Community principles of proportionality that inform prosecution policies and practices may differ even within a county, and represent those of wealthier suburban populations with political power and the ability to prioritize crime reduction without bearing the costs of punishment . Those costs are high. Criminal records, particularly felony convictions, create a broad range of legal and social barriers that persist long after time is served. Restrictions range from the loss of voting rights, parental rights, public benefits that support health and education, employment and occupational licensing, and housing – creating conditions that can in turn impact mental and physical health . Associations between the risk of criminal justice exposure and place of residence present the possibility that collateral consequences will be unequally distributed and exacerbate inequalities by race and location . Considering the significance of criminal history for the severity of punishment for subsequent offenses, including eligibility to receive drug diversion rather than a felony conviction and incarceration, the effects of living in a punitive location are likely to compound over time. Geographic differences in conviction rates also have implications for costs. Punitive charging and sentencing decisions are made by counties but costs are passed on to the state; a felony conviction can receive a sentence to state prison, while county jails and probation supervise those with misdemeanors . In essence, the decisions of more punitive counties to impose higher rates of imprisonment are subsidized by less punitive counties .Can criminal justice reforms reduce widespread geographic variation in case outcomes? Studies of the enactment of three strikes laws have found effects are limited, as local officials ignore or bend the laws to fit existing policies and practices, maintaining geographic disparities . However,commercial vertical farming system mandatory minimum sentencing laws that seek to maximize punishment may affect circumvention differently than reforms that aim to reduce punishment. Little research has evaluated how reforms to reduce punishments affect geographic disparities, a significant evidence gap considering the national trend towards decarceration reforms currently underway. One exception is a study of California’s Proposition 36, passed in 2000, which mandated that individuals arrested for non-violent drug offenses receive treatment instead of incarceration.

The study found an overall reduction in incarceration for drug possession, but relative differences between counties persisted, and were driven by local ideological dispositions and policy preferences regarding drug use . Prior to the passage of Prop 47 in 2014, possession of a controlled substance and possession of concentrated cannabis were classified as “wobbler” offenses, which are charged as felonies by default, but provide prosecutors with the discretion to reduce them to misdemeanors. This discretion was introduced through a penal code amendment in 1969 to reduce caseloads at overburdened superior courts responsible for hearing felony cases, by allowing lesser felonies to be adjudicated as misdemeanors in municipal courts at the prosecutor’s discretion . Research conducted in 2010 found the proportion of arrests for possession of a controlled substance that were charged as felonies varied across California counties from 25 to 100 percent . Even after controlling for case characteristics and criminal history, county of residence was a strong predictor of felony filings following arrest . California law does not dictate how wobblers should be prosecuted, nor does it define the quantity of drug that differentiates possession from sale, the latter of which is always a felony. Charging policies are established by district attorneys and differ across counties; the study found that some simply charged all possession cases as felonies, and others considered the quantity of the drug, prior criminal record, and concurrent charges. The extant research has found that Prop 47 led to fewer arrests, bookings, and custody time on average for Prop 47 offenses , and identified county variation in changes in arrests and jail populations following passage . However, how Prop 47 impacted geographic disparities in the severity of case dispositions has not been investigated. While the reclassification of drug possession offenses to misdemeanors may have reduced county variation in felony convictions for drug possession, felony convictions for concurrent offenses or for drug offenses that remained felonies may have increased in more punitive counties, potentially offsetting a reduction in geographic disparities. This study will assess the effect of Prop 47 on county variation in felony convictions in two ways. First, we will test whether there was a change in county variation in the probability of a felony conviction for those arrested for drug possession.

Within this group, we will examine whether there was an increase in felony convictions for concurrent offenses, which would suggest mitigation of Prop 47’s effects. Second, we will assess the change in felony conviction probability for individuals arrested for non-Prop 47 felony drug offenses, such as sale and transport, which may also result in Prop 47 convictions. California law does not specify the amount of drug that differentiates sale from possession, and those arrested for sale might have their charges reduced to possession. If this group of defendants continues to have charges reduced to possession, which is now a misdemeanor, we would see a reduction in their felony convictions. Yet prosecutors in more punitive counties may use their discretion to buffer this effect. For example, the practice of reducing sale to possession during plea bargaining could decline in these counties, potentially increasing cross-county variation in felony convictions following these arrests. Felony conviction. Separately for Prop 47 arrests and non-Prop 47 felony drug arrests, we determined whether the event resulted in a felony conviction for any offense associated with the arrest. We used any felony conviction as our primary outcome, because prosecutors have the discretion to consolidate arrest charges into an individual filing, or to alter offenses to negotiate a plea, and the charges prosecutors file may have been affected by Prop 47. For example, it is possible that prosecutors were more likely to file felony charges for non-Prop 47 offenses after passage, to counteract the drop in felonies due to reduced classification of Prop 47 offenses. By defining the outcome as any felony conviction, we attempted to account for possible changes in specific charges filed, and capture the severity of the overall case disposition following the arrest. Arrests with no disposition were assumed not to have been prosecuted. If Prop 47 shifted law enforcement practices, some individuals arrested during the pre-Prop 47 period might not have been arrested had they committed their crimes during the post-Prop 47 period. To assess the plausibility of such compositional changes in the populations arrested, we first compared pre- and post-policy groups on demographic characteristics, concurrent charges,commercial vertical farms and criminal histories, separately for Prop 47 offenses and non-Prop 47 felony drug offenses. Pearson’s chi-squared tests were used for categorical variables and Wilcoxon rank-sum tests for skewed continuous variables.

In the presence of compositional changes, we cannot estimate the effects of Prop 47 on arrest outcomes for individuals who would only have been arrested under pre-Prop 47 conditions, because a comparable group is not represented in the post-Prop 47 period. Furthermore, estimating the effect of reclassification on individuals unlikely to be arrested under the new laws would be of little value. Therefore, propensity score matching was used to assess the effect of the “treatment on the treated,” comparing arrest outcomes only among individuals who were likely to be arrested regardless of the reclassification of offenses. Each individual who was arrested after Prop 47 was matched with an individual who was approximately as likely, given their covariates, to have been arrested after Prop 47 was adopted, but was in fact arrested pre-Prop 47. We generated propensity scores using a logit model predicting the log odds that an arrest occurred during the post-Prop 47 vs. pre-Prop 47 period. Predictors included all available demographic variables, and concurrent arrest and criminal history variables likely to affect the arrest disposition. These consisted of age, gender, race/ethnicity; county and calendar month of arrest; any concurrent arrest, separately for felony or misdemeanor classifications: property, violent, sex, weapons, and other; whether the arrest included a probation or parole violation; number of prior arrests; prior arrest for a Prop 47 drug offense ; a measure of the severity of conviction history ; dummies for types of prior felony convictions, including drug, property, violent, sex, weapons, and other; any prior prison sentence and any prior jail sentence. For sale/transport arrests, we also include whether there was a concurrent Prop 47 drug offense. To accommodate non-linearities in age and the number of prior arrests, we use restricted cubic splines with five knots at equally spaced percentiles of each variable’s distribution. Propensity scores were estimated separately for arrests for Prop 47 and sale/transport offenses.Using within-county one-to-one matching without replacement, post-Prop 47 arrestees were matched on the logit of their propensity score to pre-Prop 47 arrestees, within a maximum of 0.2 of the standard deviations of the logit of the propensity score . For Prop 47 drug arrests, 5.6% of the post-Prop 47 group was dropped due to insufficient matches. For sale/transport arrests, 7.8% of the post Prop 47 group was dropped. Covariate balance across propensity-score matched treatment and control groups was checked to assess the adequacy of the propensity score models. Standardized mean differences in all covariates were less than 5% in both samples. For each arrest category, we used a set of mixed logit models to examine the variance in county probabilities of felony conviction pre- and post-Prop 47 among propensity score matched samples. First, we specified the model to include county-specific random intercepts and random coefficients for the policy effect with an unstructured covariance structure.

This generated an estimate of the covariance of pre-Prop 47 mean felony conviction probability with Prop 47 effects on conviction probability, which would indicate whether counties with higher pre-Prop 47 means declined to a greater degree, thus reducing variance in the outcome. Second, models were specified such that counties had separate random intercepts for pre- and post-Prop 47 periods, which generated an estimate of the variance in county probability of felony conviction in each period. We used a likelihood ratio test to compare the fit of this model using an exchangeable covariance structure, which restricts the variance in pre and post intercepts to be equal, to one with an unstructured covariance structure, which allows the variance to differ pre- and post-Prop 47 implementation, as a test of the change in county variance. The latter model is the most flexible and was used to generate the policy effects on the marginal probabilities of felony conviction. For county-specific estimates of outcomes pre vs. post-Prop 47, we generated empirical Bayes estimates of county random pre- and post-Prop 47 implementation intercepts from the models with unstructured covariance and calculated the linear combinations of fixed and random effects corresponding to pre- and post-Prop 47 periods. We also used fixed effects models with county dummy variables, interacted with the pre-post Prop 47 variable to generate marginal probability estimates of within-county pre-post change. A large proportion of arrests had no disposition . We assumed these cases were either never presented to the district attorney or the district attorney did not file charges, and therefore assigned an outcome of no conviction. We compared case characteristics for those with and without dispositions and found support for this assumption. Those without dispositions were less severe cases, and therefore less likely to result in conviction. For example, a larger proportion had no concurrent arrests . They were also less likely to have concurrent felony arrests . It is possible, however, that some of these cases resulted in convictions and the disposition was never reported by the court. We therefore conducted a sensitivity analysis that assumed the most severe extreme: that all sale/transport cases with missing dispositions received a felony conviction, and that all Prop 47 cases with missing dispositions received felony convictions if they occurred in the pre-Prop 47 period or if they included a concurrent felony arrest in the post period. Prop 47 arrests without concurrent felony arrests in the post period are unlikely to have received a felony conviction, since the Prop 47 offense was at that point classified as a misdemeanor.Changes in the outcomes of Prop 47 arrest events must be considered in the context of potential changes in the composition of arrestees .