12.57

200

THE HIDDEN COSTS OF PRETRIAL
DETENTION REVISITED

359.08

23.9

005

MARCH 21, 2022

56.10

100

EXECUTIVE SUMMARY

Any time a person is arrested and accused of committing a crime, a
decision has to be made. Will this person be quickly released back into
the community, or will this person be detained in jail to await the next
stage of case processing?
This decision is not a strictly “either-or” proposition. Some people facing charges are incarcerated during this pretrial stage
for months or even years, while other spend only a few hours or days in jail, and some spend no time behind bars at all.
This pretrial detention decision carries enormous consequences for the individual charged, and has serious downstream
effects throughout the entire justice system. Accordingly, the decision whether to subject someone to incarceration awaiting
further court processing is a weighty one and is often informed by a complex set of factors. Concerns about community
safety, the constitutional rights of justice-involved persons, and the need for individuals to appear in court all play an
important role.
The key question at hand is whether pretrial detention promotes future court appearance or public safety. If it does, there
may be a benefit of pretrial detention with respect to public safety. However, it is possible that pretrial detention actually
makes things worse for justice-involved individuals—a finding revealed by a recent meta-analysis of over 100 studies on the
effects of custodial versus community sanctions (Petrich et al., 2021). If pretrial detention actually inflicts harms, then not
only does it compromise public safety, but it also forces the public to bear additional costs—not only financially but also the
human costs on people and their families-of incarcerating citizens unnecessarily.
The present study builds upon prior analyses that showed pretrial detention for a period longer than three days was
associated with higher rates of failing to appear and re-arrest during the pretrial phase. However, this “three-day rule” was
generated with data that had little precision in terms of actual time spent in pretrial detention. The new data used in the
present study contain time-stamped information for more precise estimates, and more rigorous statistical analyses were
applied to assess the potential “generality” of the effects.
To that end, the present study uses data on 1,487,107 individuals booked into a jail in Kentucky between 2009 and 2018 to
address two broad research objectives:
1) to investigate the relationship between pretrial detention and failure to appear, 2) to investigate the relationship
between pretrial detention, and the length of pretrial detention, and new arrest during the pretrial phase, 3) to
investigate the relationship between pretrial detention and sentencing outcomes, and 4) to determine the extent to
which these findings are “general” across sociodemographic categories of people and across different time periods.

This work was supported by Arnold Ventures, a philanthropy dedicated to tackling some of the most pressing problems
in the United States. The authors wish to thank Tara Blair, Daniel Sturtevant, the members of the Arnold Ventures
Pretrial Policy and Research Advisory Board, Stuart Buck, and the State of Kentucky for assistance, participation,
review, and support of this project.

Electronic copy available at: https://ssrn.com/abstract=4063164

Multivariate models were generated that controlled for relevant factors such as the likelihood of FTA or rearrest, supervision
status, offense type, offense level, time at time in the community, demographic characteristics, and other factors. Three
critical findings related to the impact of pretrial detention were revealed.
1. P
 retrial detention and failure to appear (FTA): Pretrial detention—for any amount of time (not just for three days or
longer)—is not consistently associated with the likelihood of failing to appear.
2. P
 retrial detention and rearrest: Pretrial detention—for any length of time (not just for three days or longer)—is associated
with a higher likelihood of a new arrest pending trial.
3. S
 entencing outcomes: Pretrial detention is associated with an increased likelihood of receiving a sentence to jail or prison
and a longer sentence compared to those that were released pretrial. This finding held even when controlling for the
outcomes of pretrial release. That is, those that were rearrested or failed to appear on pretrial released were still less likely to
receive a sentence to incarceration and received a shorter sentence relative to those that were detained pretrial.
4. G
 enerality of effects: Race was not a significant factor in predicting either failure to appear or for rearrest, and the
relationship between pretrial detention and pretrial outcomes (failure to appear and rearrest) did not vary significantly or
consistently by race of the justice-involved person. Furthermore, the effect of pretrial detention on these outcomes was
general across time periods both before and after legal changes were implemented to begin using a new risk assessment
instrument during the pretrial phase.
PROJECT DESCRIPTION
Previous analyses of data drawn from Kentucky between 2009-2010 were conducted in 2013. These original analyses
showed that, overall, there was no significant “deterrent effect” of pretrial detention on outcomes such as failure to appear
or rearrest when the length of pretrial detention exceeded three days.1 Even further, the general pattern of findings—in
particular, the apparent “three day rule”—indicated that, if anything, longer periods of pretrial detention were associated
with worse outcomes: higher rates of failure to appear and rearrest, especially for individuals who were found to be lower
risk based on an assessment.
The present research revisits this question with more fine-grained data on the time spent in pretrial detention, a longer time
window of cases, more detailed data on the length of time spent in pretrial detention, and investigates these questions using
more advanced statistical modeling techniques that are more likely to reveal “causal” relationships.2
DATA AND METHODS
The sample used for the current study includes 1,487,107 individuals who were arrested and booked into a Kentucky jail
between 2009 and 2018. The measures in this study included the following:
•

Individuals’ demographic characteristics

•

Individuals’ likelihood of FTA or rearrest

•

 ffense characteristics including offense level (e.g., felony,
O
misdemeanor) as well as felony offense class (e.g., A, B, C, D)

•

 etails of the pretrial status (released or detained, length of
D
detention in days)

•

 ime in the community for both pretrial and postT
disposition periods

•

 entencing outcomes (sentenced to incarceration and
S
sentence length)

The key outcome variables used in this study include:
•

Failure to appear

•

Rearrest for new offense during pretrial release

The majority of individuals in the sample were white (80%) and male (71%); most were not arrested for a felony offense
(61%); and the majority were subject to pretrial detention for less than 24 hours (56%). With respect to the outcomes of
interest, only a small percentage either failed to appear (17%) or were rearrested (12%).
Multivariate analyses were used to complete these analyses, including logistic regression analysis, regression discontinuity
models3, as well as robust and bias-corrected models. The results are presented in the simplest form and are consistent
across all methods of estimation.

Electronic copy available at: https://ssrn.com/abstract=4063164

RESULTS

Failure to Appear
Table 1 displays the results of a multivariate regression analysis for failure to appear. This model controls statistically the
full range of other factors mentioned earlier that may influence the likelihood of failing to appear. Note that most of the risk
ratios for each of the lengths of pretrial detention are not statistically significant.4 The exception to this trend are the risk
ratios for pretrial detention lengths of 10, 11, and 12 days. Additional models indicated that an increase in failure is noted
when a defendant is detained for four or more days and still other models indicated that longer periods of pretrial detention
were associated with a decreased likelihood of failure to appear. Finally, regression discontinuity models at using hours in
detention revealed that there is no “optimal” time in pretrial detention as related to failure to appear.5 Overall, it appears
then that pretrial detention does not maintain a consistent relationship with failure to appear.
Table 1. Multivariate regression results for the impact of pretrial detention length and failure to appear (FTA).
Variable

Risk Ratios

p-value

Number of Charges

1.037893

≤ 0.001

Felony Charge

0.69097

≤ 0.001

Misdemeanor Charge

1.351901

≤ 0.001

Against Person

0.553353

≤ 0.001

Property

1.382691

≤ 0.001

Time at Risk

1.003432

≤ 0.001

Black

1.215701

≤ 0.001

Male

1.000688

0.945

Hours in Detention
(0/23=0)

Reference

(24/47=1)

1.061

0.05

(48/71=2)

1.060

0.374

(72/95=3)

1.108

0.186

(96/119=4)

1.124

0.055

(120/143=5)

1.137

0.011

(144/167=6)

1.118

0.057

(168/191=7)

1.092

0.191

(192/215=8)

1.114

0.142

(216/239=9)

1.110

0.019

(240/263=10)

1.186

≤ 0.001

(264/287=11)

1.259

≤ 0.001

(288/311=12)

1.185

≤ 0.001

(312/335=13)

1.128

0.028

(336/359=14)

1.044

0.597

(360/383=15)

1.212

0.009

(384/407=16)

1.131

0.119

(408/431=17)

1.165

0.011

(432/455=18)

1.157

0.03

(456/479=19)

1.179

0.077

Constant

0.071

Electronic copy available at: https://ssrn.com/abstract=4063164

Pretrial Rearrest
Table 2 displays the results for the analyses for the relationship between hours spent in pretrial detention and the odds
of being rearrested during the pretrial period. The basis of comparison here is 0-23 hours spent in pretrial detention, and
these coefficients are interpreted as a value of 1.0 is no association at all; a value below 1.0 would indicate that the odds of
being rearrested are lower for those subject to pretrial detention, and a value above 1.0 would indicate that the odds of being
rearrested are higher for those subject to pretrial detention.
Table 2. Multivariate regression results for rearrest by hours spent in pretrial detention.
Risk Ratio

p-value

Number of Charge

Variable

0.975

≤ 0.001

Felony Charge

1.056

0.179

Misdemeanor Charge

1.163

≤ 0.001

Against Person

0.875

0.011

Property

1.270

≤ 0.001

Time at Risk

1.003

≤ 0.001

Black

0.958

0.336

Male

1.188

≤ 0.001

Hours in Detention
(0/23=0)

Reference

(24/47=1)

1.237

≤ 0.001

(48/71=2)

1.289

≤ 0.001

(72/95=3)

1.448

≤ 0.001

(96/119=4)

1.432

≤ 0.001

(120/143=5)

1.459

≤ 0.001

(144/167=6)

1.428

≤ 0.001

(168/191=7)

1.432

≤ 0.001

(192/215=8)

1.484

≤ 0.001

(216/239=9)

1.504

≤ 0.001

(240/263=10)

1.528

≤ 0.001

(264/287=11)

1.581

≤ 0.001

(288/311=12)

1.500

≤ 0.001

(312/335=13)

1.406

≤ 0.001

(336/359=14)

1.407

≤ 0.001

(360/383=15)

1.513

≤ 0.001

(384/407=16)

1.456

≤ 0.001

(408/431=17)

1.518

≤ 0.001

(432/455=18)

1.490

≤ 0.001

(456/479=19)

1.375

≤ 0.001

Constant

0.055

≤ 0.001

Overall, these results show is that any time spent in pretrial detention beyond 23 hours is associated with a consistent and
statistically significant increase in the likelihood of rearrest.6 In particular, as can be seen in Figure 1, the relatively flat line
that runs across the figure indicates that, even as you move from left to right along the line (indicating increasingly more
time spent in pretrial detention), the elevated odds of rearrest remain fairly constant. These results held up even after
controlling for a person’s likelihood of rearrest.

Electronic copy available at: https://ssrn.com/abstract=4063164

Figure 1. Risk Ratio Hours Detained and Subsequent Pretrial Rearrest

Figure 1. Risk Ratio Hours Detained and Subsequent
Pretrial
(Reference is 0-23 hours, All p values < 0.001)
Reference
is Rearrest
0-23 hours

All p values < 0.001

1.800
1.600

,,,,,---------------............._
....

RISK RATIO

1.400
1.200
1.000
0.800
0.600
0.400
0.200

(2
4/
47
(4 =1)
8/
71
=2
(7
)
2/
95
=
(9
6/ 3)
11
(12 9=
0/ 4)
14
(14 3=
4/ 5)
16
(16 7=
8/ 6)
1
(19 91=
7)
2/
21
5
(2
16 =8
)
/2
(2
39
40
=9
/2
)
63
(2
=
64
1
/ 2 0)
(2 87=
88
1
/3 1)
11
(3
=1
12
2)
/3
35
(3
=
36
/3 13)
5
(3
60 9=1
/ 3 4)
8
(3
84 3=1
/4 5)
(4 07=
08
1
/4 6)
3
(4
32 1=1
7)
/4
(4 55=
56
1
/4 8)
79
=1
9)

0.000

Sentencing Outcomes
Multiple analyses were conducted investigating the likelihood of receiving a sentence to incarceration and the length of
that sentence. Models were constructed that examined the relationship between pretrial detention and release and the
likelihood of being sentenced to jail or prison, and the length of those sentences. The results of these analyses indicated
that those released pretrial were about one-half to three-quarters as likely to receive a sentence to prison or jail compared
to detained counterparts. Differences of these magnitudes persisted even for those that were released pretrial and were
rearrested or failed to appear. Further, when those released pretrial were sentenced to incarceration, they were sentenced
to shorter periods of incarceration than were those that were detained. Again, these shorter sentences were observed even
when those released were rearrested or failed to appear.

Pre/Post PSA Implementation
In 2013 Kentucky began using the Public Safety Assessment (or PSA) in all 120 counties. The PSA implemented a researchbased and validated assessment tool to assess the risk of pretrial failure (both failure to appear and rearrest).
The analyses presented in Table 4 display the results for failure to appear (FTA) and rearrest by pre- and post PSA eras. The
results show that the rates of those individuals detained and released were roughly similar across the two time periods.
They also show that rates of FTA rose slightly in the PSA era, as did new rearrests. In addition to the bivariate models
presented below, several multivariate models were also run investigating the impact of the PSA implementation on failure
to appear, rearrest, and release rates. Overall, these multivariate models indicate that the FTA rate has been trending in
an upward direction for some time and that there is not necessarily an increase in FTA rates associated with the date of
implementation for the PSA.
Further, models investigating the relationship between the PSA implementation and rearrest on pretrial yielded contradictory
results. Taken altogether, it is not likely that the implementation of the PSA had much impact on release, FTA, or pretrial
rearrest rates. Finally, and of importance, additional analyses disaggregated by year indicate that the effect of pretrial
detention on both of these outcomes (FTA and rearrest) remained constant across both the pre-PSA and PSA time periods.

Electronic copy available at: https://ssrn.com/abstract=4063164

Table 4. Bivariate results of pre-PSA and PSA eras for failure to appear (FTA) and rearrest 2010-2018.
N
Pre PSA

1

PSA

Percent

Detained

Released

Total

Detained

Released

182433

531015

713448

0.256

0.744

198609

578489

777098

0.256

0.744

N

Percent

No FTA

FTA

Total

No FTA

FTA

Pre PSA

460082

70933

531015

sssss0.866

0.134

PSA

463067

115272

578339

0.801

0.199

2

N

Percent

No RA

RA

Total

No RA

RA

Pre PSA

475651

55364

531015

0.896

0.104

PSA

500837

77502

578339

0.866

0.134

3

1 = 2 = 0.032; p = 0.858
2 = 2 = 8564.410; p < 0.001
3 = 2 = 2323.651; p < 0.001

DISCUSSION AND CONCLUSIONS
The purpose of the present study was to extend previous analyses of the effect of pretrial detention on pretrial and
sentencing outcomes using data from Kentucky. To do so, analyses were conducted with additional data and more rigorous
quantitative methods.
The analyses revealed several key findings:
1) increasing the amount of time spent in pretrial detention was not consistently related to the odds of failing to appear
in court; 2) increasing the amount of time spent in pretrial detention was consistently associated with an increased odds
of rearrest, 3) those released pretrial had a lower likelihood of receiving a sentence to incarceration and when sentenced
to incarceration received a shorter sentence, and 4) race was not a significant factor in predicting either failure to
appear or for rearrest, and the relationship between pretrial detention and pretrial outcomes ( failure to appear and
rearrest) did not vary significantly or consistently by race of the individual or by the time period under investigation.
These results are largely consistent with those found in previous analyses of data from Kentucky, where no “deterrent
effect” of pretrial detention was observed on pretrial outcomes. In addition, that no deterrent effects were revealed is also
consistent with decades of research7 on the effects of custodial sanctions (e.g., incarceration in either jail or prison) on
outcomes like recidivism. In fact, the current analyses show that, at least with respect to rearrest during the pretrial period,
longer stints in pretrial detention actually did more harm than good in terms of rearrest rates.
The key takeaway from these analyses is that incarcerating people prior to their trial does not result in better pretrial
outcomes in terms of failure to appear or rearrest. Indeed, there is no observable “deterrent effect” of pretrial detention,
and in fact there is a consistent “criminogenic effect” of pretrial detention on rearrest. This means that the costly option of
incarcerating defendants prior to trial is not being translated into a public benefit of an increase in public safety.
It is equally important to note that there is no magic amount of time spent in pretrial detention that will result in a
consistent public benefit (i.e, the “three day rule” can be safely abandoned)—the reality is that getting people out of jail
sooner rather than later is better.

Electronic copy available at: https://ssrn.com/abstract=4063164

These analyses also have important implications and recommendations moving forward. For example:
1. J
 ail time should generally be avoided during the pretrial phase. This is not to say that certain justice-involved persons cannot
ever pose a threat to public safety should they be released, but rather that, in most instances, jail is likely the most harmful
option during the pretrial stage. Thus – and recognizing that the majority of individuals are successful pretrial8 -- as a general
rule, jail should not be the default choice.
2. J
 udges need to be informed with respect to the present results concerning the consequences of pretrial detention. While the
potential for a “deterrent effect” of incarceration is enticing (i.e., that depriving someone of their liberty prevents someone
from engaging in crime after release, or through example prevents others from engaging in crime)—and one that judges
often assume will occur—the evidence suggests this estimated effect does not exist. At this point, we can say the evidence is
clear that pretrial detention is likely more harmful than beneficial.9
3. I t is important to offer resources to justice-involved individuals during the pretrial phase. Research has consistently
demonstrated that a service-delivery approach to criminal justice—one that emphasizes treatment and support—is far more
effective than one based on punishment.

REFERENCES
Apel, R. (2013). Sanctions, perceptions, and crime: Implications for criminal deterrence. Journal of Quantitative Criminology,
29, 67-101.
Gibbs, J.P. (1968). Crime, punishment, and deterrence. The Southwestern Social Science Quarterly, 48, 515-530.
Jonson, C.L. (2010). The impact of imprisonment on offending: A meta-analysis. Doctoral dissertation, University of Cincinnati.
Mears, D.P., & Cochran, J.C. (2018). Progressively tougher sanctioning and recidivism: Assessing the effects of different types
of sanctions. Journal of Research in Crime and Delinquency, 55, 194-241.
Nagin, D.S., Cullen, F.T., & Jonson, C.L. (2009). Imprisonment and reoffending. Crime and Justice: A Review of Research,
38, 115-200.
Petrich, D.M., Pratt, T.C., Jonson, C.L., & Cullen, F.T. (2021). Custodial sanctions and reoffending: A meta-analytic review.
Crime and Justice: A Review of Research, in press.
Pratt, T.C. (2019). Addicted to incarceration: Corrections policy and the politics of misinformation in the United States.
2nd edition. Thousand Oaks, CA: Sage.
Smith, P., Goggin, C., & Gendreau, P. (2002). The effects of prison sentences and intermediate sanctions on recidivism:
General effects and individual differences. Ottowa, ON: Solicitor General of Canada.
Villatez, P., Gillieron, G., & Killias, M. (2015). The effects on re-offending of custodial vs. non-custodial sanctions:
An updated systematic review of the state of the knowledge. Oslo, Norway: The Campbell Collaboration.
Weisburd, D., Farrington, D.P., Gill, C., et al. (2017). What works in crime prevention and rehabilitation:
An assessment of systematic reviews. Criminology and Public Policy, 16, 415-449.

Electronic copy available at: https://ssrn.com/abstract=4063164

ENDNOTES
1

S
 ee “The Hidden Costs of Pretrial Detention” (2013, LJAF, Christopher T. Lowenkamp, Marie VanNostrand, and
Alexander Holsinger); https://nicic.gov/hidden-costs-pretrial-detention

2

I t is notoriously difficult to establish causality in non-experimental research designs. It is therefore critically
important to control for other factors that could be related to the outcome of interest and to accommodate them into a
multivariate modeling strategy, which is the approach taken in the present research.

3

T
 he discontinuity in the regression models is the number of hours served in jail during pretrial detention.
Discontinuities were tested for at 24, 48, 72, and 96 hours of pretrial detention.

4

T
 o say that a result is “statistically significant” means that the results that were obtained are unlikely to be the result of
sampling error. That is, there can be some random fluctuations in the numbers—where they may not be exactly zero—
but we are using a statistical approach that determines whether those small deviations from zero might simple be due
to chance.

5

T
 hese results remained similar across different subsamples of the data (i.e., across different offense and individual
characteristics) as well as for different time periods “at risk.”

6

T
 hese results remained unchanged across different multivariate estimation techniques (e.g., logistic regression and
regression discontinuity analysis).

7

Several

meta-analyses of the correctional treatment/sanctions literature have been conducted over the last two decades
(see, e.g., Jonson, 2010; Petrich et al., 2021; Smith et al., 2002; Villatez et al., 2015), all of which reach the same general
conclusion: that custodial sanctions (like jail relative to remaining in the community) do not consistently reduce
recidivism.

8

I n a review of Public Safety Assessment (PSA) validation research, base rates for pretrial failure, and specifically for
experiencing an arrest pretrial for a violent charge range from 1% to nearly 10%. In Kentucky, the base rate for a new
violent arrest pretrial was 1.1% (DeMichele et al., 2018), 4% in Lucas County, Ohio (Lowenkamp et al., 2020), and 9.8% in
Los Angeles County, California (Hess & Turner, 2021). Other research (not PSA validation studies) found similarly low
rates of arrests for violent charges during pretrial. This includes Cook County, Illinois, which reported a 3% base rate
(Stemen & Olson, 2020) and Washington, D.C. with a 1% base rate for pretrial arrests for violence for supervision clients
(Pretrial Services Agency for the District of Columbia, 2020).

9

T
 here is a large body of evidence that stiffening criminal penalties in general do not provide any consistent deterrent
effects (Mears & Cochran, 2018; Nagin, Cullen, & Jonson, 2009). Even further, more recent evidence in the form of a
review of over 100 studies suggests that custodial sanctions (e.g., putting someone in jail rather than keeping them in
the community) provides no benefits in terms of crime control (Petrich et al., 2021).

0

Arnold Ventures

For more information: ArnoldVentures.org

Electronic copy available at: https://ssrn.com/abstract=4063164