When California announced in September 2023 that fast-food workers at large chains would soon earn a $20 hourly minimum wage, franchise owners had six months to prepare. Newspaper reports at the time suggested some were already trimming shifts. A new analysis suggests those anecdotes reflected a broader pattern, and that by the time the law took effect on April 1, 2024, on-site staffing at affected restaurants had already dropped by roughly 8 percent.
The study, published in Applied Economics Letters, uses anonymized cellphone location data to track how fast-food restaurants adjusted their workforces in the months between the law’s announcement and its implementation. That anticipatory window is one that most administrative labor data cannot see clearly.
A sectoral wage floor with a sharp line
California’s Assembly Bill 1228 is unusual because it targets a specific slice of the labor market. The $20 hourly wage applies only to limited-service restaurant chains with 60 or more locations nationwide. Fast-food outlets tucked inside grocery stores, hotels, or airports larger than 15,000 square feet are exempt. That means two nearly identical burger counters, one freestanding and one inside a supermarket, could face very different wage rules.
That sharp coverage line is what drew the attention of Hitanshu Pandit of Northeastern University’s School of Public Policy and Urban Affairs. Earlier studies of AB-1228 had produced mixed results, ranging from no adverse employment effects to modest declines to more substantial reductions. But most of those studies drew on payroll or government employment data that arrives quarterly or monthly and rarely identifies which restaurants belong to which chain. That makes it hard to see whether firms adjusted before the law took effect, and hard to assign treatment status based on chain size.
Using cellphone pings as a staffing proxy
Pandit’s approach relies on a commercial dataset from Advan Research that aggregates GPS pings from mobile devices across more than 5 million commercial locations in the United States. For each establishment and each week, the data record how many devices visited and how long they stayed. Visits lasting longer than four hours serve as a stand-in for worker shifts, since most customers do not linger that long at a restaurant.
To check whether that stand-in actually tracks employment, Pandit compared the county-level total of long-duration visits in 2022 against official Census Bureau employment counts for food service and retail. The correlations exceeded 0.97, suggesting that the mobile-device measure does closely track the number of people working at these locations.
The measure has limits, and Pandit acknowledges them. It captures worker-shifts rather than total hours worked. If a restaurant cut headcount but stretched remaining workers’ hours, the measure would register a decline even if total labor stayed the same. Conversely, if a restaurant kept the same number of workers but shortened shifts below the four-hour threshold, the measure would also fall. The data cannot separate these possibilities.
Comparing similar restaurants across sectors
To isolate the effect of the fast-food wage law from broader economic trends, Pandit compared covered fast-food outlets against full-service restaurants and retail establishments in the same California counties. Full-service restaurants draw from similar labor pools and face similar local demand conditions but were not subject to the $20 floor. Retail stores that employ many minimum-wage workers were also excluded from the mandate.
The analysis was restricted to California counties that follow only the state-level minimum wage schedule, excluding places with their own local wage ordinances that might muddy the comparison. The panel ran weekly from January 2023 through December 2024, spanning both the announcement in September 2023 and the implementation in April 2024.
The statistical technique, known as difference-in-differences, compares how the treated group (large chain fast-food outlets) changed relative to the comparison group (full-service restaurants and retail) before and after the policy shift. Establishment-level fixed effects account for permanent differences between individual stores, while week fixed effects control for shocks that hit all sectors at once.
What the data revealed
Across the specifications, treated fast-food outlets showed roughly an 8 percent decline in on-site staffing intensity after the September 2023 announcement, relative to comparable establishments. The estimate held whether the comparison group was full-service restaurants, retail stores, or the full combined sample.
A placebo test using the exempt “enclosed” fast-food outlets, the ones inside grocery stores and similar venues, found no significant effect. Because those restaurants were not covered by the law, they should not have adjusted staffing if AB-1228 was actually driving the observed change. The fact that they didn’t supports the interpretation that the law itself was responsible for the drop at covered outlets.
The event-study version of the analysis, which traces the timing of the effect week by week, showed no meaningful differences between the treated and comparison groups before the September 2023 announcement. After the announcement, the coefficients turned progressively more negative, stabilizing at roughly 7 to 10 percent below baseline. The decline was gradual rather than a sharp break at the April 2024 implementation date, consistent with firms adjusting staffing in advance of a predictable cost increase.
Urban and rural locations showed similar responses. The interaction term testing for urban-rural differences was small and statistically insignificant, suggesting that market density did not shape how restaurants adjusted.
How this fits with earlier research
Pandit calculates that the estimated response implies a short-run employment elasticity of roughly negative 0.3 to negative 0.4, meaning that a 1 percent increase in the wage was linked to a 0.3 to 0.4 percent decrease in on-site staffing intensity. That range sits within the estimates from Clemens, Edwards, and Meer’s National Bureau of Economic Research working paper and meta-analytic figures from Neumark and Shirley.
The author frames the contribution as complementary to studies using administrative data. Those sources can distinguish between headcount changes and hours changes, while the mobile-device data cannot. But the mobile-device data can identify which chain each location belongs to, which unemployment insurance records generally cannot, and can observe weekly rather than quarterly patterns.
One caveat worth noting is that visit attribution depends on GPS pings falling within a defined geographic boundary around each establishment. In dense commercial areas where multiple businesses share a building, that boundary can pick up noise from neighbors. Pandit notes that this concern is most relevant for the enclosed-venue placebo sample, since those outlets sit inside larger structures.
The author interprets the pattern of gradual, anticipatory adjustment as consistent with standard labor demand theory: when firms know a cost increase is coming and know when it will hit, they smooth out their response rather than making abrupt changes on the effective date.
