Files
Abstract
This dissertation is primarily about improving the criminal justice system, focusing on pretrial detention. Chapter 1 attempts to leverage detainees' (and their attorneys') utility over different detention locations in conjunction with traditional operational improvements to save costs, reduce turnarounds (people whose pretrial detention is longer than their eventual sentence), and improve efficiency. Chapter 2 develops a model to identify turnarounds before they occur so their cases can be intervened upon. Chapter 3 is an experiment which studies people's utility over time, attempting to ascertain what people care about when evaluating potential waits, and how those waits' presentations impact people's utility. In Chapter 1, we consider excessive pretrial detention, which is often caused by inefficient case processing. Pretrial detention is expensive for both the taxpayer in terms of housing costs and for detainees in terms of perceived costs. In the extreme, detainees can be incarcerated longer pretrial than their sentence requires. Using data from the Cook County Sheriff's Office, we explore the drivers of delays in case processing and policies which can reduce the consequences of excessive pretrial detention. We develop a model of detainee behavior that affects their case lengths, and hence, the duration of their pretrial duration. Taking it to the data obtained from the Cook County Sheriff's Office, we estimate detainees' perceived costs of being detained in jail, prison, and on EM. We find that prison is perceived as the most costly housing location, followed by jail, and then EM. Costly housing locations may induce unnecessary delays in case processing. We consider four counterfactual interventions and study their impact. First, we consider operational improvements to court processes that may lower the number of court visits by the detainees. Removing one court visit from detainees' cases can save the jail over 20 million dollars annually and reduce turnarounds by 10.9%. Second, we consider paying the bonds of detainees with lower level charges. Simple fund-allocation policies can reduce the pretrial jail population by 2% and can save taxpayers four times what is paid toward bail. Third, we consider split sentencing: in which sentences are split between incarceration and supervision or probation. We estimate that the jail could save 8.5 million dollars and the courts 2,900 visits annually. Finally, we consider the impact of reducing the perceived costs of being detained in prison. We find that this can shorten case lengths by 193 years annually, remove 2,523 court visits each year, and cut turnarounds by over 40%. In chapter 2, we consider targeted intervention to reduce the incidence of turnarounds. But because turnarounds account for less than 5% of the detainee population, detainees who receive this intervention would need to be selected carefully. This paper attempts to score detainees using data available to jails to predict turnarounds before they happen and prioritize intervention. We develop a scoring method that predicts turnarounds before they occur, using data about the detainee, their case, and their current case length. We also extend this scoring method to prioritize detainees whose cases are predicted to end after a lead time of up to 28 days. These scoring methods rely on two tools: First, a classification algorithm which determines detainees' probability of being a turnaround given their attributes and case length. Second, a proportional hazards model which predicts detainees' probability of their case ending at a certain case length given their current case length. Testing this scoring method with immediate intervention on 100 detainees each month for four months in 2016 results in 58 turnarounds identified per month, 10.1 years of dead days removed each month, and an associated excess housing cost of over 525,000 dollars per month. Incorporating a 28 day lead time for the intervention to be effective results in 52 turnarounds identified per month, 8.2 years of dead days removed each month, and an associated excess housing cost of over 429,000 dollars per month. Finally, in Chapter 3, in a conjoint analysis study, we analyze the relative import of mean duration, variability, line length, and reward for ``everyday" waits: those of moderate duration (less than twenty minutes) and modest reward (approximately five dollars). We find that mean duration and variability are the key drivers of people's disutility over waits. The latter suggests that customers are risk averse with their time, a phenomenon rarely included in queueing models. We also find that the information about a wait---how it is presented and customers' beliefs about it---strongly influences customer utility. We identify three primary information effects: (1) Mean duration appears twice as costly when wait times are presented in aggregate (like ride sharing apps) than when presented per-person (like grocery store lines). (2) People familiar with a wait defer to their prior beliefs in lieu of posted statistical information. And as a corollary to the previous item, (3) posting information about a wait's duration or variability does little to induce more sensitivity to that feature for customers familiar with the wait. The interaction of information and fundamentals can connect people's utility over waits to their behavior in queueing systems. We capture these interactions in a utility function for modelers desiring an empirically grounded specification of people's utility. Finally, we provide a series of managerial insights for practical use by managers and researchers alike.