A common but erroneous theme in the media about recent labor market trends is that technology (the robots!) threatens job growth and is the cause of wage stagnation and inequality. Politicians, policymakers, and pundits echo this as well. These insights come from research on the “job polarization hypotheses”—the claim that computerization leads to the “simultaneous growth of high-education, high-wage and low-education, low-wages jobs at the expense of middle-wage, middle education jobs” and, correspondingly, to wage polarization. It is noteworthy, therefore, that MIT Professor David Autor, the leading intellectual architect of the job polarization hypothesis, has presented a paper at the Federal Reserve Bank of Kansas City’s economic policy symposium in Jackson Hole, Wyo., which finds that job polarization did not occur in the 2000s and that, in any case, job polarization is not necessarily connected to wage polarization.

This confirms the findings of others, such as Beaudry, Green, and Sand and my own research with Heidi Shierholz and John Schmitt. One can only applaud Autor for updating his analysis of employment and wage trends, and acknowledging the lack of occupational job polarization in the 2000s and its failure to be able to explain wage trends. One can only hope that the economics commentariat will follow suit and ramp up their exploration of other causes of stagnant and unequal wage growth. Immodestly, I would suggest our recent paper, Raising America’s Pay, as a starting point.

In 2010, Autor wrote the influential paper The Polarization of Job Opportunities in the U.S. Labor Market: Implications for Employment and Earnings for the Center for American Progress and the Hamilton Project. This paper laid out what became the conventional wisdom: “the structure of job opportunities in the United States has sharply polarized over the past two decades, with expanding job opportunities in both high-skill, high-wage occupations and low-skill, low wage occupations, coupled with contracting opportunities in middle-wage, middle-skill white-collar and blue-collar jobs” and “this pattern of employment polarization has a counterpart in wage growth.”

In his latest paper delivered today, Autor takes on a large task, offering “a conceptual and empirical overview of the evolving relationship between computer capability and human skill demands.” In so doing, he updates his analysis and examines employment and wage trends for the periods 1979-89, 1989-99, 1999-2007, and 2007-12, and is thus able to examine the 2000’s both before and during the Great Recession.

His analysis (see Figure 7 below) examines occupational employment growth, with occupations ranked by wage on the horizontal axis and the changes in each occupation’s employment share on the vertical axis. It shows (orange and green lines) that employment in high-wage occupations did not expand in the 2000s. Instead, Autor writes:

Growth of high-skill, high-wage occupations (those associated with abstract work) decelerated markedly in the 2000s, with no relative growth in the top two deciles of the occupational skill distribution during 1999 through 2007, and only a modest recovery between 2007 and 2012. Stated plainly, the U‑shaped growth of occupational employment came increasingly to resemble a downward ramp in the 2000s.

That is, instead of a shrinking middle with an expansion at the top and bottom of the occupation hierarchy for 1999-2007 (see the orange line), he finds an expansion of low-wage occupations (in the bottom thirty percent of occupational employment, ranked by wages), a shrinkage in the middle (from the 30^th to the 80^th percentile of occupational employment, ranked by wages) and no expansion in the upper twenty percent. Only low wage occupation expanded. The 2007-2012 period (the green line) had similar trends, though there was a slight expansion of high wage occupations. Simply put, there has been no occupational employment polarization since 1999.

Autor points out that the consequence of this trend is an oversupply of “highly educated workers,” who have taken jobs that do not require such an education: “These patterns suggest that the set of abstract task-intensive jobs is not growing as rapidly as the potential supply of highly educated workers. As Beaudry, Green, and Sand (2013, 2014) highlight, the coalescence of these forces has likely led highly educated workers to seek less educated jobs, which in turn creates still greater challenges for the lower educated workers competing for routine and manual task-intensive work.”

Further, Autor’s analysis of wage trends (see Figure 6 below) shows there is no direct correspondence between these employment patterns and wage patterns. He finds that wages grew more slowly in the entire 2000s (green and orange lines) than in the 1990s (blue line) or 1980s (red line), and that the wages grew more the higher the wage level: the highest-wage earners fared better than middle-wage earners who fared better than low-wage earners, with wages absolutely falling among the lowest-paid ten percent. For low-wage workers this is the exact opposite of what one would expect based on the “polarization hypothesis,” since their more rapid employment expansion should have been associated with faster, not slower, growth in wages. Rather, Autor finds that, “occupational employment trends do not drive wage patterns or wage inequality” and “technological and skill deficiency explanations of wage inequality have failed to explain key wage patterns over the last three decades, including the 2000s.” This is consistent with some of the major findings of my research with Heidi Shierholz and John Schmitt.

In the new paper, Autor directly acknowledges the failure of occupational employment patterns to predict wage patterns, something he did not do in his earlier work, writing that “while computerization has strongly contributed to employment polarization, we would not generally expect these employment changes to culminate in wage polarization except in tight labor markets.”

Many observers have erroneously claimed that occupational job polarization explained middle class wage stagnation. Indeed, many observers have treated occupational employment trends as the secret DNA for understanding labor market trends. Hopefully, this broad agreement that occupational employment trends have no strong counterpart in wage trends will encourage these observers to look elsewhere for explanations for wage trends.

Autor makes advances on his earlier work by trying to explain the divergence between occupational employment and wage patterns. The problem to solve is why changes in occupational employment patterns (the rising and falling shares of total employment of particular occupations) bear no relation to the relative wage changes of (particular) occupations (as we found), and therefore do not contribute much to our understanding of changes in wage patterns. Autor tries to solve the puzzle by referencing the impact of three mitigating forces: complementarity, demand elasticity, and labor supply. I think he confuses matters a bit, since two of his three mitigating factors are already reflected in the changes in occupational employment patterns. Specifically, complementarity (expanding the role of computers to not just substitute for some but not other occupations, but also to raise demand for some but not others as “complements”) and demand elasticity (whether efficiencies in products and services will lead to more or fewer expenditures on them) are captured by changes in employment.

Changes in supply certainly can be a mitigating factor, so the explanation becomes that low-wage occupations do not see their wages rise even when they expand more than other occupations because most everybody is qualified to do those jobs. Okay. But changes in supply (how many people are available to do that occupation’s work) do not answer why the majority of wage inequality occurs within detailed occupations, as we find, especially in recent years. Moreover, on first glance it does not seem that supply shifts (middle-wage workers bumping down into low-wage occupations) readily explain why low wage occupations’ wage growth differed relative to that of middle-wage occupations across the various time periods: comparable in the 1980s, much better in the 1990s, and much worse in the 2000s. In my view, even augmenting the polarization hypothesis with these three mitigating factors, looking at employment shifts across occupations (including the supply of workers to them) provides very little insight into wage trends.

In the end, changes in occupational employment patterns are a dim flashlight indeed for shining light on key wage patterns. In fact, we found that occupations have become a less, not more, important determinants of wage pattern in recent years. Both academics and the media ought to cast a wider net for understanding trends. Given the weakness of the polarization hypothesis, there is no plausible story for connecting technological change, robots, or computers, to wage stagnation and growing wage inequality. This would suggest that policymakers and the punditry should widen their scope and consider other factors. Heidi Shierholz, John Schmitt, and I even have a new paper that is relevant: Wage Inequality: A Story of Policy Choices.