8 percent of residential real estate was held by this top 1 percent. From this acquisition of assets, we subtract the acquisition of household debt using the same methodology.

Then, the SCF shares are applied to data from the Financial Accounts of the United States on the economy-wide net acquisition of those assets by the household sector. So, in 2013, roughly $650 billion in corporate equities was acquired by the U.S. household sector. Assuming that the acquisition was proportional to the cross-sectional share held by each income group, we can measure asset acquisition by income percentile. Finally, we can divide these measures of total asset acquisition (or, savings) by data on incomes from the Congressional Budget Office dataset on effective tax rates and household incomes by percentile. Using the CBO data is one innovation of this paper, as it provides a more comprehensive measure of household income by percentile than what has been used by Maki and Palumbo (2001).

Figure E in the paper reports the average savings rates by income group between 1989 and 2013. While the relative ranking of savings is apparent in each year of the data, there is substantial volatility between individual years in both average savings as well as savings by income group. Much of this volatility seems clearly linked to large movements in asset prices—the bubbles in the stock market and residential real estate markets that characterized the late 1990s and early 2000s, respectively.

Figure A1 shows the estimates for each year of savings by income group. What is apparent is the large swings in savings behavior of the top 1 percent. Between 1989 and 2001, their savings rates declined substantially. From 2001 to 2004, their savings rate increased markedly. However, what is also apparent is that the rich always save orders of magnitude more than the bottom 80 percent.

Figure A1

Endnotes

1. The estimate is 4 percentage points when measuring how much the drag on demand that began in 1979 was slowing growth by 2007, 3.1 percentage points when measuring how much the drag on demand that began in 1989 was slowing growth by 2007, and 2 percentage points when measuring how much the drag on demand that began in 1979 was slowing growth by 2012.

2. In 2016 dollars, the federal minimum wage peaked in 1968 at just under $10, or about 25 percent higher than today’s federal minimum wage of $7.25. A recent proposal would raise the federal minimum to $15 by 2024, which would be roughly equivalent to $12.50 in 2016 dollars, or about 25 percent higher than the 1968 peak (Cooper 2017).

3. See Bivens (2016) for the measure of the “inequality tax.” The 20 percent refers to the peak level of this tax in 2007, and the collapse in top 1 percent incomes following the Great Recession reduced this inequality tax for a number of years. Given that most measures of inequality have begun marching upward post-2011, it seems a safe bet that the 2007 level of the inequality tax is either with us again today or will be soon.

4. This analysis focuses on the impact of inequality on demand through 2007, instead of 2012 or 2013, because top incomes by 2012 and 2013 still had not recovered from the combination of the stock and housing markets declines associated with the Great Recession. Including only the still-depressed top 1 percent shares of 2012 or 2013 in this analysis would underestimate the effect of rising inequality. The latest year we use is 2012 even though income data from 2013 does exist. We chose not to use the 2013 data because of the pronounced income-shifting for tax reasons that occurred in this year that lowered reported capital incomes. Specifically, asset holders shifted capital income realizations to 2012 to avoid some tax changes that were set to become law in 2013. To get a sense of the extent of this tax-shifting, note that the 2012 top 1 percent share of total income was 17.3 percent, a full 2.3 percentage points higher than the 15.0 percent share held by the top 1 percent in 2013.

5. As noted in Mishel et al. 2012, the rise in capital income’s share is driven overwhelmingly by a higher profit rate, not a rise in capital-output ratios.

6. This presumes, of course, that overall income growth over the period would not have been hurt by not allowing inequality to rise. This is a fair presumption, and Bivens 2016 showed that there is no evidence to support worries that a more equal distribution of income growth in the past generation would have somehow impeded average growth rates.

7. See Bivens (2017) and Ball (2014) for some empirical support for the view that prolonged demand-side weakness eventually bleeds over into serious damage to the economy’s supply side.

8. The difference between potential GDP in 2017 as projected by the CBO in 2008 and potential GDP in 2017 as estimated by the CBO in 2017 is roughly $2 trillion. If the damage inflicted by demand shortfalls that led to the Great Recession and subsequent slow recovery is responsible for a quarter of that gap, it would be $500 billion. (See Bivens 2017 and Yagan 2017 for examples of how demand weakness has bled into slower growth in potential GDP.

9. Holston, Laubach, and Williams (2016) provide a recent example of the analysis and labeling of the “falling R-star” phenomenon.

10. From a household perspective, savings (income minus consumption spending) by definition must show up as the acquisition of an asset, even if that asset is just a larger balance in a checking account.

11. We noted previously in footnote 4 why we do not use 2013 as an end-point in our analysis of incomes. The savings rates from the SCF are constructed using averages from 1989 to 2013. We use the 2013 data in this case because the SCF is only available every 3 years.

12. The dynamic where an increase in desired savings actually induces a demand-shortfall that leads to lower realized savings is well-known to macroeconomists; it has traditionally been called the paradox of thrift. In the decades before the Great Recession, it was largely thought to be a theoretical curiosity, but Eggersston and Krugman (2012) have noted that it is quite possible, particularly when the ZLB binds.

References

Aguiar, Mark, and Mark Bils. 2015. “Has Consumption Inequality Mirrored Income Inequality?” American Economic Review vol. 105, no. 9, 2725–2756.

Alichi, Ali, Kory Kantenga, and Juan Sole. 2016. “Income Polarization in the United States.” International Monetary Fund Working Paper WP/16/121.

Ball, Lawrence. 2014. “Long-Term Damage from the Great Recession in OECD Countries.” National Bureau of Economic Research (NBER) Working Paper no. 20185.

Bivens, Josh. 2016. Progressive Redistribution without Guilt: Using Policy Changes to Shift Economic Power and Make Incomes Grow Fairer and Faster. Economic Policy Institute.

Bivens, Josh. 2017. A “High-Pressure” Economy Can Help Boost Productivity and Provide Even More “Room to Run” for the Recovery. Economic Policy Institute.

Bivens, Josh, Elise Gould, Lawrence Mishel, and Heidi Shierholz. 2014. Raising America’s Pay: Why It’s Our Central Economic Policy Challenge. Economic Policy Institute.

Bivens, Josh, and John Irons. 2008. A Feeble Recovery: The Fundamental Economic Weaknesses of the 2001–07 Expansion. Economic Policy Institute.

Board of Governors of the Federal Reserve System, Effective Federal Funds Rate [FEDFUNDS], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/FEDFUNDS, October, 2017

Bureau of Economic Analysis (BEA). 2017. “Table 1.1.6. Real Gross Domestic Product, Chained Dollars.” National Income and Product Accounts. Accessed October, 2017.

Bureau of Economic Analysis (BEA) National Income and Product Accounts. 2017, Table 2.1 Accessed Accessed October 2017.

Congressional Budget Office (CBO). 2008. The Budget and Economic Outlook: Fiscal Years 2008 to 2018.

Congressional Budget Office (CBO). 2015. Estimated Impact of the American Recovery and Reinvestment Act on Employment and Economic Output in 2014.

Congressional Budget Office (CBO). 2016. The Distribution of Household Income and Federal Taxes, 2013.

Congressional Budget Office (CBO). 2017. The Budget and Economic Outlook: 2017 to 2027.

Cooper, David. 2017. Raising the Minimum Wage to $15 by 2024 Would Lift Wages for 41 Million American Workers.Economic Policy Institute.

Cynamon, Barry, and Steven Fazzari. 2015. “Rising Inequality, Demand, and Growth in the U.S. Economy.” Working Paper.

Eggertsson, Gauti, and Paul Krugman. 2012. “Debt, Deleveraging, and the Liquidity Trap: A Fisher-Minsky-Koo Approach.” Quarterly Journal of Economics vol. 127, no. 3, 1469–1513.

Federal Reserve Board. Various years. Financial Accounts of the United States. Tables B.103, R. 103, F.101, and L.6.

Federal Reserve Board. Various years. Survey of Consumer Finances. Microdata.

Holston, Kathryn, Thomas Laubach, and John Williams. 2016. “Measuring the Natural Rate of Interest: International Trends and Determinants.” Working Paper 2016-11. Federal Reserve Bank of San Francisco.

Krugman, Paul. 2013. “Secular Stagnation, Coalmines, Bubbles and Larry Summers.” New York Times, November 16.

Maki, Dean, and Michael Palumbo. 2001. “Disentangling the Wealth Effect: A Cohort Analysis of Household Savings in the 1990s.”

Mishel, Lawrence, Josh Bivens, Elise Gould, and Heidi Shierholz. 2012. The State of Working America, 12th Edition. An Economic Policy Institute book. Ithaca, N.Y.: Cornell Univ. Press.

Rachel, Lukasz, and Thomas Smith. 2015. “Secular Drivers of the Real Interest Rates.” Bank of England Staff Working Paper no. 571.

Summers, Lawrence. 2016. “The Age of Secular Stagnation.” Foreign Affairs, February.

Taylor, Lance, Armon Rezai, Rishabh Kumar, N.H. Barbosa-Filho, Laura Carvalho. 2015. “Wage Increases, Transfers, and the Socially Determined Income Distribution in the USA.” Working Paper, Institute for New Economic Thinking (INET).

Yagan, Danny. 2017. “Employment Hysteresis from the Great Recession.” National Bureau of Economic Research (NBER) Working Paper no. 23844.