Did Proposition 47’s effect on theft translate into harm to California’s retail sector?
California’s 2014 reform reclassified certain nonviolent offenses from felonies to misdemeanors. This thesis looks beyond its effects on crime to ask what happened to retail businesses, payroll, and employment.
The estimates point to higher reported larceny, but no clear effect on retail business density. Retail payroll and employment appear to have increased modestly. These statewide results do not rule out losses for individual stores or communities.
Larceny per capita
≈ +22%
p < 0.001
Retail businesses per capita
no clear change
p = 0.19, not significant
Retail payroll per capita
≈ +4.6%
p < 0.001
Retail employment per capita
≈ +2.7%
p < 0.001
Difference-in-differences estimates in log points, California vs. 15 comparison states, 2008–2019. Synthetic control and synthetic DiD agree on larceny; on employment the three methods disagree (see Finding 3).
Background
Passed by voters on November 4, 2014, Proposition 47 (the Safe Neighborhoods and Schools Act) reclassified low-level drug possession and petty theft from felonies to misdemeanors. Theft of property worth up to $950 — including shoplifting — became a misdemeanor; the old felony threshold was $400.
It followed Brown v. Plata, the Supreme Court ruling that ordered California to relieve prison overcrowding. Savings from lower incarceration went to a fund for drug and mental-health treatment, victim services, and dropout prevention.
Years later, Prop 47 became a shorthand for retail theft: news coverage of San Francisco store closures blamed it for a wave of shoplifting. This thesis tests whether the retail sector was actually harmed.
Findings
Finding 1
Reported larceny increased
Difference-in-differences, synthetic control, and synthetic difference-in-differences all point to an increase in larceny per capita relative to comparison states after the reform. This is a relative increase; it does not mean larceny rose in every year.
Larceny rose in California after Prop 47
Reported larcenies per 1,000 residents. California moved with the average of the 15 comparison states until 2014; afterwards theft kept falling elsewhere but not in California. The shaded band spans the comparison states.
The larceny effect is hard to explain away
HonestDiD 95% confidence intervals for the larceny effect as the allowed violation of parallel trends after the reform (M̄) grows. The interval stays above zero through M̄ = 1.5 and first includes zero at M̄ = 2.
Finding 2
No clear change in business density
The difference-in-differences estimate for retail businesses per capita is not statistically significant, and the synthetic-control methods give mixed results. The evidence does not establish a meaningful effect on business density during the study period.
Retail business density was unaffected
Retail establishments per capita, indexed to 2014 = 100 to remove level differences between states. California stays inside the range of the comparison states; no break opens after the reform.
Finding 3
Modest increases in payroll and employment
The labor-market estimates are generally positive, especially per capita. Evidence is less consistent for some per-establishment measures, and noisy employment pre-trends and weaker payroll placebo results call for caution.
The methods disagree on retail employment
Estimated effect on retail employment per capita (log points) with 95% intervals. Only difference-in-differences finds a significant effect; synthetic DiD is near zero, and the synthetic-control estimate rests on a fit that collapses onto a single comparison state, so it can’t be interpreted.
Approach
Study period
2008–2019: seven pre-treatment years and five post-treatment years. The main analysis excludes the pandemic period.
Comparison
California and 15 comparison states that did not change their felony theft thresholds during the study period.
Data
Census business and population data, Bureau of Economic Analysis retail GDP, and FBI reported larceny data. Outcomes are measured per capita or per establishment.
Methods
Difference-in-differences, synthetic control, and synthetic difference-in-differences, with the last used as the preferred specification. Event studies and placebo tests help assess the results.
What explains the labor-market results?
A mediation analysis asks whether increased larceny explains the labor-market effects. It finds little consistent evidence for that pathway. The one statistically significant indirect effect—lagged larceny on employment per capita—is sensitive to unobserved confounding. Security hiring and other explanations remain hypotheses, not established mechanisms.
Limits
The analysis ends in 2019. It does not explain pandemic-era or more recent retail outcomes, and five post-treatment years may miss slower changes in business entry and exit.
Statewide averages combine urban and rural areas and different kinds of retailers. They can conceal concentrated effects on particular firms or places.
The study measures reported larceny, business density, payroll, and employment—not store profitability or shoplifting losses directly. Causal interpretations depend on the comparison groups and assumptions of each method.