Credible research designs for minimum wage studies

Read the full PDF in your browser (opens in a new tab)

We assess the NSW (Neumark, Salas and Wascher 2014) critiques of our minimum wage studies that found small effects on teen employment. Data from 1979-­2014 contradict NSW, and show that the disemployment suggested by a model assuming parallel trends across U.S. states mostly reflects pre­-existing trends. A data-­driven LASSO procedure that optimally corrects for state trends produces a small employment elasticity (­-0.01); even a highly sparse model rules out substantial disemployment, contradicting NSW’s claim that we discard too much information. Synthetic controls do place more weight on nearby states—confirming the value of regional controls—and generate an elasticity of ­-0.04. A similar elasticity (-­0.06) obtains from a design comparing contiguous border counties, which we show to be good controls. NSW’s preferred matching estimates mix treatment and control units, obtain poor matches, and find employment declines the most where the relative minimum wage falls. These findings refute NSW’s key claims.