Methodological Rift Over California's $20 Fast-Food Wage: Job Losses vs. Null Employment Effects
The academic literature evaluating the employment impact of California’s landmark $20 fast-food minimum wage (AB 1228, effective April 1, 2024) has split into two highly rigorous but fundamentally opposed camps. The debate centers on whether the policy led to a significant contraction in employment or had virtually no impact on headcounts.
On one side, a prominent study by economists Jeffrey Clemens, Olivia Edwards, and Jonathan Meer published by the National Bureau of Economic Research (NBER Working Paper No. 34033) concludes that the wage increase caused a substantial drop in employment. Analyzing Quarterly Census of Employment and Wages (QCEW) data, the authors find that:
"In unadjusted data from the Quarterly Census of Employment and Wages, we find that employment in California's fast food sector declined by 2.7 percent relative to employment in the fast food sector elsewhere in the United States from September 2023 through September 2024. Adjusting for pre-AB 1228 trends increases this differential decline to 3.2 percent, while netting out the equivalent employment changes in non-minimum-wage-intensive industries further increases the decline. Our median estimate translates into a loss of 18,000 jobs in California's fast food sector relative to the counterfactual." (From NBER Working Paper 34033)
On the other side, researchers Denis Sosinskiy and Michael Reich at UC Berkeley’s Institute for Research on Labor and Employment (IRLE) argue that the policy resulted in a "null employment effect." Using granular cellphone mobility data from Advan (tracking shifts longer than 4 hours), Square payroll data, and QCEW data, they counter that:
"Our preferred specification employs a triple-difference method that compares the deseasonalized fast food industry in California to control states as well as to trends in the full-service restaurant industry... our employment estimate centers around zero and is not statistically significant." (From Sosinskiy & Reich (2026))
Reconciling the Methodological Differences
The core of this disagreement lies in several key methodological choices made by the respective research teams:
- Choice of Reference Date and Anticipation Effects: Clemens, Edwards, and Meer (CEM) chose September 2023—the month AB 1228 was enacted—as their baseline, arguing that employers anticipated the wage floor and began cutting staff immediately. However, Sosinskiy and Reich point out that nearly 60% of CEM’s estimated job losses occur before the policy was actually implemented in April 2024. They argue that this "anticipation effect" is actually a statistical illusion caused by other confounding factors.
- Failure to Control for Demographic and Economic Shocks: California experienced significantly slower population growth and differing GDP growth trajectories compared to the rest of the U.S. between 2022 and 2024. When Sosinskiy and Reich use the employment-to-population ratio (rather than raw employment levels) and control for local GDP growth and non-restaurant employment, the pre-implementation "job loss" vanishes.
- Inappropriate Control Groups in Triple-Difference (DDD): CEM compared fast-food employment to "non-minimum-wage-intensive industries" as a control, which exhibited non-parallel pre-trends. Sosinskiy and Reich argue that full-service restaurants (which share food-away-from-home demand and retail real estate pressures but were not subject to the $20 mandate) provide a more valid control group. When using full-service restaurants as a control, the triple-difference estimate centers around zero.
- Data Noise: CEM relied on monthly QCEW data, which is highly volatile and prone to reporting lags. Sosinskiy and Reich demonstrate that quarterly QCEW data, combined with granular mobility-based shift data, provides a much more stable and reliable measure of employment.
This methodological rift shows how easily rigorous econometric models can be weaponized to support opposite narratives, depending on decisions regarding reference dates, control groups, and demographic controls.