Our analysis of January 1 state minimum wage changes understated the total increase in wages for workers throughout the country
In December, we published a “snapshot” estimating that 4.5 million workers throughout the country were likely to receive a raise at the beginning of the year as a result of higher state minimum wages going into effect. We estimated that these increases would raise the annual income of affected workers by roughly $5 billion. Subsequently, Mark Perry at the American Enterprise Institute published a blog post critiquing those estimates. He claimed our research methods were flawed and opaque, and that the wage increases for workers impacted by state minimum wage increases in 2018 will be much smaller than our original $5 billion. All of these claims are wrong.
In regards to our research methods, we do make one modeling decision that may strike some as overly optimistic: we assume that new minimum wage levels will largely be enforced. But in regards to the size of minimum-wage-driven raises in 2018 our methods contain one hugely conservative choice. We didn’t fully account for minimum wage changes in New York, and subsequently left out significant wage increases going to workers in New York City and its surrounding counties. We’ll say some more on both of these issues below.
It’s worth noting first that while EPI does not typically publish methodological statements for snapshots—they’re short pieces with a single, informative graphic with minimal accompanying text—the methodology employed in our December snapshot is the same methodology we have used for years in modeling the impact of higher state and federal minimum wages. We’ve published this methodology multiple times, the most recent being this past April in appendix B of our analysis of the proposal to raise the federal minimum wage to $15 by 2024.
While there is always room for improvement, we think this methodology is pretty good. In fact, when the Congressional Budget Office estimated the number of workers likely to be directly affected by a federal minimum wage increase to $10.10 in 2014, their estimate was nearly identical to ours.
While we’ve published our methods before, more transparency is generally better, so the next several paragraphs provide a quick overview of our minimum wage analysis methodology.
EPI uses statistical software to analyze publicly available microdata from the outgoing rotation group (ORG) of the Current Population Survey (CPS), published by the Bureau of Labor Statistics. The CPS-ORG has data on workers’ reported usual weekly and hourly earnings, as well as their usual weekly hours. Using the CPS data, we identify workers in the sample likely to be directly affected by a minimum wage increase as those with hourly wages between the existing minimum wage and the enacted or proposed new one.
While conceptually straightforward, there are still challenges in accurately identifying these likely-affected workers because of measurement error in the data. For example, survey respondents often “clump” their reported hourly wage values around round numbers—e.g., if someone is paid $8.30 per hour, they’ll simply report their hourly wage as $8.00 per hour. (You can read more about this phenomenon, and how we deal with it, in the methodological statement for EPI’s State of Working America Data Library.) Also, some workers may only report their pay in weekly terms, and if they misreport the hours they typically work, it will lead to an erroneous implied hourly wage. For minimum wage analyses, these measurement issue are particularly acute since low-wage workers are prone to conditions that exacerbate measurement error: turnover is high in low-wage industries so workers are frequently changing jobs, work is more likely to be seasonal or part time, and many low-wage jobs suffer from inconsistent hours. (See Methodological Appendix C of this report for additional discussion of this issue.)
To account for this measurement error, we typically include anyone who reported earning at least 90 percent of the existing minimum wage in our sample of directly-affected workers. We exclude every observation with a wage value below this 90 percent cutoff point on the assumption that whatever is causing them to report being paid significantly less than the minimum wage—such as being exempt from the minimum wage or being the victim of wage theft—is likely to preclude them from benefiting from the minimum wage hike.
Having identified the sample of likely-affected workers, we calculate their expected raise by simply subtracting their existing hourly wage from the new (or proposed) minimum wage. Because we’re using microdata, we don’t have to make any assumptions about these workers’ work hours—we know the hours they typically work from the data. Thus, we can calculate the increase in aggregate pay for affected workers by multiplying the implied raise for each worker by their reported usual weekly work hours times 52 weeks per year.1
In last few years, we have noticed—at least in a number of states that have raised the minimum wages—that the number of observations reporting wages below the legal minimum wage has grown, and in some cases the gap between their reported hourly wage and the statutory minimum wage has also grown. (We noticed this first in Massachusetts, where the minimum wage rose incrementally from $8.00 in 2014 to $11.00 in 2017.) Having not yet done a thorough analysis, we speculate that three possible factors might be driving this phenomenon: 1) growing nonresponse in the CPS could be increasing overall measurement error; 2) there could be some lag in employer compliance with higher minimum wages; and 3) there could be some lag in some employees’ awareness of the change in their hourly pay.
Because of this trend, and to ensure we are still capturing the full universe of likely-affected workers, in recent analyses we have lowered the cutoff point for the sample of affected workers to those who report earning an hourly wage of at least 80 percent of the existing minimum wage. Doing so has improved our estimation of the number of affected workforce when analyzing multi-year minimum wage increases, making estimates in a particular state more consistent year-after-year when updating estimates with newer years of CPS microdata.
However, lowering this cutoff point has also complicated how we calculate the expected wage increase for affected workers. Including observations with hourly wage values significantly lower than the existing minimum wage creates instances where the implied raises for these observations may be too large. For example, a worker in Massachusetts who—for whatever reason—reports being paid $8.75 per hour in 2016 (when the minimum wage was $10, after it was raised from $8 in the preceding two years) probably got a pay hike when the minimum wage rose from $10 to $11. Mechanically, however, our standard model would estimate their raise as $2.25 pay hike, despite only a $1 increase in the minimum wage. This is almost certainly an erroneous result (although it’s possible that some such estimates are accurate if, for example, enactment and execution of a minimum wage hike were accompanied by greater enforcement and compliance). By specifying our raise model to increase these workers’ wages all the way to the new minimum wage, we’re effectively assuming full compliance with the law. This may be an overly optimistic assumption given estimates of noncompliance in low-wage industries, but we think it is reasonable to make this assumption if we want to report the full potential wage increase for workers and their employers.
When we model typical legislative or ballot-measure minimum wage increases (e.g., increases of $0.50 or $1.00), the affected wage band is large enough to pick up significantly more “normal” wage values (i.e., at or above the existing minimum wage) than subminimum values, thus rendering this subset of potentially artificially large raises inconsequential in the overall analysis. However, in instances where the change in the minimum wage is relatively small—such as Alaska’s $0.04 inflation adjustment—the paucity of workers in the affected band (and the corresponding small CPS sample size) means that even a handful of these artificially large raises is quite likely to overstate the calculated aggregate value of the increase.
To assess the potential impact of this issue, we went back and reran all our estimates for the states with relatively small inflation-indexed minimum wage increases, this time assuming that all directly-affected workers received an hourly wage increase equal to exactly the increase in the minimum wage—that is, we did not assume that workers earning below the existing minimum wage would be bumped all the way up to the new minimum. As shown in Table 1, changing the model in this way reduced the total aggregate increase in wages for all affected workers in states that raised their minimum wages from $5 billion to $4.7 billion.
States with minimum wage increases effective January 1, 2018 (revised)
|State||Share of workforce directly benefiting||Type of increase||New minimum wage as of Jan. 1, 2018||Amount of increase||Total workers directly benefiting||Total increase in annual wages|
|New Jersey||2.3%||Inflation adjustment||$8.60||$0.16||91,000||$21,519,000|
|South Dakota||2.7%||Inflation adjustment||$8.85||$0.20||10,000||$2,391,000|
Note: "Legislation" indicates that the new rate was established by the legislature or through a ballot measure. "Inflation adjustment" indicates that the new rate was established by a formula, reflecting the change in prices over the preceding year. Directly affected workers will see their wages rise because the new minimum wage rate exceeds their current hourly pay. This does not include additional workers who may receive a wage increase through "spillover" effects, as employers adjust overall pay scales. *The New York estimates only reflect the change in the minimum wage in upstate New York. New York City and its surrounding counties had separate minimum wage increases that are not captured in this estimate. New York's minimum wage increase took effect on December 31, 2017. Totals may not sum due to rounding.
Source: EPI analysis of Current Population Survey microdata 2016
As we were reviewing these estimates, we noticed one other problem in our calculations. When we estimated the impact of New York’s minimum wage increase, we took a conservative approach to modeling the impact of the increase there. We only modeled the minimum wage increase that applies to upstate counties, where the minimum wage rose from $9.70 to $10.40. But, New York’s minimum wage law specifies three different minimum wages: one for New York City, one for the downstate counties immediately surrounding New York City, and another for the upstate counties. On Long Island and in Westchester County, the minimum wage rose from $10.00 to $11.00, and in New York City, the minimum wage went from $11.00 to $13.00. By not including for the increase in the New York City and downstate counties minimum wages, we significantly underestimated the total wage increase for workers in New York.
Unfortunately, the CPS is not well-suited for analyzing minimum wage increases in sub-state geographies, so estimating the precise impact of the New York City increase and the surrounding counties would take significantly more time and would require use of a different model and data source. Fortunately, prior to the enactment of New York’s minimum wage law, we did produce separate estimates of the likely impact of a minimum wage proposal that was similar—although not exactly the same—as the minimum wage bill that was ultimately passed. In the proposal we analyzed, the New York City minimum wage would have been raised to $12.00 in 2017, and $13.50 in 2018. (The enacted law raised the city minimum to $11.00 in 2017 and $12.50 in 2018.) Still, our estimates from this step in the proposed New York City minimum wage provide some sense of the additional wages going to workers in New York that we missed in our original snapshot.
We estimated that raising the minimum wage in New York City from $12 to $13.50 would have raised all affected workers’ earnings by roughly $2.1 billion (see Appendix Table A1). Now, it is important to note that this includes wage increases for workers “indirectly affected” by the minimum wage hike—i.e., workers already being paid slightly more than the new minimum wage, who are likely to get some raise as employers adjust their overall pay scales. (For simplicity’s sake, we did not include any indirect effects in our estimates for the December snapshot—another reason to believe that, if anything, our December estimates were too low, not too high. Had we included these “indirect” impacts, the count of affected workers and the total wage increase for affected workers would be even larger.) The wage increase for just directly-affected workers in New York City would be smaller than $2.1 billion—although wage increases for directly-affected workers would still be the bulk of the overall change in the wage bill.
Based on these figures, it’s reasonable to estimate that the minimum wage hike in New York City alone at the start of this year likely raised workers’ annual wages somewhere in the ballpark of $1.5–$2 billion. Consequently, our estimate of the total aggregate annual wage increase for workers getting a raise on January 1, 2018 should have been upwards of $6 billion.
To sum up, our research methods for estimating the effect of minimum wages will likely never be perfect, but they’re transparent and match closely with what other experts’ calculations show in most instances. More importantly, low-wage workers will indeed get non-trivial boosts to their paychecks from minimum wage increases across a number of states in 2018—even more than we initially reported.
1. You may be thinking “but not all affected workers work year-round.” This is true, and it’s why we don’t make any claims about the impact on individual workers’ earnings—and when we do so in other reports, we try to be careful to specify that effects on individual workers’ earnings (such as the average increase in annual wages) are specifically for those workers working year-round. Nevertheless, when describing the aggregate change in wages for all affected workers, we multiply all observations’ implied weekly raise by 52 because we’re pooling a full calendar year of CPS data and if a low-wage worker only worked a portion of the year, they would necessarily be excluded from the sample in the months in which they did not work.