The Minimum Wage, EITC, and Criminal Recidivism

Notes: “Recidivate 1 Year” indicates those released prisoners who returned to prison within one year of their release (analogous for “Recidivate 3 Year”). Column 2 has fewer observations to allow everyone to have three years of post-release data, whereas Column 1 only requires one year of post-release data.

Notes: Each observation is a state–year–month, so there are 9,180 observations (51 states including DC × 15 years × 12 months). A change in the state minimum wage could come from a state-level law or a federal minimum wage change. Minimum wages are measured in real 2011 dollars. Note: 17 states had no changes other than the federal minimum wage increases. State EITC percent is the percent of the federal EITC that the state EITC represents.

Notes: The dependent variable is return to prison in the same state within one or three years of release (indicated in the column heading). Minimum wage is measured in nominal dollars in the state and month the offender was released. All specifications include state and year fixed effects, as well as the individual and time-varying state-level controls outlined in Section III. State EITC policy is included as a dummy variable, but its coefficient is not shown. See Table 6 for those result stratified by gender. Mean recidivism rates are the mean of the dependent variable for the respective column. Robust standard errors clustered at the state level are shown in parentheses (43 clusters). p-values from 1,000 wild-cluster bootstrap iterations are shown for the main minimum wage coefficient as suggested by Cameron et al. (2008) in cases with a small number of clusters, typically ≤30 (our analysis is near but not belowthis threshold).

* p < 0.10,

↵** p < 0.05,

↵*** p < 0.01 (based on cluster-robust standard errors).

Notes: The dependent variable is return to prison in the same state within one or three years of release as indicated by the panel title. Minimum wage is measured in nominal dollars and is measured in the state and month the offender was released. All specifications include state and year fixed effects, as well as the individual and time-varying state-level controls outlined in Section III. State EITC policy is included as a dummy variable, but its coefficient is not shown. For baseline means, see Table 4. Robust standard errors clustered at the state level are shown in parentheses (43 clusters). p-values from 1,000 wild-cluster bootstrap iterations are shown for the main minimum wage coefficient, as suggested by Cameron et al. (2008) in cases with a small number of clusters, typically ≤30 (our analysis is near but not below this threshold). Results are based off of Table 4 Columns 1 and 3, which are repeated in Column 1 “Main.” Additions to these regressions are indicated in the column headers. Future minimum wage change is the maximum minimum wage within the window 13–24 months after release minus the minimum wage 12 months after release; it is meant to measure a future minimum wage change that does not occur within one year of release—hence it is only included in the one-year recidivism results. Bound minimum wage is an indicator for a state–year that was bound by one of the three federal changes in 2007, 2008, or 2009—that is, had a state minimum wage below the new federal level at the beginning of the year.

↵* p < 0.10,

↵** p < 0.05,

↵*** p < 0.01 (based on cluster-robust standard errors).

Notes: The dependent variable is return to prison in the same state within one or three years of release (indicated in the column heading). Minimum wage is measured in nominal dollars. State EITC is an indicator for the existence of a state top-up. State EITC percent is the percent of the federal EITC available to those in that state measured in percentage points. All are measured in the state and month the offender was released. All specifications include state and year fixed effects, as well as the individual and time-varying state-level controls outlined in Section III. Robust standard errors clustered at the state level are shown in parentheses (43 clusters). p-values from 1,000 wild-cluster bootstrap iterations are shown for the main minimum wage coefficient and the EITC coefficients, as suggested by Cameron et al. (2008) when the number of clusters is small.

↵* p <0.10,

↵** p < 0.05,

↵*** p < 0.01.

↵a To address the smaller number of treated clusters for EITC estimates, the EITC wild boostrap errors are estimated at the state–year subcluster (MacKinnon and Webb 2018; Roodman et al. 2019).

Notes: In Columns 1–8 the dependent variable is return to prison for a certain crime type within one or three years of release (as indicated by column headers). Columns 9–20 are one- and three-year rates of returning to prison within subsamples based on either education level or race/ethnicity. >HS means any college, not necessarily a college degree. Minimum wage is measured in nominal dollars in the state and month the offender was released. Mean recidivism rates are the mean of the dependent variable for the respective column. All specifications include state and year fixed effects, as well as the individual and time-varying state-level controls outlined in Section III. State EITC policy is included as a dummy variable but its coefficient is not shown. Robust standard errors clustered at the state level are shown in parentheses (43 clusters). p-values from 1,000 wild-cluster bootstrap iterations are shown for the main minimum wage coefficient, as suggested by Cameron et al. (2008) in cases with a small number of clusters, typically ≤30 (our analysis is near but not below this threshold). Outcome heterogeneity by gender is examined in Table 6.

↵* p < 0.10,

↵** p < 0.05,

↵*** p < 0.01 (based on cluster-robust standard errors).

Notes: Only female offenders are included in this table. In Columns 1–8 the dependent variable is return to prison for a certain crime type within one or three years of release (as indicated by column headers). Remaining columns the dependent variable is return to prison within one or three years of release (as indicated in column headers). Columns 9–20 are subsample results based on either education level or race/ethnicity. >HS means any college, not necessarily a college degree. State EITC is an indicator for the existence of a state top-up and is measured in the state and month the offender was released. Mean recidivism rates are the mean of the dependent variable for the respective column. Robust standard errors clustered at the state level are shown in parentheses (43 clusters). p-values from 1,000 wild-cluster bootstrap iterations are shown as well, as suggested by Cameron et al. (2008) when the number of clusters is small.

↵* p < 0.10,

↵** p < 0.05,

↵*** p < 0.01 (based on cluster-robust standard errors).

↵a To address the smaller number of treated clusters for EITC estimates, the EITC wild boostrap errors are estimated at the state–year subcluster (MacKinnon and Webb 2018; Roodman et al. 2019). Results for men can be found in Online Appendix Table G.4.

Notes: Data from the Current Population Survey Outgoing Rotation Groups, population stratification, and years are indicated in the panel titles. Panel A focuses on teenagers (15–19 years old),the targeting population from Allegretto et al. (2011) and Neumark et al. (2014). Panels B and C focus on “low-skill” Black men and women, with low skill here indicating the absence of postsecondary education. Each cell is a different regression. Elasticities are calculated by dividing the coefficient by the mean employment rate for the relevant population. Controls included in all regressions: age, nonseasonally adjusted unemployment rate, marital status, education, race/ethnicity, gender, quarter FE, state FE, and additional trends or FE as noted (in Panel A all specifications also include proportion of population aged 15–19 as in Allegretto et al. 2011). Column 1 provides a baseline estimation. Columns 2–4 replicate the key specifications of Allegretto et al. (2011). Columns 5–7 include different polynomial time trends, replicating the key specifications from Neumark et al. (2014). Regressions are weighted using the person-level weight wtfinl. Standard errors clustered on state in parentheses.

↵* p < 0.10,

↵** p < 0.05,

↵*** p < 0.01.