New Seattle Minimum Wage Research Contradicts Previous Research
On April 1, 2015, Seattle began it’s experiment with a $15 minimum wage. The minimum wage increase is being implemented as a series of smaller increases over a number of years until it reaches $15. The Seattle City Council commissioned a group of researchers from the University of Washington to study the impacts of the policy and issue reports.
The first report was released in July 2016, and found a decrease in the likelihood that a low wage person was employed after the minimum wage passed. You can find other more in depth summaries of the results all over the internet.
A second report put out by researchers at Berkeley’s Institute for Labor and Employment Research came to starkly different conclusions.
Employment in food service, however, was not affected, even among the limited-service restaurants, many of them franchisees, for whom the policy was most binding.
How did two sets of researchers come to such different conclusions? What differences in the studies might account for these differing results? It would have been nice if the Berkeley team added a section that discussed the differences — there is only one other paper on this topic, and they do come to very different conclusions.
From what I can tell, there are two big differences between these studies.
1. The biggest difference is almost certainly how they built their “Synthetic Seattle” Control.
In order to determine the impacts of a minimum wage increase, both groups use a “synthetic control” method. They create an artificial version of Seattle using a weighted average of different counties (that didn’t experience a minimum wage change) that match trends with the real Seattle in the pre-treatment period. Then they look at the post treatment trends in the real Seattle and the synthetic Seattle, and can more confidently attribute differences to the minimum wage change.
The UW team built their control city using similar counties in Washington State only, while the Berkeley team uses counties in any state that meets certain requirements for their minimum wage laws — mostly Florida, Colorado, Ohio, Missouri and Washington.
The advantage of using counties in other states is that they are far enough away that it’s more certain that the Seattle minimum wage increase didn’t impact their control cities, and by expanding their list of potential counties they can get a stronger employment trend match between their “Synthetic Seattle” and the real one in the pre-treatment period.
The advantage to using similar counties in Washington is that unobserved macro differences between Seattle and the “Synthetic Seattle” are possibly smaller during the treatment period than between Seattle and a “Synthetic Seattle” built from counties in other far away states. The Berkeley team notes that:
Unobserved factors, such as Seattle’s hot labor market compared to that in Synthetic Seattle (Tu, Lerman and Gates 2017), may have positively affected Seattle’s low-wage employment during this period.
And it seems like this would be less likely with a “Synthetic Seattle” built only from counties in Washington.
The Berkeley team also notes that if there are unobserved differences between their control and exposed cities, when additional data becomes available, they will be able to account for it. It will be interesting to see what comes of this when that data becomes available.
2. The sample they study is different.
The UW team had income and hours data, and so could follow a cohort of low wage workers across industries over time from just before the minimum wage law passed to the end of 2015. They compared the outcomes of that cohort to the outcomes of previous Seattle cohorts over 6 quarters, and to the cohorts of a Control “Synthetic Seattle” which didn’t have a minimum wage increase.
The Berkeley team studied total employment in the restaurant industry in Seattle after the minimum wage law was passed compared to their Control “Synthetic Seattle” that didn’t have a minimum wage increase.
I don’t know if either of these methods is better than the other. I kinda prefer the UW cohort method because it measures changes for individual workers as opposed to aggregate employment trends in an industry that mixes workers that are bound by the minimum wage and workers that are not.
There are other differences, the Berkeley team has an additional quarter of data, they use different data sets (though the UW team compares their data set to the one Berkeley uses and it doesn’t appear too different), but these seemed to me like the biggest differences.
In the end, these are still early days in this minimum wage experiment, and hopefully a clearer picture will emerge as more time goes by. Or, this will become just another episode in the minimum wage debate saga where one group with one set of methods says one thing and another group with another set of methods says another, and everyone just chooses the side that matches their politics.