University of Southern California Dissertations and Theses

Essays on applied microeconomics

(USC Thesis Other)

Essays on applied microeconomics

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON APPLIED MICROECONOMICS
by
Amy Mahler
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2023
Copyright 2023 Amy Mahler
Acknowledgments
I would like to thank my main advisor, Jeff Weaver, for his methodical thinking, technical sup-
port, and clear guidance throughout the dissertation generating process. I am also grateful for the
knowledge and expertise of Vittorio Bassi, Matt Kahn, and Emily Nix. My work has improved
tremendously from the feedback of Geert Ridder, Paulina Oliva, Rob Metcalfe, Jeffrey Nugent, Si-
mon Quach, and Augustin Bergeron. Many thanks also to Nancy Nicosia, Ashlesha Datar, Andrew
Ayres, and Ellen Hanek for their mentorship in producing policy-relevant research.
This endeavor would not have been possible without the invaluable insight of public defenders
throughout the U.S. I am especially grateful to Damon Preston, Cara Cape, and Dawn Gasser
of Kentucky’s Department of Public Advocacy, as well as Steve Hanlon, Erik Stilling, Mary Fox,
Deepak Budwani, Issac Merkle, Ruth Petsch, Geoff Burkhart, and many others.
My work and overall well-being has benefitted tremendously from the friendship, comradery, and
commiseration of the USC PhD students. My coffee breaks, random walks, and coworking days
with Monira Al Rakhis have transformed this program’s twists and turns into humor and wisdom.
My friends and peers Juan Espinosa, Fatou Thioune, Amanda Ang, Jingyi Fang, Tao Chen, Gaia
Rigodanza, Karim Fajury, Rajat Kochhar, Ruozi Song, and Clement Boulle have both dear friends
and invaluable allies against the trials of the first year and beyond.
I am profoundly grateful for the support of my best friend Becky Harris, whose effusive encourage-
ment, wisdom, humor, and strategic thinking helped me to and through this program. Her parents,
Bob Harris and Sue Primmer, were also exceptional at placing everything in perspective and as a
vision for “the other side” of a rigorous academic program.
I could not have undertaken this journey without my partner Andreas Aristidou, whose growth
mindset and persistence inspire me every day. From his core exam advice in my second year to his
unwavering support and belief in me on the job market, Andreas has been my role model, advisor,
ii
champion, and best friend throughout this chapter of my life. I am thrilled to continue building our
future together.
Lastly, I’m deeply grateful for the support of my parents, Anna and Don Mahler, and brother Matt.
Their support and belief in me during many career phases has been invaluable to my success. I
would also like to thank my dog and bird for the entertainment, exercises in cultivating patience,
and moral support.
iii
Table of Contents
Acknowledgments ............................................................................. ii
List of Tables.................................................................................. vi
List of Figures................................................................................. vii
Abstract....................................................................................... viii
1 The Impact of Working Conditions on Case Outcomes: Evidence from the U.S. Public
Defense System ............................................................................ 1
Chapter One: Introduction ....................................................................... 1
Chapter One: Background........................................................................ 5
Chapter One: Data................................................................................ 6
Chapter One: Empirical Strategy ................................................................ 8
Chapter One: Results ............................................................................. 10
Case Outcomes .............................................................................. 10
Heterogeneous Effects ....................................................................... 11
Chapter One: Conclusion......................................................................... 13
Chapter One: Figures ............................................................................. 14
Chapter One: Tables .............................................................................. 18
2 Workloads and Productivity in the U.S. Public Defense System............................ 23
Chapter Two: Introduction ....................................................................... 23
Chapter Two: Data................................................................................ 26
Chapter Two: Empirical Strategy................................................................ 27
Chapter Two: Results............................................................................. 28
iv
Chapter Two: Conclusion......................................................................... 30
Chapter Two: Figures............................................................................. 31
Chapter Two: Tables.............................................................................. 37
3 Can Holistic Defense Rehabilitate Criminal Offenders? ..................................... 44
Chapter Three: Introduction ..................................................................... 44
Chapter Three: Background...................................................................... 46
Chapter Three: Data.............................................................................. 48
Chapter Three: Empirical Strategy .............................................................. 49
Chapter Three: Results ........................................................................... 50
Chapter Three: Conclusion....................................................................... 51
Chapter Three: Figures ........................................................................... 53
Chapter Three: Tables ............................................................................ 54
References..................................................................................... 57
v
List of Tables
1.1 Summary Statistics ......................................................................... 18
1.2 IV First Stage ............................................................................... 19
1.3 IV Estimates: Case Outcomes ............................................................. 20
1.4 Robustness with Different Controls: Convicted .......................................... 21
1.5 Robustness with Different Controls: Log(Incarc. Days) ................................. 22
2.1 Summary Statistics ......................................................................... 37
2.2 IV First Stage ............................................................................... 38
2.3 IV Estimates: Overall Tasks ............................................................... 39
2.4 IV Estimates: Client Communication Tasks .............................................. 40
2.5 IV Estimates: Case Building Tasks ....................................................... 41
2.6 IV Estimates: Negotiation Tasks .......................................................... 42
2.7 IV Estimates: Administrative Tasks ...................................................... 43
3.1 Summary Statistics ......................................................................... 54
3.2 Difference-in-Difference Estimates: Drug Treatment Outcomes ......................... 55
3.3 Difference-in-Difference Estimates: Case Outcomes ...................................... 56
vi
List of Figures
1.1 Balance Test: Assigned Case IV............................................................ 14
1.2 IV Estimates: Nonlinear Workload Effects ................................................ 15
1.3 Robustness: Alternative IV Windows...................................................... 16
1.4 Heterogeneous Effects: Case Outcomes .................................................... 17
2.1 Balance Test: Assigned Case IV............................................................ 31
2.2 Heterogeneous Effects: Overall Tasks ...................................................... 32
2.3 Heterogeneous Effects: Extensive Margin of Task Completion........................... 33
2.4 Heterogeneous Effects: Extensive Margin of Task Completion........................... 34
2.5 Heterogeneous Effects: Intensive Margin of Task Completion............................ 35
2.6 Heterogeneous Effects: Intensive Margin of Task Completion............................ 36
3.1 Rollout of ASW Program Across Offices................................................... 53
vii
Abstract
This dissertation contributes to our understanding of the incentives and constraints faced by public
defenders – attorneys for low income defendants – in the United States. Across the three chapters,
I study the issues faced by public defenders and their clients from different angles. In Chapter 1,
I investigate how the case outcomes of low-income criminal defendants are impacted by the high
workloads of public defenders. In Chapter 2, I explore how the attorney’s workload impacts the
precisewayhe/sheallocatestimeacrosscase-relatedtaskssuchasclientcommunication,negotiation,
and evidence review. Chapter 3 measures how a program which placed social workers in public
defense offices impacted clients facing substance abuse and/or mental health charges. The efficacy
of these attorneys is an important issue to both economists and policymakers, as both criminal
records and prison time have serious downstream consequences for defendants’ labor market and
health outcomes.
viii
Chapter 1
The Impact of Working Conditions on Case Outcomes: Evidence
from the U.S. Public Defense System
Chapter One: Introduction
Justice is said to be provided through the fair and impartial administration of the law. But there
is ample evidence of disparities in the provision of legal services in the U.S. criminal justice system.
Black individuals are incarcerated at six times the rate of whites and face longer sentences for similar
crimes; more broadly, the outcomes of similar cases systematically diverge by judge and defendant
characteristics (Rehavi and Starr, 2014; Shulman et al., 2013; Bindler and Hjalmarsson, 2020). For
defendants, conviction and sentencing have far-reaching consequences for labor market, health, and
overall quality of life outcomes (Hjalmarsson, 2008; Geller et al., 2011; Lovenheim and Owens, 2014;
Aizer and Doyle, 2015; Agan and Starr, 2018).
While empirical evidence has focused on how judges and prosecutors influence case outcomes, far
less is known about the contribution of public defenders (PD), who represent low-income individuals
facing criminal charges. PDs are key inputs in the determination of case outcomes, representing
about 80% of felony cases in the U.S., and are 4.7 times as likely to represent a Black client as a
White client. There is evidence of disparate case outcomes: PD clients face a 17% higher probability
of receiving a prison sentence compared to the clients of private attorneys (Harlow, 2000; Hartney,
2009). As such, understanding how PDs contribute to client outcomes is critical to improving equity
in the larger criminal justice system.
1
Prior work on PDs explores how attorney characteristics impact client outcomes, but there is little
evidence on a large issue facing PDs nationwide: excessive caseloads. The causes of high caseloads
are straightforward: public defense systems are significantly under-funded compared to police or
prosecution, and often lack caseload limits and/or the authority to refuse cases (Justice Policy Insti-
tute, The, 2011). The consequences of caseloads are less clear. Associations have been documented
between higher PD caseloads and worse client outcomes, but may be driven by unobserved differ-
ences between cases (Gottlieb, 2021; Iyengar, 2007). The causal role of workload in determining
client outcomes may be substantial: prior literature finds that high caseloads of judges and physi-
cians significantly worsen defendant and patient outcomes, respectively (Yang, 2016; Rao, 2020;
Hoe, 2022; Freedman et al., 2021;Woodworth, 2020). Effective policies to address workload issues
would also require knowledge of whether workload pressure falls differentially on certain cases in
the PD’s caseload, since biases in decision-making may be exacerbated under high work demands
(Dyrbye et al., 2019).
This paper provides the first causal evidence of the impact of PD workloads on client outcomes,
and whether these effects vary by defendant and case characteristics. I do so by leveraging quasi-
random variation in caseloads in two U.S. states. In many jurisdictions, PD case assignment is
mediated by the discretion of the judge or PD office management. I use data from 16 counties in
which cases are assigned through a set of non-discretionary rules: the PD is assigned to either a
particular courtroom or a specific week of the month; she takes all indigent cases that arrive either
in that courtroom or during that week. As such, new cases are assigned to PDs working under very
different caseload conditions.
I estimate the workload effect using an instrumental variable approach that leverages these offices’
case assignment rules. To measure the PD’s workload during a given case, I calculate the average
number of open cases in each day, averaged over the duration of the case: the average daily caseload
(ADCL). However, the effect of this measure on case outcomes may be biased upwards by reverse
causality: the PD may not close cases due to unobservable complexity in the index case. I therefore
instrument the ADCL measure with the count of cases assigned to the PD in the three months
prior to the index case. I test whether this IV measure has the desirable features of an IV. Balance
checks confirm that the IV is conditionally independent of the index case’s characteristics, while a
2
relevance test shows it is highly relevant to the ADCL since most of the assigned cases would still
be open. Furthermore, I include PD and year-week fixed effects in the IV specification to address
identification concerns related to nonrandom assignment of PDs to offices and other shocks that
might impact PD workloads. As such, identification of the workload effect comes from deviations
from the PD-and year-week-specific levels of average workload.
The empirical analysis focuses on 40,525 felony cases worked by 209 PDs in two states between
2014-2018. I document several key results. First, when the PD’s workload increases by 1 standard
deviation, the defendant faces a 3% increased likelihood of getting convicted, and receives a 5%
longer prison sentence (conditional on conviction), indicating that PD workloads do in fact worsen
case outcomes. From nonlinear analyses, the workload effect appears to be strikingly linear, imply-
ing that policies that redistribute cases across PDs – equalizing caseloads across PDs, for example
– will not improve case outcomes on average. Rather, reducing overall caseloads may improve the
functioning of the criminal justice system for low-income defendants: back-of-the-envelope calcula-
tions suggest that if an office in the sample hired 1 additional public defender, its average conviction
rate would decrease by 0.12 to 1%.
Second, I examine whether certain cases experience worse caseload effects, since PDs may allocate
effortawayfromgroupsthattraditionallyfacediscrimination, suchasracialminorities. Surprisingly,
I find that workloads do not differentially affect the case outcomes of non-white defendants, nor
do these effects vary by gender or age. These results are encouraging for policymakers hoping
to improve the public defense system, as it simplifies the problem to one of improving resources,
without having to also address bias in defense provision.
However, I find evidence that the experience level of the PD matters for the workload effect. For
those PDs who I observe starting their employment during the data period, I find that the clients
of the PD in her first 6 months of employment experience a workload impact that is 6 times larger
than that experienced by the clients of the same PD later on in her career. Conversely, there are
no differential workload effects for the clients of PDs who end their employment during the data
period. Taken together with the main results, these findings indicate that while workloads have a
3
negative incremental effect on defendant case outcomes, PDs are in fact learning and improving in
their ability to manage cases under workload constraints as they gain experience on the job.
This paper contributes to several bodies of literature. First, this study contributes to work on public
defense provision in the U.S. These studies compare attorney performance under different incentive
structures, comparing the case outcomes of full-time, salaried PDs to private attorneys appointed
to the same types of cases (assigned counsel). Iyengar (2007) looks at the federal court system,
Anderson and Heaton (2011) in Philadelphia felony courts, Cohen (2014) and Roach (2014) in large
U.S.counties, Shem-Tov(2022)inSanFrancisco, andAganetal.(2021)inBexarCounty, TX.These
papers unanimously conclude that full-time public defenders “do better” by their clients – obtaining
lower conviction rates and shorter sentence lengths – than their assigned counsel counterparts, but
donotdelveintotheconsequencesofthedifficultworkingconditionsinwhichnearlyallPDsoperate.
This study fills this gap by focusing on the workloads of full-time PDs and the consequences for
their clients.
Second, my findings build on studies of workloads and their impacts on employee well-being and
work quality by examining heterogeneity in its effects and underlying mechanisms. In the economics
of crime literature, Yang (2016) and Rao (2020) find that quasi-random increases in the caseloads
of judges worsen defendant case outcomes. In the private sector, Hamermesh et al. (2014) and
Hummels et al. (2016) find that reductions in hours worked have positive effects on worker health
and well-being. A larger literature studies the effects of crowding on patient outcomes in healthcare
settings such as emergency departments and public health clinics. Many of these studies find
small or null effects of crowding on patient outcomes (Freedman, 2016; Kovacs and Lagarde, 2022;
Maibom et al., 2021; Marks and Choi, 2019). A few recent papers find worsened patient health
outcomes due to crowding, or improvements when new healthcare facilities reduce crowding (Hoe,
2022; Freedman et al., 2021; Woodworth, 2020;Harris et al., 2020). This paper builds on this body
of work by exploring the impact on case outcomes in the novel context of public defense, as well as
studying heterogeneity in the workload effects.
Finally, this study adds to a large body of work that seeks to explain disparities in the criminal
justice system. These studies primarily explore how racial discrimination among criminal justice
4
actors such as judges and police officers drive worse criminal justice outcomes for Black individuals
(Anwar et al., 2012; Arnold et al., 2020, Goncalves and Mello, 2020, Park, 2017, Feigenberg and
Miller, 2021, Depew et al., 2017). This study widens the lens on these disparities in two ways.
First, I consider the role of the public defense system’s high workloads in driving these disparities
nationwide, sincePDsaremorelikelytorepresentclientsofcolor. Second, myfindingthatworkloads
negatively impact case outcomes for PD clients implies that there may also be large disparities for
low income defendants from all racial/ethnic backgrounds.
The remainder of this paper is organized as follows. The next section gives further detail on the
public defense systems within the two states. Section 3 describes the data source and outlines
our baseline sample. The empirical strategy and its accompanying identifying assumptions are
described in Section 4. Section 5 presents the main results of workload effect on case outcomes.
The last section concludes.
Chapter One: Background
The empirical setting for this paper consists of the state-run public defense systems in Kentucky and
the Anonymous state (ANON). Below I situate these systems within the context of indigent defense
systems throughout the 50 states and discuss the workload issues facing their public defenders.
Severalfeaturesof thepublicdefensesystemsinKYand ANONrenderthemadvantageoustostudy-
ing the workload effect. Throughout the U.S., public defense systems might be operated/financed
by the state, by individual counties, or some combination of the two – prior descriptive work has
found systems with central operation and state financing to be more effective at delivering public
defense services
1
. KY and ANON are 2 of the 28 states who administer PD services at the state
level through a centralized-organized and financed system.
2
Within each system, public defense
might be provided by either full-time public defenders or private attorneys who take indigent cases
on a per-fee basis (assigned counsel). Prior economic literature finds overwhelmingly that full-time,
salaried PDs obtain more favorable case outcomes – lower conviction rates, shorter sentences – for
1
https://www.in.gov/publicdefender/files/Right-to-Counsel-Services-in-the-50-States.pdf
2
https://gideonat50.org/in-your-state/
5
defendants than assigned counsel attorneys (Iyengar, 2007; Anderson and Heaton, 2011; Cohen,
2014; Roach, 2014; Shem-Tov, 2022; Agan et al., 2021). Since both the KY and ANON systems
employ solely full-time, salaried PDs, their structures are favorable to providing high-quality public
defense.
Relevant to the study of employee workloads, PDs in these states face exceptionally high caseloads.
TheAmericanBarAssociationdefinedannualmaximumsforPDcaseloadsassuch: caseloadsshould
be no higher than the work equivalent of either 150 felony cases or 400 misdemeanors.
3
PDs in KY
and ANON regularly exceed these standards by several multiples. According the KY system’s 2017
Annual Litigation Report, the average trial attorney caseload included 459 new case assignments, a
caseload volume 55 percent above national standards.
4
ANON’s system faces similar challenges: a
2014studyfoundthatthestatewouldneedanadditional270publicdefendersinordertoadequately
represent the state’s indigent defendants.
Chapter One: Data
I use two data sources from the internal case records of the centralized, state-run public defense
systems in Kentucky and the Anonymous state, respectively. For a given case, both data sources
contain a case identifier, PD identifier, office identifier, open/close dates, charge information (sever-
ity level, number of charges), crime type (drug, violent, property, sex, family-related, or other), and
defendant demographics (race, gender, age).
The Kentucky data source contains case-level information for all cases assigned to PDs between
2017-2018, for all of the 52 public defense offices in the state. The KY data includes rich detail on
the case’s outcome, including whether the defendant was convicted, incarcerated, charged a fine,
put on probation, used a plea bargain. If the defendant was incarcerated, the data also records the
sentence duration.
3
https://sixthamendment.org/the-right-to-counsel/national-standards-for-providing-the-right-to-
counsel/sufficient-time-to-ensure-quality-representation-aba-principle-4/
4
https://www.lanereport.com/86735/2018/02/public-defenders-seek-support-for-gov-bevins-funding-proposal/
6
The Anonymous state data source contains case-level information for all cases assigned to PDs
between 2014-2016, for 4 out of 39 public defense offices in the state. The data details whether the
defendant was convicted and/or used a plea bargain. Notably, the ANON data does not include
information on incarceration or other sentencing details.
While the two data sources detail many case-level outcomes, this paper’s main analysis focuses on
the workload impact on a small set of policy-relevant outcomes. With respect to case outcomes, I
analyze whether the defendant was convicted or used a plea bargain using both data sources. From
the KY data alone, I examine whether the defendant was incarcerated and for how long.
I construct the caseload measure to proxy for the workload of the PD during a given case. While a
simple count of the PD’s assigned cases would capture the inflows of cases to her current caseload, it
would omit how long cases stay open/active, thus under- or over-estimating the PD’s true workload.
Instead, I define the caseload measure as the average daily caseload (ADCL) – the average number
of cases the PD has open during the timespan of the index case. Further, I restrict the timespan in
which cases in the PD’s caseload can be count as “open” by examining when PDs tend to complete
75% of their tasks for a given case using the task-level Anonymous state dataset. This “active”
timespan is defined as the first 90 days for a misdemeanor or low-level felony and the first 180 days
for a high-level felony. Thus the ADCL workload measure can approximate well for how actively
busy the PD is with other cases during the index case.
For the analysis, I pool the KY and Anonymous state data sources and restrict the pooled sample
to improve statistical power and the validity of the results. I justify pooling the sample with
the following rationale – both datasets come from states with centralized, state-run public defense
systems with long histories of substantial caseload pressure, the charge level and crime types are
defined similarly in both states, and PDs face similar variability and levels in their caseloads. I
restrict this pooled sample by keeping only felony cases,
5
removing cases lasting less than 2 weeks,
removing PDs with fewer than 30 cases in the entire data period, removing defendants who are
minors, and removing cases with more than 12 charges, as these cases tend to have greater levels of
unobserved complexity. The final pooled analysis sample covers 55,479 cases taken by 267 PDs in
22 offices throughout both states.
5
Felony cases are quasi-randomly assigned to PDs in both states – this point is elaborated in Section 1.
7
Summary statistics for the variables in this pooled sample are presented in Table 1.1. The majority
of defendants are male and white, and between their mid-20s to mid-30s. 89% of cases are low-level
felony drug or property crime cases with 3 charges on average. Defendants face an average 80%
probability of conviction, 69% probability of using a plea bargain, and 54% chance of incarceration.
Chapter One: Empirical Strategy
The ideal experiment to estimate the impact of workload on case outcomes would take two similar
groupsofPDs, allstartingsimilarcasesatthesametime, andrandomlyassignonegrouptocaseload
X, the other to caseload X +1. The estimated difference between the two index cases’ outcomes
would be the causal impact from the PD working on 1 additional case.
Thequasi-experimentanalyzedinthispaperutilizesaninstrumentalvariablestrategythatleverages
the rigid rules that PD offices in both states use to assign cases to PDs in order to obtain such causal
estimates. Thesewithin-officerulesvaryinprecisedetailacrosstheXXofficesintheanalysissample,
but fall into two broad categories. The first category states that PDs are assigned to a courtroom
and take all
6
cases that flow into that courtroom. I argue that this rule creates quasi-random
variation in the type and number of cases in the PD’s caseload because the courtroom itself is
randomly assigned cases in both KY and ANON. The second rule category assigns the PD to a
fixed week of the month in which she is “on call” to take on new cases for all the local courtrooms
served by the PD office. This rule would similarly generate quasi-random variation in PD caseloads
since new cases flow into the PD office based on the time/location of the crime committed and the
defendant’s financial status, and the PDs do not have discretion over which week they receive cases.
Despite the quasi-random variation in caseload generated by these rules, OLS estimation of the
impact of the proposed caseload measure, the average daily caseload (ADCL), on a case’s outcome
may be biased. The ADCL measure includes both cases that were quasi-randomly assigned to
the PD, but also cases have not been closed yet by the PD. The portion of caseload that persists
because the PD could not close those cases would bias OLS estimates upwards if the PD was unable
6
Some offices assign 2 PDs to a courtroom and have them divide cases by whether the last digit of the case ID is
even or odd, or whether the defendant’s last name starts with A-M or N-Z.
8
to close them due to unobservable complexity in a given case. This would induce bias due to reverse
causality.
To address this potential bias, I introduce an IV, the PD’s count of assigned cases prior to the
index case, and investigate the instrument’s validity. My preferred measure is the count of cases
assigned over the previous 3 months, as this is the most correlated with the ADCL measure; I
conduct robustness tests using assigned case counts other time horizons. Table 1.2 shows evidence
for the Relevance Assumption, since the ADCL is highly correlated with the assigned case count
IV with an F-statistic of 152. This result is intuitive given that most cases in the sample close
between 3-9 months after they open – a large portion of the variation in the ADCL is new cases
coming in. I test the Conditional Independence Assumption by individually regressing each of the
case and defendant characteristics on the IV measure. Figure 1.1 shows the coefficient estimates
from these regressions, which are all near zero with standard errors between -0.0025 and 0.0025.
The Exclusion Restriction would be violated if the assigned case count IV had a direct impact on
the case outcome that was distinct from the impact via ADCL. Such an impact could be notable if
PDs accrue substantial psychological burnout from the count of assigned cases, even if they do not
actively work on those cases during the index case.
I estimate an IV model for public defender i with case c as follows:
First Stage
ADCL
ic
=α 0
+α 1
ACic+γX
c
+θ i
+δ yw
+ϵ ic
(1.1)
IV Regression
Y
ic
=β 0
+β 1
ˆ
ADCL
ic
+γX
c
+θ i
+δ yw
+ϵ ic
(1.2)
The parameter of interest is β 1
from Equation 2.2: this represents the impact of workload measure
ˆ
ADCL
ic
on case outcome Y
ic
, where
ˆ
ADCL
ic
is the predicted value from Equation 2.1. One
potential threat to identification is that PDs with higher unobserved defense skill could be employed
by offices with less severe cases. To account for such time-invariant unobserved characteristics, I
control for PD fixed effects, θ i
. I also include year-week fixed effects δ y
w to account for resource-
9
constraining events that impact the entire PD office, such as crime waves that increase caseloads
across the board. Thus the identifying variation for β 1
comes from PD- and year-week-specific
deviations in the PD’s (predicted) average daily caseload. To improve precision, both equations
control for a vector of case and defendant characteristics, X
c
, which includes defendant race, gender,
and age, as well as charge class, number of charges, and type of crime. I cluster standard errors at
the public defender level.
Chapter One: Results
In this section, I report the causal estimates of the impact of a 1 case increase in PD’s ADCL
on the following case outcomes: whether the defendant was convicted, used a plea bargain, was
incarcerated, and the incarceration sentence length. I then explore how these estimates vary by
defendant demographic characteristics, case severity, and public defender experience.
Case Outcomes
Table 1.3 shows positive and statistically significant impacts of caseload on the likelihood that the
defendant experiences worse case outcomes.
In Column (1), the coefficient of ADCL can be interpreted as the impact of the PD having one
additional open case on whether the defendant of the index case is convicted. This estimate is
significant at 5%, and can be translated into an approximately 2.7% change from the average
likelihood of conviction if the ADCL measure increases by 1 standard deviation (25 cases). This
case outcome can be viewed as the ultimate “win” for the PD over the opposing counsel, since a
non-conviction would clear the defendant’s charges and release him from custody, if applicable.
Column (2) shows that higher caseloads decrease the likelihood that the defendant uses a plea
bargain, and is significant at 10%. This coefficient represents a 2.2% change from the mean of
71%, given a 1 standard deviation increase in workload. In plea bargaining, PDs and prosecutors
negotiate a deal where defendants plead guilty to a lesser charge in exchange for a reduced sentence.
10
Since PDs have career incentives to take cases to trial or get them dismissed, this result suggests
that workloads weaken the PD’s bargaining power with the prosecution.
Column (3) indicates that caseloads do not significantly impact on the defendant’s likelihood of
incarceration, although the sign is positive, as we would anticipate. This result is intuitive because
the majority of the sample is composed of cases with violent and/or drug-related charges, which are
subject to mandatory minimum prison sentences ranging from 1-20 years, depending on the charge
class.
7
Column (4) shows that defendants receive significantly longer prison sentences when caseloads are
higher. In terms of percentage changes from the mean incarceration length, defendants receive
prison sentences that are 5% longer with a 1 standard deviation caseload shock.
A simple back-of-the-envelope calculation suggests that adding an additional PD to an office in the
sample could mitigate these effects. For offices with a relatively higher ADCL among the PDs, such
as the Cynthiana Office in KY, hiring an additional PD would reduce the average case’s ADCL by
10 cases, resulting in a 1% reduced likelihood of conviction for the average defendant.
These results are robust to alternative specifications involving different sets of controls and alter-
native time windows for the assigned caseload instrument. First, I remove all controls and fixed
effects from the regression for the convicted and incarceration length outcomes respectively. Tables
1.4 and 1.5 show the results of this exercise. In each set of results, the coefficient magnitude shrinks
as controls are added, but remains significant. Second, instead of the 3 month time window chosen
for the main specification’s assigned case IV, I define this IV over 1, 2, 4, and 5 months windows
and reestimate the main results. Figure 1.3 shows coefficient magnitudes and confidence intervals
for these alternative IVs that are comparable to the 3 month IV of the main analysis.
Heterogeneous Effects
I investigate whether caseloads have differential impacts on case outcomes depending on the demo-
graphic characteristics of the defendant, the severity of the case, or the experience of the PD by
7
https://apps.legislature.ky.gov/law/statutes/statute.aspx?id=39556
11
interacting these characteristics with the ADCL and IV measures. These results are summarized in
Figure 1.4.
I find limited evidence of differential effects of caseloads on cases with defendant from different de-
mographic backgrounds. In particular, I do not find that higher caseloads worsen case outcomes for
non-white relative to white defendants. I find some evidence that the workload effect on likelihood
of conviction or plea bargaining is lower for male defendants than for females, significant at 10%.
I also find some evidence that caseloads have a weaker effect on the likelihood of incarceration for
older defendants compared to younger ones.
I proxy the severity of the case with three measures: whether the case is classified as a high-level
felony (charge class A or B), whether the charge count is above the median for the sample (3
charges), and whether the charges included a violent or sex offense. I find some evidence that,
when workloads rise, severe cases are more likely to end in a plea bargain. There is also some
evidence that defendants with higher charge counts face lower-magnitude impacts on the likelihood
of incarceration.
Lastly, for the subsample of PDs whose career I observe starting or ending in the data period
(N=123), I estimate whether the workload effect was intensified or muted during the first or last
6 month of the PD’s career. Panel (d) of Figure 1.4 shows that for PDs in the first 6 months of
starting at the PD office, caseloads have a 7 times larger effect on incarceration length than the
caseload effect for those same PDs with greater than 6 months of experience. This evidence suggests
that PDs may learn to manage high workloads better as they gain experience working in an office.
On the other hand, PDs who are within 6 months of leaving the PD office are not differentially
impacted by higher caseloads relative to their performance on cases earlier in their careers. If PDs
are leaving their roles due to “burnout”, as has been documented in other high-volume public service
roles, this does not appear to impact their clients’ case outcomes.
12
Chapter One: Conclusion
Thereisagrowinginterestinunderstandingthedriversofthewell-documentedinequitiesoftheU.S.
criminal justice system. Public defense is an important institution within this system, represents
the majority of felony defendants, and their clients face more severe sentences compared to those
defended by private attorneys. A often-reported central issue in public defense is that of high
caseloads, but empirical assessment is needed to guide policymakers.
This paper documents the causal effects of high caseloads in the public defense systems of two U.S.
states. Using quasi-random variation in caseloads, I show that a defendant is more likely to be
convicted and receives a longer prison sentence when her PD juggles more additional cases. The
caseload impact is surprisingly evenly distributed across cases with different demographic charac-
teristics and severity levels. However, the clients of relatively new PDs are harder-hit by increasing
caseloads, suggesting that there is a learning curve to caseload management.
These results raise important questions about how defendant welfare might be impacted by various
policy levers. For instance, if an additional PD is added to one of the offices in the sample, the
conviction rate would decline by between 0.12 and 1%, but new PDs may contribute to defendants
receiving longer prison sentences over the short-term. To sum up, while high work demands has
clear negative impacts on defendants in the public defense system, more work is needed to evaluate
the welfare trade-offs involved in these settings.
13
Chapter One: Figures
Figure 1.1: Balance Test: Assigned Case IV
Notes: Each plotted point represents the coefficient from a bivariate OLS regression of the case covariate
on the 3-month assigned cases IV, with 95% confidence intervals represented by the capped lines. Each
regressionincludespublicdefenderandyear-weekFEsandstandarderrorsclusteredatthepublicdefender
level.
14
Figure 1.2: IV Estimates: Nonlinear Workload Effects
(a) Convicted (b) Plea Bargain
(c) Incarcerated (d) Log(Incarc. Days)
Notes: Each plotted point represents the coefficient estimate of β 1 from the IV regression (Equation 2.2)
of the case outcome on an indicator for whether the average daily caseload was above the percentile
threshold for the index case, instrumented with the 3-month assigned case IV. Regressions include case
controls specified in Equation 2.2, as well as public defender and year-week fixed effects. 95% confidence
intervals for each estimate are displayed.
15
Figure 1.3: Robustness: Alternative IV Windows
(a) Convicted (b) Log(Incarc. Days)
Notes: Each plotted point represents the coefficient estimate from the IV regression of the case outcome
on the average daily caseload, instrumented with the number of assigned cases over the previous 1-5
months. Regressions include case controls specified in Equation 2.2, as well as public defender and
year-week fixed effects. 95% confidence intervals for each estimate are displayed.
16
Figure 1.4: Heterogeneous Effects: Case Outcomes
(a) Convicted (b) Plea Bargain
(c) Incarcerated (d) Log(Incarc. Days)
Notes: Each plotted point represents the coefficient estimate from the IV regression of the case outcome
on the average daily caseload, interacted with the defendant- or case-covariate listed on the Y-axis.
Regressions include case controls specified in Equation 2.2, as well as public defender and year-week fixed
effects. 95% confidence intervals for each estimate are displayed.
17
Chapter One: Tables
Table 1.1: Summary Statistics
Mean Std. Dev. Obs.
Defendant Characteristics
White 0.65 0.48 39127
Male 0.73 0.44 39173
Age 34.24 10.29 39291
Case Characteristics
High-Level Felony 0.11 0.31 40525
Charge Count 3.24 2.42 40525
Drug Crime 0.32 0.47 40525
Violent Crime 0.09 0.29 40525
Property Crime 0.40 0.49 40525
Sex Crime 0.03 0.17 40525
Family Crime 0.10 0.30 40525
Other Crime 0.09 0.29 40525
Case Outcomes
Convicted 0.79 0.40 38310
Plea Bargain 0.69 0.46 40483
Incarcerated 0.54 0.50 20473
Incarc. Days 504.41 893.19 19696
18
Table 1.2: IV First Stage
(1)
Avg. Daily Caseload
Assigned Cases in Prev. 3 Months 0.634
∗∗∗ (0.0314)
F-Stat 408.1
Obs. 34943
19
Table 1.3: IV Estimates: Case Outcomes
(1) (2) (3) (4)
Convicted Plea Incarcerated Log(Incarc. Days)
Avg. Daily Caseload 0.000863
∗∗ 0.000611
∗ 0.000418 0.00653
∗∗ (0.000419) (0.000352) (0.000480) (0.00315)
Y mean 0.799 0.711 0.567 3.339
% Change from Y Mean 2.741 2.215 1.904 5.016
Obs. 33060 34906 17596 16957
Notes: (1) ADCL is the average daily caseload, an estimate of the β parameter in Equation (3); (2) All specifica-
tions include PD and year-week FEs and control for defendant race, gender, and age, charge class, # of charges,
and crime type; (3) Standard errors are clustered at public defender level; (4) Percentage change from Y mean is
scaled by the standard deviation of ADCL; (5) Larger sample size for conviction and plea bargaining outcomes
due to their availability in the case records of both KY and ANON states, while incarceration was only recorded
for KY.
20
Table 1.4: Robustness with Different Controls: Convicted
(1) (2) (3) (4) (5) (6)
Avg. Daily Caseload 0.00163
∗∗∗ 0.000923
∗∗∗ 0.000832
∗∗ 0.000727
∗ 0.000801
∗∗ 0.000863
∗∗ (0.000263) (0.000300) (0.000403) (0.000395) (0.000398) (0.000419)
PD FE
Year-Week FE
Crime Ctrl
Charge Ctrl
Def. Ctrl
Obs. 34199 34199 34199 34199 34199 33060
21
Table 1.5: Robustness with Different Controls: Log(Incarc. Days)
(1) (2) (3) (4) (5) (6)
Avg. Daily Caseload 0.0128
∗∗∗ 0.00670
∗∗ 0.00686
∗∗ 0.00602
∗ 0.00792
∗∗ 0.00653
∗∗ (0.00445) (0.00261) (0.00337) (0.00336) (0.00318) (0.00315)
PD FE
Year-Week FE
Crime Ctrl
Charge Ctrl
Def. Ctrl
Obs. 17736 17736 17736 17736 17736 16957
22
Chapter 2
Workloads and Productivity in the U.S. Public Defense System
Chapter Two: Introduction
Governmentsarecentraltofacilitatingeconomicdevelopmentwithinthestate. Butwhileeconomists
have thought considerably about overall government quality, less attention has been paid to the
employees providing public services (Finan et al., 2015). Public employees are reputed to lower
government service quality by being lazy, corrupt, or both. But employee performance is in part
determined by institutional features, which may shape the quality of how much and how well em-
ployees execute their functions.
One institutional issue that reportedly plagues many public service areas is the employee’s workload.
Prior work has documented the workload effect in criminal justice and healthcare systems: studies
have found that judge vacancies worsen defendant case outcomes (Yang, 2016; Rao, 2020) and that
emergency department crowding negatively affects patient health outcomes (Hoe, 2022; Freedman
et al., 2021; Woodworth, 2020; Harris et al., 2020). But these studies cannot observe the meditating
factor of public employee performance under workload pressure.
This issue matters to both economists and policymakers. In the economics literature, theoretical
work has established that workers juggling more tasks leads to longer project lengths (Coviello
et al., 2014; Coviello et al., 2015; Marchewka et al., 2020). However, we do not know how specific
tasks within projects are prioritized under workload pressure. For policymakers looking to offset
negative consequences of workloads in public service systems, there are trade-offs between investing
in alleviating workload directly versus investing in employee monitoring and/or performance pay.
23
This paper seeks to answer these questions in the context of public defense, a critical area of public
service with notoriously high workloads. Public defenders (PD) are attorneys provided by the
state to low income individuals facing criminal charges. They represent a large and vulnerable
population in the criminal justice system: 80% of felony defendants in a U.S. are assigned a PD,
and Black defendants are nearly 5 times as likely to be represented by the PD compared to White
defendants (Harlow, 2000; Hartney, 2009). Prior work on PDs explores how attorney characteristics
impact client outcomes, but there is little evidence on a large issue facing PDs nationwide: excessive
caseloads. The causes of high caseloads are straightforward: public defense systems are significantly
under-funded compared to police or prosecution, and typically lack caseload limits and/or the
authority to refuse cases (Justice Policy Institute, The, 2011). The impact of workload on PD
performance on case-related tasks has not been causally identified, but due to the facts that PDs
face low levels of monitoring and receive compensation in the form of flat annual salaries, PDs have
discretion to reallocate time across cases and tasks, with potential consequences for clients.
This paper provides the first causal evidence of the impact of workloads on PD performance, and
explores whether these effects vary by defendant and case characteristics. I explore these questions
using granular time use data from an Anonymous U.S. state and an empirical strategy based on
strict rules governing how cases are assigned to PDs. For a given case, the data describes each
task the PD completed for a given case, the task date, and hours worked on the task. Causal
identification comes from the fact that within certain public defense offices in this state, PDs are
assigned to either the week of the month or a particular courtroom and take all indigent cases in that
unit. Since cases are randomly assigned to courtroom, these assignment rules yield quasi-random
variation in the PD’s caseload when he receives a new case.
I estimate how the PD’s workload impacts the intensive (time spent on tasks) and extensive margins
(number of tasks completed) for work on a case. On the intensive margin, I find that when the
PD’s workload increases by 1 standard deviation, PDs reduce hours worked per case by 18%. This
overall reduction significantly affects particular tasks: PDs spend less time communicating with the
client (-13%), reviewing evidence (-24%), negotiating with the prosecutor (-21%), and appearing
in court (-19%). At the extensive margin, PDs reduce the number of tasks completed by 15%.
Notably, they are significantly less likely to review case-related evidence (-13%) and conduct legal
24
research (-22%). Since these tasks are completed independently by the PD, the fact that they are
de-prioritized with increased workloads suggests that monitoring and/or performance-based pay
could improve PD performance on these tasks.
HeterogeneityanalysesdonotrevealevidencethatPDsworklesshardforclientsfromanyparticular
demographic group, and PDs reduce the total hours spent and tasks completed for cases by a larger
margin when they are new on the job. Interestingly, I find that PDs overall respond to rising
workloads by reducing time spent on and tasks completed for more serious felony cases, compared
to their less-serious counterparts. While the case outcome analyses do not reveal disparities for these
clients, these outcomes may be too broad to capture the relevant margin of sentencing severity for
these high-level felony cases. As such, policymakers should be aware that the welfare of high-level
felony clients may be at greater risk when workloads rise.
This paper contributes to two bodies of literature. First, I contribute to work on productivity in the
public sector. Prior work has focused primarily on education and healthcare sectors. In education,
Muralidharan and Singh (2020) surveys recent evidence on the management of school employees,
findingthatperformance-basedpaymayimproveserviceprovision, whileBranchetal.(2012)studies
the value-added of school principals to student achievement outcomes. In healthcare, Janke et al.
(2019) investigate whether hospital CEOs affect the performance of large and complex public sector
organizations, and find little evidence of CEOs’ impact on hospital production. Closest to this
study is Kahn and Li (2020), who document that judges are less productive on polluted days. This
paper contributes to this literature by estimating of workload on granular aspects of an individual
employee’s performance via the time use data.
Second, I contribute to a small literature on the impact of multitasking on worker performance.
Coviello et al. (2014) model task juggling (multitasking) by workers and derive the result that more
task juggling leads to longer overall project times; Coviello et al. (2015) empirically verify this
result, finding that larger judge caseloads increase the length of trials in the Italian court system.
Through a lab experiment, Marchewka et al. (2020) show that multitasking under time pressure
worsened performance on tasks among participants. To this body of evidence, this paper adds a
25
deeper analysis of which tasks are prioritized when public defenders are induced to multitask more
due to workload constraints, and finds a similar worsening of performance in the field context.
Chapter Two: Data
This section describes the data source for this paper, which variables it contains, and how it was
collected.
The data comes from the internal case records of the centralized, state-run public defense systems
in an Anonymous U.S. state, covering cases for four PD offices between 2014-2016. The data is
anonymous due to the sensitivity of the information it contains, which is detailed further below.
First, for a given case, the data contains standard case details: a case identifier, PD identifier,
office identifier, open/close dates, charge information (severity level, number of charges), crime type
(drug, violent, property, sex, family-related, or other), and defendant demographics (race, gender,
age).
The data’s sensitivity comes from the fact that each observation is a task that the PD completed
for a given case, and includes the task’s date, hours spent on it, and a detailed description of
the task itself. For example, I observe that on 12/29/2014, PD 59 spent 25 minutes on a client
phone call for case 000010FE. The primary categories of tasks include communicating with the
client in person, communicating with the client via phone or email, reviewing discovery (material
or testimonial evidence), performing legal research, negotiating with the prosecutor, appearing in
court, and completing administrative tasks.
For the outcome measures, for each of the task categories described, I create an indicator variable
that records whether task was completed for the case, and a continuous variable that records time
spent on the task in hours. Table 2.1 summarizes these outcome measures as well as defendant and
case characteristics. Per case, PDs complete 11 tasks on average and spend about 6 hours total on
case-related tasks. At the extensive margin, PDs are most likely to communicate with the client
in person and appear in court. They spend most task time on client communication, in court, and
26
negotiating for a given case. Descriptively, the sample consists of defendants who are non-white,
male, and in their 30s, and of cases with low-level drug or property crime charges.
Thisdatawasgeneratedduringthe2014-2016periodinwhichtheAnonymousState’spublicdefense
systemadministrationmandatedthatPDslogtheircase-relatedtasks, aswellasthetimetheyspend
on those tasks, in order to better understand their caseload pressures. These findings were reported
to the state legislature in order to better understand PD workloads.
Chapter Two: Empirical Strategy
The quasi-experiment analyzed in this paper draws on the fact that, within the public defense
offices studied, cases are assigned to PDs by strict rules. These rules fall into two categories: PDs
are assigned to a particular courtroom and take all indigent cases therein, or they are assigned to
a week of the month and take all cases arriving during that week. As such, neither the PD nor
the office has discretion over whether a PD is assigned an arriving case. Since PDs work within
offices and offices serve sets of counties, a case’s arrival on a particular PD’s desk is determined only
by the location where the offense was committed. For the purposes of identification here, the case
assignment is therefore quasi-random.
I use an instrumental variable strategy for the analysis due to some endogeneity in the treatment.
The workload treatment is defined as the average daily caseload (ADCL), the average number of
cases the PD had open over the life of the index case. Since this is a measure of open cases,
the ADCL may be endogenous if the PD cannot close cases due to some unobserved complexity.
I therefore instrument the ADCL with the count of cases the PD was assigned during the three
months prior to the index case.
I estimate an IV model for public defender i with case c as follows:
First Stage
ADCL
ic
=α 0
+α 1
ACic+γX
c
+θ i
+δ yw
+ϵ ic
(2.1)
27
IV Regression
Y
ic
=β 0
+β 1
ˆ
ADCL
ic
+γX
c
+θ i
+δ yw
+ϵ ic
(2.2)
The parameter of interest is β 1
from Equation 2.2: this represents the impact of workload measure
ˆ
ADCL
ic
on the task outcome Y
ic
, where
ˆ
ADCL
ic
is the predicted value from Equation 2.1. A
potential threat to identification is that PDs with higher unobserved defense skill could be employed
by offices with less severe cases. To account for such time-invariant unobserved characteristics, I
control for PD fixed effects, θ i
. I also include year-week fixed effects δ y
w to account for resource-
constraining events that impact the entire PD office, such as crime waves that increase caseloads
across the board. Thus the identifying variation for β 1
comes from PD- and year-week-specific
deviations in the PD’s (predicted) average daily caseload. To improve precision, both equations
control for a vector of case and defendant characteristics, X
c
, which includes defendant race, gender,
and age, as well as charge class, number of charges, and type of crime. I cluster standard errors at
the public defender level.
I evaluate the IV assumptions to ensure the relevance and validity of the instrument. Evidence for
theRelevanceAssumptionis evaluatedin Table2.2, which showsthat the ADCLishighly correlated
with the assigned case count IV with an F-statistic of nearly 300. Since most of the cases in the
sample close between 3-9 months, this high correlation makes sense – this set of assigned cases are
still open. I test the Conditional Independence Assumption by individually regressing each of the
case and defendant characteristics on the IV measure. Figure 2.1 shows the coefficient estimates
from these regressions, which are all near zero with standard errors between -0.0025 and 0.0025.
The Exclusion Restriction would be violated if the assigned case count IV had a direct impact on
the case outcome that was distinct from the impact via ADCL.
Chapter Two: Results
To evaluate how PDs reallocate time and attention across tasks when workloads rise, I utilize the
IV strategy and time use data from the Anonymous State and report the results in this section.
28
I begin at the high-level by exploring the extent to which PDs reduce the overall intensive (time
spent) and extensive (tasks completed) margins of effort towards case-related tasks. Using the same
IVspecificationandcontrolsinequation 2.2, Iestimatethecaseloadeffecton(1)thetotalnumberof
hours spent on the tasks and (2) the total count of tasks completed. As evidenced by Table 2.3, PDs
reduce their intensive margin of effort considerably: PDs spent 18% less total time on a case when
workloads rise by 1 standard deviation. This overall decline in total hours falls disproportionately
on particular tasks: PDs spend less time communicating with the client (-13%), reviewing evidence
(-24%), negotiating with the prosecutor (-21%), and appearing in court (-19%), as shown in Tables
2.4, 2.5, and 2.7.
An interesting behavioral observation comes from the extensive margin of task completion. First,
I find that with a 1 standard deviation increase in workload, PDs reduce the number of tasks
completed by 15% (Table 2.3). Which tasks are deprioritized to the point of removal? Table
2.5 shows that PDs cut “behind the desk” tasks: they are 13% less likely to review discovery
(case evidence), and 22% less likely to conduct legal research. These tasks are notably those that
the PD completes alone, in contrast to tasks such as appearing in court, communicating with the
client, or negotiating with the prosecutor. These findings suggest that increasing monitoring and/or
implementing performance pay may improve PD time allocation across tasks to ensure these less-
observed tasks are completed.
Lastly, in Figures 2.2 - 2.6, I explore whether PD time use differs by the previously explored
dimensions of heterogeneity: defendant characteristics, case severity, and PD experience. I do
not find evidence that defendant demographics makes any difference on how PDs spent case time.
However, it does appear that PDs with relatively less experience experience stronger workload
effects: I find some evidence that new PDs spend less time and complete fewer tasks on cases in
their first 6 months, compared to their cases later on. Interestingly, I also find strong evidence
that higher caseloads induce PDs across experience levels to spend disproportionately less time
and complete fewer tasks for more serious cases compared to low-level ones. This finding may
indicate some optimizing behavior on the part of PDs, since low-level, non-violent cases represent
the majority of their caseloads.
29
Chapter Two: Conclusion
This paper finds that as workloads rise, public defenders’ performance is impaired in completing
case-related tasks. I find that while PDs reduce time spent for all tasks, they are more likely to
altogether remove less observed tasks such as the case building tasks of legal research and discovery
review. Furthermore, I find that the relatively serious felony cases experience a disproportionately
greater reduction in time spent and tasks completed under workload pressure. Employee time use
under workload pressure needs further exploration in other public service areas, as the quality of
public service provision depends heavily on the working conditions faced by public employees.
30
Chapter Two: Figures
Figure 2.1: Balance Test: Assigned Case IV
Notes: Each plotted point represents the coefficient from a bivariate OLS regression of the case covariate
on the 3-month assigned cases IV, with 95% confidence intervals represented by the capped lines. Each
regressionincludespublicdefenderandyear-weekFEsandstandarderrorsclusteredatthepublicdefender
level.
31
Figure 2.2: Heterogeneous Effects: Overall Tasks
(a) Total Hours (b) Task Count
Notes: Each plotted point represents the coefficient estimate from the IV regression of the task outcome
on the average daily caseload, interacted with the defendant- or case-covariate listed on the Y-axis.
Regressions include case controls specified in Equation 2.2, as well as public defender and year-week fixed
effects. 95% confidence intervals for each estimate are displayed.
32
Figure 2.3: Heterogeneous Effects: Extensive Margin of Task Completion
(a) Client In Person Communication (b) Client Other Communication
(c) Discovery (d) Legal Research
(e) Negotiation
Notes: Each plotted point represents the coefficient estimate from the IV regression of the task outcome
on the average daily caseload, interacted with the defendant- or case-covariate listed on the Y-axis.
Regressions include case controls specified in Equation 2.2, as well as public defender and year-week fixed
effects. 95% confidence intervals for each estimate are displayed.
33
Figure 2.4: Heterogeneous Effects: Extensive Margin of Task Completion
(a) In Court (b) Admin Tasks
Notes: Each plotted point represents the coefficient estimate from the IV regression of the task outcome
on the average daily caseload, interacted with the defendant- or case-covariate listed on the Y-axis.
Regressions include case controls specified in Equation 2.2, as well as public defender and year-week fixed
effects. 95% confidence intervals for each estimate are displayed.
34
Figure 2.5: Heterogeneous Effects: Intensive Margin of Task Completion
(a) Client-In Person Hours (b) Client Other Communication Hours
(c) Discovery Hours (d) Legal Research Hours
(e) Negotiation Hours
Notes: Each plotted point represents the coefficient estimate from the IV regression of the task outcome
on the average daily caseload, interacted with the defendant- or case-covariate listed on the Y-axis.
Regressions include case controls specified in Equation 2.2, as well as public defender and year-week fixed
effects. 95% confidence intervals for each estimate are displayed.
35
Figure 2.6: Heterogeneous Effects: Intensive Margin of Task Completion
(a) In Court Hours (b) Admin Tasks Hours
Notes: Each plotted point represents the coefficient estimate from the IV regression of the task outcome
on the average daily caseload, interacted with the defendant- or case-covariate listed on the Y-axis.
Regressions include case controls specified in Equation 2.2, as well as public defender and year-week fixed
effects. 95% confidence intervals for each estimate are displayed.
36
Chapter Two: Tables
Table 2.1: Summary Statistics
Mean Std. Dev. Obs.
Defendant Characteristics
White 0.31 0.46 13994
Male 0.77 0.42 14009
Age 33.33 10.62 13956
Case Characteristics
High-Level Felony 0.19 0.39 14009
Charge Count 1.96 1.46 14009
Drug Crime 0.25 0.43 14009
Violent Crime 0.15 0.36 14009
Property Crime 0.46 0.50 14009
Sex Crime 0.05 0.21 14009
Family Crime 0.10 0.30 14009
Other Crime 0.07 0.25 14009
Tasks Completed
Task Count 10.97 12.76 14009
Client in Person 0.72 0.45 14009
Client Other Communication 0.56 0.50 14009
Discovery 0.42 0.49 14009
Legal Research 0.15 0.36 14009
Negotiation 0.61 0.49 14009
In Court 0.76 0.43 14009
Administrative 0.21 0.41 14009
Task Hours
Total Hours 5.79 9.48 14009
Client In Person Hours 1.17 1.67 14009
Client Other Communication Hours 0.63 1.21 14009
Discovery Hours 0.57 1.60 14009
Legal Research Hours 0.23 1.31 14009
Negotiation Hours 1.33 4.00 14009
In Court Hours 1.39 1.97 14009
Administrative Hours 0.13 0.39 14009
37
Table 2.2: IV First Stage
(1)
Avg. Daily Caseload
Assigned Cases in Prev. 3 Months 0.575
∗∗∗ (0.0333)
F-Stat 299.3
Obs. 12451
38
Table 2.3: IV Estimates: Overall Tasks
(1) (2)
Total Hrs Task Count
ADCL -0.0849
∗∗∗ -0.131
∗∗∗ (0.0251) (0.0357)
Y mean 5.588 10.357
% Change -18.382 -15.244
Change in Minutes -61.635 -94.730
Obs. 12451 12451
39
Table 2.4: IV Estimates: Client Communication Tasks
(1) (2) (3) (4)
In Person In Person Hrs Other Comm. Other Comm. Hrs
ADCL -0.00119 -0.0119
∗∗∗ -0.000824 -0.00690
∗∗ (0.00115) (0.00370) (0.00128) (0.00334)
Y mean 0.708 1.146 0.563 0.633
% Change -2.033 -12.584 -1.769 -13.189
Change in Minutes -0.864 -8.653 -0.598 -5.009
Obs. 12451 12451 12451 12451
40
Table 2.5: IV Estimates: Case Building Tasks
(1) (2) (3) (4)
Discovery Discovery Hrs Legal Research Legal Research Hrs
ADCL -0.00421
∗∗ -0.0111
∗∗∗ -0.00270
∗∗∗ -0.00505
(0.00165) (0.00380) (0.000892) (0.00305)
Y mean 0.402 0.555 0.151 0.219
% Change -12.688 -24.151 -21.585 -27.880
Change in Minutes -3.058 -8.036 -1.958 -3.661
Obs. 12451 12451 12451 12451
41
Table 2.6: IV Estimates: Negotiation Tasks
(1) (2)
Negotiation Negotiation Hrs
ADCL -0.00249 -0.0224
∗∗ (0.00151) (0.00912)
Y mean 0.599 1.284
% Change -5.026 -21.113
Change in Minutes -1.808 -16.269
Obs. 12451 12451
42
Table 2.7: IV Estimates: Administrative Tasks
(1) (2) (3) (4)
In Court In Court Hrs Admin Tasks Admin Hrs
ADCL -0.00274
∗ -0.0212
∗∗∗ -0.00112 -0.000310
(0.00138) (0.00603) (0.00138) (0.00112)
Y mean 0.751 1.376 0.191 0.108
% Change -4.414 -18.674 -7.107 -3.480
Change in Minutes -1.988 -15.412 -0.814 -0.225
Obs. 12451 12451 12451 12451
43
Chapter 3
Can Holistic Defense Rehabilitate Criminal Offenders?
Chapter Three: Introduction
Criminal justice policy in the U.S. has historically relied on punitive sentencing for offenders. But
policy debate has questioned the efficacy of these sanctions (Kuziemko, 2013; Aizer and Doyle,
2015; Bhuller et al., 2016; Rose and Shem-Tov, 2021). Furthermore, the increasingly punitive
sanctions associated with the “War on Drugs” has led to tremendous growth in the correctional
population, as well as significant racial disparities in incarceration rates (Raphael and Stoll, 2013;
Neal and Rick, 2014; Lofstrom and Raphael, 2016). In recent years, policymakers have made
efforts to promote alternatives to incarceration for nonviolent offenders to reduce the correctional
population without impacting public safety, such as drug, mental illness, or other specialty-focused
courts (Evans Cuellar et al., 2006; Seward et al., 2021; Mitchell et al., 2012; Golestani et al., 2021),
felony diversion (Augustine et al., 2022; Mueller-Smith and T. Schnepel, 2021), or restorative justice
conferencing (Shem-Tov et al., 2021).
A recent, potentially powerful and cost-effective alternative policy that has received less attention
are holistic defense systems. The programs work through existing public defense systems, and have
been discussed and adopted in recent years across the country (Giovanni, Thomas, 2012). Holistic
defense is defined as a public defense system that provides client-centered, high quality defense,
meets clients’ social service needs, involves enhanced consideration of collateral consequences and
other legal issues, collaborates with community programs, and provides systemic advocacy (Ostrom
and Bowman, 2020).
44
This paper aims to measure the impact of a holistic defense program, Kentucky’s Alternative Sen-
tencing Worker (ASW) Program, on client case outcomes such as incarceration and drug treatment.
The ASW program comprises a holistic defense system that brings a multidisciplinary team with
the task of providing alternatives to traditional incarceration sentences. The program focuses in
particular on achieving client buy-in to engage in rehabilitation, treatment, and other social services
through one-on-one interviews with clients and partnerships with community organizations. But
while the features of the program appear promising, its efficacy in diverting clients from incarcera-
tion and promoting engagement with community services has not been formally evaluated.
To study the impact of ASW on client outcomes, I use data from the Kentucky Administrative
Office of the Courts and exploit the staggered timing of the program’s roll-out across counties for
causal identification. Specifically, I use a difference-in-difference model that compares treated offices
to untreated offices, before vs. after the program is implemented at the treated offices. Identifying
variation comes from within-office differences before and after ASW program entry, and within-
year-week differences between treated and untreated offices.
I find that the ASW program did not have a significant impact on a client’s likelihood of receiving
a drug- or alcohol-related sentence, but did impact the case’s sentencing outcomes of receiving jail
or prison time. Results from difference-in-difference models show null effects of the program on
indicator outcome variables for (1) whether the client was sentenced to drug/alcohol education and
(2) whether the client was sentenced to drug treatment, as well as whether the case was dismissed
or diverted from prosecution. However, the program’s roll-out did significantly increase the client’s
likelihood of receiving a jail sentence by 4%, and significantly decreased the likelihood of receiving
a prison sentence by 3%. Since jails hold defendants sentenced to 1 year or less, while prison is
for 1 year or longer, the evidence suggests that the ASW program was effective in obtaining less
time incarcerated – in jail rather than prison – but was not significantly impactful in moving clients
towards substance abuse-related education or treatment.
Thispapercontributestotwobodiesofliterature. First,Iaddtoliteraturethatevaluatestheefficacy
of programs aimed at providing alternatives to incarceration. These programs include evaluations
of specialty courts, including those focused on substance abuse cases, mental health cases, cases
45
involving veterans, or cases involving domestic abuse (Evans Cuellar et al., 2006; Seward et al.,
2021; Mitchell et al., 2012; Golestani et al., 2021). Other work measures on the impact of felony
diversion on recidivism, finding evidence of effective reduction in recidivism rates (Augustine et al.,
2022; Mueller-Smith and T. Schnepel, 2021). Finally, Shem-Tov et al. (2021) studies restorative
justice conferencing, where the victim, the accused, and supporters of both parties participate in
a structured conference, and finds it effective in reducing recidivism. This paper adds to this
literature by measuring the impact of holistic defense, which works through adding resources to the
public defense attorney, rather than working through the prosecution or court, which may be more
cost-effective and systemically sustainable than the aforementioned programs.
Second, this paper adds to a growing literature on mental health interventions in the criminal
justice system. Most relevant to the present study is the work of Seward et al. (2021), who study
the impact of a mental health court that pairs public defenders and social workers on the health,
reoffending, and recidivism outcomes of defendants. They conclude that the contribution of the
social worker to client mental health is greater than that of the public defender himself. Deza et al.
(2022) measure the effect of local access to mental healthcare on crime and crime-related costs,
finding significant reductions in both for an additional ten offices in a county. Lastly, in the context
of Liberia, Blattman et al. (2017) show that an intervention pairing cognitive behavioral therapy
and cash led to large decreases in crime and violence.
Chapter Three: Background
Historically, court systems have used community alternatives to incarceration for drug offenders,
such as court-mandated treatment. While studies have found these policies to have some effect
on drug offender rehabilitation, they have been implemented through the prosecutor, whose role is
adversarial to that of the defendant. Two issues have been pointed out: first, defendant rehabili-
tation depends entirely on the prosecutor, who may have more punitive tastes; second, even if the
prosecutor is on board, the defendant may not be motivated to complete the treatment since he
was sentenced to it by someone in an adversarial role. The Alternative Sentencing Worker (ASW)
Program was proposed to resolve both issues by (1) working through the defendant’s attorney to
46
create the rehabilitative plan, rather than relying on the prosecutor, and (2) giving the client an
active role in determining their interest in/motivation to carry out this plan.
Specifically, the ASW Program provides the Kentucky public defense offices with social workers,
who work with both public defenders and their defendants to create alternative sentencing plans to
present in court. The primary goal of the program is to “maximize clients’ liberty interest while at
the same time attaining client engagement in constructive use of probation or diversion sentences in
lieu of incarceration” (Department of Public Advocacy, 2016). This program was rolled out across
all but one of the 35 public defense offices in Kentucky between 2000-2017. The number of offices
receiving a social worker for the first time is shown in Figure 3.1: most offices receive a social worker
in 2015, but there are some early and late adopters.
Once assigned to an office, the social worker plays roles on individual cases and for the office as
a whole. For an individual case, there are three components to each social worker’s role. First,
the social worker interviews the defendant to assess her needs and collects information on her
social history from family, friends, and coworkers. This interview uses a counseling technique called
Motivational Interviewing (MI), which is a communication method for enhancing a client’s intrinsic
motivationtochangebyunderstandingandresolvinghisambivalencetochange(MillerandRollnick,
2009; Morgenstern et al., 2017; Fcfp and Richard, 2007). In this context, the social worker elicits
the client’s desire to enroll in and complete drug treatment and/or mental health services – aiming
to facilitate rather than direct the client towards these services. Second, the social worker assists
the public defender in creating an Alternative Sentencing Plan, which would summarize the social
worker’s assessment of the defendant’s background and motivation to participate in a treatment
plan, providing support for a non-incarceration sentence. The public defender presents this plan to
the court. Finally, if the Alternative Sentencing Plan is accepted by the court, the social worker
facilitates the client’s treatment through making referrals to providers and following up with clients
12 months later to see their progress.
For the office as a whole, the social worker serves as a liaison to community programs, developing
closer working relationships and easing referral procedures to promote cooperation among service
providers and the court system. Social workers also provide consultations for public defenders about
47
client needs and approaches with cases that were not directly assigned a social worker. As such, the
social worker may in fact have some impact on all cases flowing through their home office.
Chapter Three: Data
To determine the impact of the ASW program on defendant outcomes, I use the universe of criminal
court records from Kentucky’s Administrative Office of the Courts. The analysis sample is at the
case level and consists of 912,304 cases from the 35 PD offices in KY between 2002 - 2022. For a
given case, the record lists the case identifier, county, open/close dates, charge information (severity
level, number of charges), crime type (drug, violent, property, sex, family-related, or other), and
defendant demographics (race, gender, age).
Table 3.1 shows summary statistics for the cases in the sample. The majority of defendants are male
and in their thirties. 85% of defendants are white, reflecting the racial composition of the state.
1
Sample cases have approximately three charges, and 70% of the sample faces either low-level (Class
D) felony or Class A misdemeanor charges. Cases are mostly drug- or property-related crimes, such
as drug possession or robbery.
The primarily aim of the ASW program is to promote education and/or treatment over carceral
outcomes such as jail or prison time. Thus the hypothesized direct effect would include both an
increase in rehabilitation-focused sentencing outcomes and a decrease in carceral outcomes. In the
data, the rehabilitative outcomes include getting sentenced to alcohol/drug education (these are
grouped together in the KY court records) or sentenced to drug treatment, and approximately
20% of cases receive at least one of these sentences. The other direct effect would be on carceral
outcomes, which appear frequently in case sentencing: 9% of cases receive both jail and prison time,
60% receive only jail time, 20% receive only prison time.
The ASW program may also have indirect effects on case outcomes if social workers increase the
public defender’s bargaining power with the judge and/or prosecutor. I proxy this by estimating
the effect of the program rollout on whether the case’s charges are dismissed altogether, or the case
1
https://www.census.gov/quickfacts/KY
48
is diverted from prosecution, such that charges are either dismissed or dropped altogether. While
about a third of cases have charges dismissed, diversion is rare at approximately 1% of cases.
The KY court records data has the limitation that I do not know whether the case actually had a
social worker involved, so the effect measured will be intent-to-treat.
Chapter Three: Empirical Strategy
How did the ASW program impact case outcomes in treated offices? To answer this question, I
exploit the staggered timing of the program’s rollout across different geographic regions over time.
If the program has achieved its aims, then clients facing drug charges in treated offices should have a
higher chance of drug education and/or treatment-related sentencing, and a lower chance of prison
time.
The “naïve” OLS estimation of the ASW program’s effect would compare cases who received a social
worker to those who did not. There are two sources of bias in such estimates: first, cases are not
randomly assigned to social workers. The Directing Attorney (head) of the PD office assigns social
workers to cases, prioritizing Circuit Court cases where it is likely they can prevent the client from
going to prison (Department of Public Advocacy, 2022). As such, there are likely unobserved factors
that affect both the likelihood of being assigned a social worker as well as the case’s outcome that
would bias the estimates. Second, the social workers provide resources to the office as a whole, so
even the cases not actually assigned a social worker may also be “treated” by the program’s entry
into the office, since all attorneys may learn of the community resources and defending arguments
to use from the treated cases.
As such, I will use a difference-in-difference strategy that exploits the staggered timing of the ASW
program’s rollout. This model will serve as a baseline which will then be extended to a triple-
difference model. Recall that the final sample consists of N=912,304 cases in public defense offices
between 2002-2022. I start from the following estimating equation for case i, office j, year-week t
Y
ijt
=β DiD
ASW
jt
+X
′
ijt
Θ+ δ j
+λ t
+ϵ ijt
(3.1)
49
Y
ijt
representsanindicatorvariablethattakesonavalueofoneintheeventofeachthefollowingcase
sentencing outcomes: drug/alcohol education, drug/alcohol treatment, diversion, jail, and prison.
The indicator ASW
jt
is equal to one for the cases and offices exposed to the ASW program. As
such, the parameter of interest, β , represents the differential effect of the ASW program on case
outcomes for cases in treated relative to untreated offices. X
′
ijt
is a set of observable case and
defendant characteristics, which include defendant age, gender, and race, number of charges, charge
severity, and crime type. I also include δ j
and λ t
which represent office and year-week fixed effects,
respectively. Finally, standard errors are clustered at the office level to allow for dependence in case
outcomes within offices.
Identifyingvariationcomesfromtwosources: withinofficedifferencesbeforeandafterASWprogram
entry, and ASW versus non-ASW differences in the same year-week. β is identified as long as the
outcomes of cases in treated offices follow similar trends to cases in untreated offices.
Recent literature documents that two-way fixed effects models can be susceptible to bias in set-
tings with staggered treatment adoption (Callaway and Sant’Anna, 2021; Borusyak et al., 2023;
Goodman-Bacon, 2021; Sun and Abraham, 2021). More precisely, unless strong assumptions on
treatment homogeneity hold, standard TWFE models may not represent the weighted average of
unit-level treatment effects due to the fact that regressions make both “clean” comparisons between
treated and not-yet-treated units as well as “forbidden” comparisons between units that have both
been treated already. In this study’s context, the homogeneity assumption entails early and late
ASWprogramadoptersexperiencethesamepathoftreatmenteffects. Thismaynotbetrue. Future
versions of this paper will re-estimate the main regressions using the interaction weighted estimator
of Sun and Abraham (2021). This estimator purges the bias discussed above by comparing ASW
offices only to non-ASW offices (and removing yet-to-adopt ASW offices).
Chapter Three: Results
In this section, I report the difference-in-difference estimates of the impact of the ASW program
on sentencing outcomes. Table 3.2 shows that the program did not significantly impact client’s
50
likelihood of receiving a drug/alcohol-related sentence. The outcome variables are indicators for
whether the client was sentenced to complete a drug/alcohol education program or to receive drug
treatment. Columns (1) and (2) show these null results for the likelihood of getting sentenced to
drug/alcohol education and drug treatment respectively. While the education outcome’s coefficient
estimate is fairly large and positive, as expected, the estimate is imprecise so as to be insignificant.
More surprisingly, the coefficient for the drug treatment outcome is negative (albeit small and
insignificant), which is not what we would expect from a program aimed at advocating for clients
with addiction issues to receive education and/or treatment. One possible explanation for these
surprising results is that the drug-related interventions are not part of the client’s court-mandated
sentence, per se, but are rather worked out by the ASW separate from court intervention.
However, Table 3.3 shows that the ASW program had significant effects on case outcomes. While
Columns (1) and (2) illustrate null effects of the program on the likelihood that the case is dismissed
or diverted from prosecution, Columns (3) and (4) show significant effects on the likelihood of
receiving jail or prison sentences. These effects run in opposite directions: while the program
increasesthelikelihoodthattheclientwillbesentencedtoprisonbynearly4%, theclient’slikelihood
of a prison sentence decreases by about 3%. While both jail and prison sentences mean clients are
confined, the length of stay differs: clients would remain in jail for less than a year, but in prison
for at least a year.
Taken together, the results tell a story where clients do not receive more drug-focused interventions,
but the ASW intervention does yield lesser sentences in the form of jail rather than prison time.
Discussions with policymakers in the state have indicated that both jails and prisons often feature
drug detoxification programs, so this may still be part of the client’s sentence, but this is not made
explicit in the court records.
Chapter Three: Conclusion
While criminal justice policy has traditionally relied in punitive sentencing for criminal offenders,
recent programs promoting alternatives to incarceration for non-violent offenders have interested
51
policymakers and stakeholders concerned by the growth of and disparities within the correctional
population. While prior work has examined specialized courts, diversion, and restorative justice
conferencing, the recent movement within the public defense system towards holistic defense pro-
gramming – and its potential for both efficacy and cost-efficient diversion from incarceration – has
been under-explored.
This paper aims to fill this gap by measuring the impact of one such holistic defense program, the
Alternative Sentencing Worker Program of Kentucky, on client case outcomes. Using administrative
data from the KY court system and a difference-in-difference strategy exploiting the staggered
rolloutoftheprogramacrosscounties, Ifindnulleffectsoftheprogramonsubstanceabuseeducation
andtreatmentoutcomes, butsignificantimpactsonclientprisonandjailoutcomes: witha4%higher
likelihood of receiving jail time, and a 3% lower likelihood of prison time. These results indicate
that the program does not appear to be achieving its objective of increasing client use of substance
abuse rehabilitation resources, it is facilitating reductions in incarceration time overall.
To conclude, my findings show that holistic defense programs can reduce sentence length among
drug offenders, but do not significantly impact the likelihood of substance abuse rehabilitation, and
as such can be an effective alternative to traditional criminal justice practices.
52
Chapter Three: Figures
Figure 3.1: Rollout of ASW Program Across Offices
53
Chapter Three: Tables
Table 3.1: Summary Statistics
Mean Std. Dev. Obs.
Defendant Characteristics
Male 0.71 0.45 901451
Asian 0.00 0.04 895263
Black 0.15 0.35 895263
White 0.85 0.35 895263
Age 33.99 10.64 902734
Case Characteristics
Charge Count 3.40 4.23 912304
Class A Felony 0.01 0.07 911742
Class B Felony 0.04 0.19 911742
Class C Felony 0.09 0.29 911742
Class D Felony 0.37 0.48 911742
Class A Misdemeanor 0.33 0.47 911742
Class B Misdemeanor 0.16 0.36 911742
Violation 0.01 0.10 911742
Local Ordinance 0.00 0.02 911742
Drug Crime 0.45 0.50 912304
Property Crime 0.23 0.42 912304
Family Crime 0.05 0.23 912304
Juvenile Crime 0.00 0.01 912304
Sex Crime 0.01 0.10 912304
Violent Crime 0.11 0.31 912304
Traffic Crime 0.05 0.22 912304
Other Crime 0.09 0.29 912304
Case Outcomes
Alcohol/Drug Education 0.14 0.34 688544
Drug Treatment 0.06 0.24 677451
Dismissed 0.34 0.47 911703
Diverted 0.01 0.12 911703
Jail 0.70 0.46 887701
Prison 0.30 0.46 887701
54
Table 3.2: Difference-in-Difference Estimates: Drug Treatment Outcomes
(1) (2)
Drug/Alc Educ Drug Treat
ASW x Post 0.0146 0.00247
(0.0203) (0.0143)
Y mean 0.198 0.066
Obs. 261047 252531
Notes: (1) Covariates include defendant race, gender, and age, case charge class, # of charges, and offense codes;
(2) Fixed effects included for office and year-week; (3) Standard errors clustered at PD office level.
55
Table 3.3: Difference-in-Difference Estimates: Case Outcomes
(1) (2) (3) (4)
Dismiss Divert Jail Prison
ASW x Post -0.0123 -0.00222 0.0351
∗∗∗ -0.0318
∗∗∗ (0.00937) (0.00529) (0.0127) (0.00726)
Y mean 0.357 0.017 0.703 0.248
Obs. 360829 360829 345836 345836
Notes: (1) Covariates include defendant race, gender, and age, case charge class, # of charges, and offense codes;
(2) Fixed effects included for office and year-week; (3) Standard errors clustered at PD office level.
56
References
Agan, A., Freedman, M., and Owens, E. (2021). Is Your Lawyer a Lemon? Incentives and Selection
in the Public Provision of Criminal Defense. The Review of Economics and Statistics, 103(2):294–
309.
Agan, A. and Starr, S. (2018). Ban the Box, Criminal Records, and Racial Discrimination: A Field
Experiment*. The Quarterly Journal of Economics, 133(1):191–235.
Aizer, A. and Doyle, J. J. (2015). Juvenile Incarceration, Human Capital, and Future Crime:
Evidence from Randomly Assigned Judges *. The Quarterly Journal of Economics, 130(2):759–
803.
Anderson, J. M. and Heaton, P. (2011). How Much Difference Does the Lawyer Make? The Effect
of Defense Counsel on Murder Case Outcomes. SSRN Electronic Journal.
Anwar, S., Bayer, P., and Hjalmarsson, R. (2012). The Impact of Jury Race in Criminal Trials. The
Quarterly Journal of Economics, 127(2):1017–1055.
Arnold, D., Dobbie, W., and Hull, P. (2020). Measuring Racial Discrimination in Bail Decisions.
page 83.
Augustine, E., Lacoe, J., Raphael, S., and Skog, A. (2022). The Impact of Felony Diversion in San
Francisco. Journal of Policy Analysis and Management, 41(3):683–709.
Bhuller, M., Dahl, G. B., Løken, K. V., and Mogstad, M. (2016). Incarceration, Recidivism and
Employment.
Bindler, A. and Hjalmarsson, R. (2020). The Persistence of the Criminal Justice Gender Gap:
Evidence from 200 Years of Judicial Decisions. The Journal of Law and Economics, 63(2):297–
339.
Blattman, C., Jamison, J. C., and Sheridan, M. (2017). Reducing Crime and Violence: Exper-
imental Evidence from Cognitive Behavioral Therapy in Liberia. American Economic Review,
107(4):1165–1206.
Borusyak, K., Jaravel, X., and Spiess, J. (2023). Revisiting Event Study Designs: Robust and
Efficient Estimation. arXiv:2108.12419 [econ].
57
Branch, G., Hanushek, E., and Rivkin, S. (2012). Estimating the Effect of Leaders on Public
Sector Productivity: The Case of School Principals. Technical Report w17803, National Bureau
of Economic Research, Cambridge, MA.
Callaway, B. and Sant’Anna, P. H. (2021). Difference-in-Differences with multiple time periods.
Journal of Econometrics, 225(2):200–230.
Cohen, T. H. (2014). Who is Better at Defending Criminals? Does Type of Defense Attorney Matter
in Terms of Producing Favorable Case Outcomes. Criminal Justice Policy Review, 25(1):29–58.
Coviello, D., Ichino, A., and Persico, N. (2014). Time Allocation and Task Juggling. American
Economic Review, 104(2):609–623.
Coviello, D., Ichino, A., and Persico, N. (2015). THE INEFFICIENCY OF WORKER TIME USE:
The Inefficiency of Worker Time Use. Journal of the European Economic Association, 13(5):906–
947.
Department of Public Advocacy (2016). Kentucky Department of Public Advocacy: Alternative
Sentencing Worker Program.
Department of Public Advocacy (2022). Alternative Sentencing Worker Case Priorities.
Depew, B., Eren, O., and Mocan, N. (2017). Judges, Juveniles, and In-Group Bias. page 31.
Deza, M., Maclean, J. C., and Solomon, K. (2022). Local access to mental healthcare and crime.
Journal of Urban Economics, 129:103410.
Dyrbye, L., Herrin, J., West, C. P., Wittlin, N. M., Dovidio, J. F., Hardeman, R., Burke, S. E.,
Phelan, S., Onyeador, I. N., Cunningham, B., and van Ryn, M. (2019). Association of Racial
Bias With Burnout Among Resident Physicians. JAMA Network Open, 2(7):e197457.
Evans Cuellar, A., McReynolds, L. S., and Wasserman, G. A. (2006). A cure for crime: Can
mental health treatment diversion reduce crime among youth? Journal of Policy Analysis and
Management, 25(1):197–214.
Fcfp, M.-T. L. and Richard, C. (2007). The motivational interview.
Feigenberg, B.andMiller, C.(2021). RacialDivisionsandCriminalJustice: EvidencefromSouthern
State Courts. American Economic Journal: Economic Policy, 13(2):207–240.
Finan, F., Olken, B., and Pande, R. (2015). The Personnel Economics of the State. Technical
Report w21825, National Bureau of Economic Research, Cambridge, MA.
Freedman, S. (2016). Capacity and Utilization in Health Care: The Effect of Empty Beds on
Neonatal Intensive Care Admission. American Economic Journal: Economic Policy, 8(2):154–
185.
58
Freedman, S., Golberstein, E., Huang, T.-Y., Satin, D. J., and Smith, L. B. (2021). Docs with their
eyes on the clock? The effect of time pressures on primary care productivity. Journal of Health
Economics, 77:102442.
Geller, A., Garfinkel, I., and Western, B. (2011). Paternal Incarceration and Support for Children
in Fragile Families. Demography, 48(1):25–47.
Giovanni, Thomas (2012). Community-Oriented Defense: Start Now.
Golestani, A., Owens, E., and Raissian, K. (2021). Specialization in Criminal Courts: Decision
Making, Recidivism, and Re-victimization in Domestic Violence Courts in Tennessee. SSRN
Electronic Journal.
Goncalves, F. and Mello, S. (2020). A Few Bad Apples? Racial Bias in Policing. page 107.
Goodman-Bacon, A. (2021). Difference-in-differences with variation in treatment timing. Journal
of Econometrics, 225(2):254–277.
Gottlieb, A. (2021). Making Gideon Count? Public Defender Resources and Felony Case Outcomes
for Black, White, and Latinx Individuals. Race and Justice, page 215336872110064.
Hamermesh, D. S., Kawaguchi, D., and Lee, J. (2014). Does Labor Legislation Benefit Workers?
Well-Being after an Hours Reduction. page 42.
Harlow, C. W. (2000). Defense Counsel in Criminal Cases. page 12.
Harris, M. C., Liu, Y., and McCarthy, I. (2020). Capacity constraints and time allocation in public
health clinics. Health Economics, 29(3):324–336.
Hartney, Christopher, L. V. (2009). Racial and Ethnic Disparities in the US Criminal Justice
System.
Hjalmarsson, R. (2008). Criminal justice involvement and high school completion. Journal of Urban
Economics, 63(2):613–630.
Hoe, T. P. (2022). Does Hospital Crowding Matter? Evidence from Trauma and Orthopedics in
England. American Economic Journal: Economic Policy, 14(2):231–262.
Hummels, D., Munch, J., and Xiang, C.(2016). NoPain, NoGain: WorkDemand, Work Effort, and
Worker Health. Technical Report w22365, National Bureau of Economic Research, Cambridge,
MA.
Iyengar, R. (2007). An Analysis of the Performance of Federal Indigent Defense Counsel. Technical
Report w13187, National Bureau of Economic Research, Cambridge, MA.
59
Janke, K., Propper, C., and Sadun, R. (2019). The Impact of CEOs in the Public Sector: Evi-
dence from the English NHS. Technical Report w25853, National Bureau of Economic Research,
Cambridge, MA.
Justice Policy Institute, The (2011). System Overload: The Costs of Under-Resourcing Public
Defense.
Kahn, M. E. and Li, P. (2020). Air pollution lowers high skill public sector worker productivity in
China. Environmental Research Letters, 15(8):084003.
Kovacs, R. and Lagarde, M. (2022). Does high workload reduce the quality of healthcare? Evidence
from rural Senegal. Journal of Health Economics, 82:102600.
Kuziemko, I. (2013). How should inmates be released from prison? An assessment of parole versus
fixed-sentence regimes *. The Quarterly Journal of Economics, 128(1):371–424.
Lofstrom, M. and Raphael, S. (2016). Crime, the Criminal Justice System, and Socioeconomic
Inequality. Journal of Economic Perspectives, 30(2):103–126.
Lovenheim, M. F. and Owens, E. G. (2014). Does federal financial aid affect college enrollment?
EvidencefromdrugoffendersandtheHigherEducationActof1998. JournalofUrbanEconomics,
81:1–13.
Maibom, J., Sievertsen, H. H., Simonsen, M., and Wüst, M. (2021). Maternity ward crowding,
procedure use, and child health. Journal of Health Economics, 75:102399.
Marchewka, M., Nesterak, J., Sołtysik, M., Szymla, W., and Wojnarowska, M. (2020). Multi-
tasking Effects on Individual Performance: An Experimental Eye-Tracking Study. EUROPEAN
RESEARCH STUDIES JOURNAL, XXIII(Issue 1):107–116.
Marks, M. and Choi, M. K. (2019). Baby Boomlets and Baby Health: Hospital Crowdedness,
Hospital Spending, and Infant Health. American Journal of Health Economics, 5(3):376–406.
Miller, W.R.andRollnick, S.(2009). TenThingsthatMotivationalInterviewingIsNot. Behavioural
and Cognitive Psychotherapy, 37(2):129–140.
Mitchell, O., Wilson, D. B., Eggers, A., and MacKenzie, D. L. (2012). Assessing the effectiveness of
drug courts on recidivism: A meta-analytic review of traditional and non-traditional drug courts.
Journal of Criminal Justice, 40(1):60–71.
Morgenstern, J., Kuerbis, A., Houser, J., Levak, S., Amrhein, P., Shao, S., and McKay, J. R. (2017).
Dismantling motivational interviewing: Effects on initiation of behavior change among problem
drinkers seeking treatment. Psychology of Addictive Behaviors, 31(7):751–762.
Mueller-Smith, M. and T. Schnepel, K. (2021). Diversion in the Criminal Justice System. The
Review of Economic Studies, 88(2):883–936.
60
Muralidharan, K.andSingh, A.(2020). ImprovingPublicSectorManagementatScale? Experimen-
tal Evidence on School Governance in India. Technical report, Research on Improving Systems
of Education (RISE).
Neal, D. and Rick, A. (2014). The Prison Boom and the Lack of Black Progress after Smith and
Welch. Technical Report w20283, National Bureau of Economic Research, Cambridge, MA.
Ostrom, B. J. and Bowman, J. (2020). Examining the Effectiveness of Indigent Defense Team Ser-
vices: A Multisite Evaluation of Holistic Defense in Practice. Justice System Journal, 41(2):139–
184.
Park, K. H. (2017). Do Judges Have Tastes for Discrimination? Evidence from Criminal Courts.
The Review of Economics and Statistics, 99(5):810–823.
Rao, M. (2020). Whither Justice?: Judicial Capacity Constraints Worsens Trial and Litigants’
Outcomes. page 33.
Raphael, S. and Stoll, M. A. (2013). Assessing the Contribution of the Deinstitutionalization of the
Mentally Ill to Growth in the U.S. Incarceration Rate. The Journal of Legal Studies, 42(1):187–
222.
Rehavi, M. M. and Starr, S. B. (2014). Racial Disparity in Federal Criminal Sentences. Journal of
Political Economy, 122(6):1320–1354.
Roach, M. A. (2014). Indigent Defense Counsel, Attorney Quality, and Defendant Outcomes.
American Law and Economics Review, 16(2):577–619.
Rose, E. K. and Shem-Tov, Y. (2021). How Does Incarceration Affect Reoffending? Estimating the
Dose-Response Function. Journal of Political Economy, 129(12):3302–3356.
Seward, J., Vigliotti, V. S., and Cunningham, S. (2021). Social Workers and Suicidality in Jail:
Evidence from Travis County’s Mental Health Court.
Shem-Tov, Y. (2022). Make or Buy? The Provision of Indigent Defense Services in the United
States. The Review of Economics and Statistics, 104(4):819–827.
Shem-Tov, Y., Raphael, S., and Skog, A. (2021). Can Restorative Justice Conferencing Reduce
Recidivism? Evidence From the Make-it-Right Program.
Shulman, E. P., Steinberg, L. D., and Piquero, A. R. (2013). The Age–Crime Curve in Adolescence
and Early Adulthood is Not Due to Age Differences in Economic Status. Journal of Youth and
Adolescence, 42(6):848–860.
Sun, L. and Abraham, S. (2021). Estimating dynamic treatment effects in event studies with
heterogeneous treatment effects. Journal of Econometrics, 225(2):175–199.
61
Woodworth, L.(2020). Swamped: EmergencyDepartmentCrowdingandPatientMortality. Journal
of Health Economics, 70:102279.
Yang, C. S. (2016). Resource Constraints and the Criminal Justice System: Evidence from Judicial
Vacancies. American Economic Journal: Economic Policy, 8(4):289–332.
62

Abstract (if available)

Abstract This dissertation contributes to our understanding of the incentives and constraints faced by public defenders – attorneys for low income defendants – in the United States. Across the three chapters, I study the issues faced by public defenders and their clients from different angles. In Chapter 1, I investigate how the case outcomes of low-income criminal defendants are impacted by the high workloads of public defenders. In Chapter 2, I explore how the attorney’s workload impacts the precise way he/she allocates time across case-related tasks such as client communication, negotiation, and evidence review. Chapter 3 measures how a program which placed social workers in public defense offices impacted clients facing substance abuse and/or mental health charges. The efficacy of these attorneys is an important issue to both economists and policymakers, as both criminal records and prison time have serious downstream consequences for defendants’ labor market and health outcomes.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays in development economics

PDF

Essays on political economy and corruption

PDF

Three essays on the microeconometric analysis of the labor market

PDF

The impact of minimum wage on labor market dynamics in Germany

PDF

Essays on narrative economics, Climate macrofinance and migration

PDF

Essays on development economics and adolescent behavior

PDF

Essays on wellbeing disparities in the United States and their social determinants

PDF

Essays in labor economics: demographic determinants of labor supply

PDF

Essays on econometrics

PDF

Essays on the microeconomic effects of taxation policies

PDF

Essays on environmental economics: education, employment, and experiments

PDF

Essays in climate change adaptation: role of market power in incentivizing adaptation behavior

PDF

The impact of digital transformation on urban communities, welfare, and distributional outcomes

PDF

Essays on development and health economics: social media and education policy

PDF

Sustaining open source software production: an empirical analysis through the lens of microeconomics

PDF

Essays in environmental economics

PDF

Growth and development in Africa: macroeconomic and microeconomic perspectives

PDF

How age, gender, and education impact our path toward carbon-neutrality

PDF

Essays on education and institutions in developing countries

PDF

Essays on effects of reputation and judicial transparency

Asset Metadata

Creator Mahler, Amy (author)

Core Title Essays on applied microeconomics

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Degree Conferral Date 2023-05

Publication Date 05/08/2023

Defense Date 05/07/2023

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics of crime,labor economics,OAI-PMH Harvest,public defense,public economics

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Weaver, Jeffrey (committee chair), Bassi, Vittorio (committee member), Kahn, Matthew (committee member), Nix, Emily (committee member)

Creator Email amahler@usc.edu,amy.mahler@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113102924

Unique identifier UC113102924

Identifier etd-MahlerAmy-11793.pdf (filename)

Legacy Identifier etd-MahlerAmy-11793

Document Type Dissertation

Format theses (aat)

Rights Mahler, Amy

Internet Media Type application/pdf

Type texts

Source 20230508-usctheses-batch-1039 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu