This paper estimates the effects of reduced hours and job loss on criminal offense using administrative court records
matched to payroll and credit data from 2000-2011 in Texas. I find that negative employment shocks increase criminal behavior substantially.
These effects are concentrated among young men, who experience a roughly 50% increase in offense risk in the year following separation.
Effects are also generally larger for those with prior criminal records, lower prior earnings, and worse credit. Across these series of
estimates, I find that crime responses are particularly pronounced in the months immediately following displacement - a finding that is obscured by
aggregation in prior work. Effects are exacerbated when looking specifically at economically-motivated crime and are sensitive to local economic and
criminal conditions. Consistent with predictions arising from the Becker-Ehrlich model, these crime effects do not fully abate even when an individual
is reattached to the labor market following displacement. Estimates of criminal offense also persist in instances of partial separation, whereby
individuals are not fully laid off, and effects are nearly proportional to earnings lost. Subsequent estimates of the relationship between earnings and
crime reveal that decreasing earnings by 20% increases relative crime risk by 6-8% for young men one year post-separation. Costs to the state
associated with increased propensity for crime following displacement are large relative to costs associated with depressed employment
opportunities, indicating that crime is an important component of the social cost of job displacement.
We leverage variation in the timing of unconditional housing recipiency by homeless
individuals in Los Angeles County to determine the effects of housing the homeless
on their employment, earnings, and benefits absorption. Placement into 2-year Rapid
Re-Housing (RRH) increases extensive-margin labor market participation by nearly
60% from a baseline of 19pp, while Permanent Supportive Housing (PSH) recipients
exhibit a 25% increase in extensive-margin employment from a baseline of 7pp. We
find little evidence of heterogeneous response based on family-status for RRH recipients,
but we do find a mildly positive employment effects for heads-of-households in
PSH. We characterize earnings and benefits responses based on ex-post employment
transition type and find that both U2E and E2E transitioners report income increases
of approximately USD 800-1000 and USD 200 per month respectively while exhibiting
little-to-no change in benefits absorption. These groups outnumber E2U transitioners
by a factor of between 2.5-5. Finally, we estimate the program cost-offset specifically
through earnings effects during program tenure (ignoring other externalities). We estimate
the cost-offset of these policies during program tenure specifically attributable
to earnings effects at 1% for RRH and 0 but nonnegative for PSH (5-10% and 1-9%
for RRH and PSH recipients respectively employed post-event).
This paper attempts to recover the causal effects of criminal debt assigned at sentencing (LFOs) on subsequent criminal behavior.
We construct an estimator that uses judge stringency as an instrumental variable. Additionally utilizing the uniform criminal sentencing guidelines
in North Carolina since 2009, we circumvent part of the ''multi-dimensionality'' problem that would otherwise make estimation of this relationship infeasible.
Preliminary evidence shows that $100 of additionally assigned debt decreases the probability that an individual will reoffend within 12 months by 0.6-0.7pp (1.4-1.8%),
though this estimate likely masks significant heterogeneity. This is in contrast to OLS estimates, which predict a (zero or) positive relationship between
assigned debt and subsequent reoffense. We explore mechanisms that may explain this difference.
I outline how standard one-sided censoring problems in decentralized data systems present themselves in a variety of common settings.
Using administrative State and Federal criminal data in North Carolina and Texas, as well as data informing residence in other states, I show that this one-sided
censoring is broadly non-random in the crime setting. This non-random outcome censoring results in biased estimates of recidivism using standard procedures, but can be corrected
using additional information. Unsurprisingly, this bias shrinks as the data becomes more uniform or complete.
We test the effects of assignment to a collaborative model of post-release community supervision (PRCS), which emphasizes release planning,
prioritizes the officer-client relationship, and invites the client to actively participate in their reentry process.
We find that clients randomly assigned to the collaborative model are 17pp more likely to report to their first probation meeting within the required 48 hours
following release. In the longer term, we find that intervention clients are 14pp less likely to have their probation revoked in the year following release.
Our results broadly demonstrate that a collaborative model of post-release community supervision helps high-risk clients successfully complete their supervision term.
Works in Progress
CCT Killed the RDD Star(s)?
One drawback to the regression discontinuity design is that the estimators require a choice of
bandwidth, which is often chosen ad hoc. There are several bandwidth selectors in the literature; however,
most follow a procedure that leads to bias in the distributional approximation of the estimator. Calonico,
Cattaneo, and Titiunik (2014) propose adjusted confidence intervals for regression discontinuity treatment effects
that are robust to large bandwidths. In this short paper, I analyze how this procedure impacts the findings of a number of widely influential quasi-experiments.
The Los Angeles Riots, Recovery, and Misallocation
In the aftermath of the 1992 Los Angeles Riots, the city embarked on a recovery and revitalization effort, aiming to address
the systemic issues that underpinned the riots. A significant component of this recovery strategy was the establishment of the Los Angeles Revitalization
Zones (LARZ), a policy initiative designed to revitalize the most affected areas. This paper investigates the impact of these riots on subsequent revitalization efforts.
Precise data on arrests and structural damage is utilized to measure exposure, while distance from pre-determined command posts is utilized as an instrument for
this exposure. Riot exposure is found to significantly reduce household income, individual income, and measures of local investment 10 years out. LARZ assignment is shown to
be misallocated under mild assumptions regarding the social planner's objective function.
