The minimum wage debate is far from over. Recently, two competing studies on the phased-in minimum wage hike in Seattle reached opposing conclusions about the effect on employment, wages and hours worked.
A National Bureau of Economic Research working paper from researchers at the University of Washington finds that the recent minimum wage increases from $9.47 to $11 in 2015 to $13 in 2016 resulted in a 9-percent decline in hours worked in low-wage jobs, while hourly wages rose by 3 percent. On net, low-wage workers lost an average of $125 per month in 2016.
{mosads}In contrast, the study from IRLE Berkeley finds that average wages rose marginally for certain types of restaurants following the minimum wage hike and had insignificant dis-employment effects. How can two studies focused on the same city, tracking the same minimum wage change in law and using the same “synthetic control” methodology arrive at vastly differing results?
Some of the differences are the result of the industries covered. The NBER working paper includes in their analysis all low-wage workers across different industries that earn below a certain hourly wage, such as $13 or $19. The authors worry that if the analysis only focused on workers who are either at or below the minimum wage, it would overstate the dis-employment effects if employers moved some workers to wages above the threshold minimum wage.
Moreover, looking across all sectors allows them to capture workers who might lose their jobs in one industry but find employment elsewhere. Administrative data reported in the paper show that employment at jobs paying less than $13 per hour declined by 34 percent between the last quarter of 2014 and the third quarter of 2016. Over the same period, hours worked for those earning under $13 an hour declined by 33 percent, while average wages for this group rose by 5 percent.
In contrast, the IRLE Berkeley study focuses only on the food services and restaurant industry, since that industry is likely to have the largest fraction of low-wage workers affected by the minimum wage hike. As mentioned, this paper found no negative employment effects for this industry.
Interestingly, the administrative data reported in the NBER paper shows that, while overall employment in the food service and restaurant industry increased between 2014 and 2016, employment declined for workers earning under $13 per hour. In fact, the NBER paper also reaches the conclusion that, when looking only at the restaurant industry, the overall employment change was negligible.
But more importantly, a significant difference arises from the application of the synthetic control method. The synthetic control method requires the construction of an appropriate “control” group, basically a “Synthetic Seattle”. In other words, a control group is a group of geographic units that matches Seattle so closely before the minimum wage increase that any difference in outcomes between Seattle and “Synthetic Seattle” can plausibly arise only from the minimum wage increase.
To construct the control group, the NBER authors look only at other Public Use Microdata Areas (PUMAs) within Washington state. A PUMA is a geographic area within a county with a population of approximately 100,000 people. To see which of these PUMAs is closest to the city of Seattle, they match on pre-minimum wage hike variables such as average wages, number of jobs, quarterly hours worked and payroll.
The IRLE Berkeley paper uses as a potential “Synthetic Seattle” all counties from across the country where minimum wages were indexed to inflation. They appear to only use population to match Seattle with these other counties.
Neither of these approaches is ideal. Typically, constructing the pool of potential matches involves looking at more than just one variable and also more than just the outcomes of interest. For instance, in my co-authored paper using the synthetic control approach to identify the effect of “Right to Work” laws on income inequality, we used more than 10 variables ranging from per capita incomes to state tax rates to the number of people receiving public assistance to match and construct the synthetic state.
Both papers seem to fall short on this score. While the Berkeley paper only appears to match on population, the NBER paper only appears to include the outcome variables, such as wages, jobs, payroll and hours worked, as the matching variables.
Also, it is problematic to limit the pool of potential matches to only Washington state as is done in the NBER study, or to states with minimum wages indexed to inflation as is done in the Berkeley paper. The ideal approach is for the methodology itself to dictate which geographic units are an ideal match ex-ante, and to assign all others a zero weight. These could include PUMAs or counties outside Washington state, or in states where wages are not indexed to inflation.
As I have written earlier, I expect that nearly doubling the minimum wage will have significant dis-employment effects. Some of this appears in the data from the NBER paper on changes in employment for low-wage workers. But in the absence of a convincing methodology, it is hard to ascribe all of this decline in employment to a hike in the minimum wage — as the NBER paper suggests — or to conclude that minimum wages have no negative employment effects — as the IRLE Berkeley paper does.
I will end as I began. The minimum wage debate is far from over.
Aparna Mathur (@aparnamath) is a resident scholar in economic policy studies at the American Enterprise Institute.
The views expressed by contributors are their own and not the views of The Hill.