Incarceration, Recidivism, and Optimal Sentencing

1
Incarceration, Recidivism, and Optimal Sentencing

• December 2008

ECONorthwest 888 SW Fifth Ave., Ste.
1460 Portland, OR 97204 503-222-6060 www.econw.com
2
Overview

• The policy context for optimal sentencing
• Survey of the literature on length of
  incarceration and recidivism
• Summary of findings from Oregons DOC
  prison-release data
• Conclusions

3
The policy context for optimal sentencing

• As established in administrative rule, Oregons
  sentencing policy seeks to impose appropriate
  punishment and insure public safety.
• The first of five basic sentencing principles
  listed in the rules is that
• The response of the corrections system to crime
  must reflect the resources available for that
  response. --OAR 213-002-0001
• Maximizing the societal benefits given the
  resources available requires understanding the
  magnitude of both the cost of sanctions and of
  the resulting benefits.

4
An economic framework for optimal sentencing
policy

• Cost-benefit analysis provides a framework to
  guide policy that seeks to maximizes benefits
  given limited resourcesone possible definition
  for optimal sentencing policy.
• Estimating costs (i.e., the costs of
  incarceration and other criminal sanctions) is
  largely a budgeting and accounting exercise.
• Analyzing the determinants of recidivism helps to
  estimate the benefits of alternative sentencing
  practices and address on of the question
  motivating this project What works to prevent
  crime?

5
What does the theory suggest about the benefits
of incarceration?

• Impacts of incarceration on crime
• General deterrence
• Incapacitation
• Specific deterrence (recidivism)
• Impacts of length of incarceration on recidivism
• Rehabilitation prison creates better people
• Aging out of crime
• Social bonding effects prison creates worse
  people

6
One perspective on optimal sentencing

• A common assumption is that the impact of
  sentence length on recidivism exhibits
  diminishing returns.
• Maximizing public safety suggests a sentence
  length corresponding to B. A cost-benefit
  analysis could suggest alternative optimal
  sentences.
• Key empirical questions
• What does the length-recidivism curve look like?
• Does it vary by offense? By individual?

7
What do the DOC data say?

• Definitions matter

Source ECONorthwest analysis of Oregon DOC data.
8
What do the DOC data say?

• Characteristics help to predict recidivism

Source ECONorthwest analysis of Oregon DOC data.
9
What do the DOC data say?

• Time served helps to predict recidivism

Source ECONorthwest analysis of Oregon DOC data.
10
Multiple risk factors require multivariate
analysis

• Time in prison helps predict recidivism, but
  offender characteristics also predict time in
  prison. Multivariate analysis is required to
  disentangle the effect of time in prison from
  those of other risk factors.
• The fundamental problem Individuals who appear
  similar but serve different lengths of time may
  differ in ways not captured by available data.
• Randomized controlled trials are the gold
  standard for solving the fundamental problem.
  Implementing RCTs is often problematic in a
  criminal justice context.

11
What does the literature say?

• Broadly speaking, two related strands of research
  address the question of optimal sentencing
• Analyses of incarceration versus a
  community-based sentence
• Analyses of sentence length, time served, and
  recidivism (our focus)
• The results on length of incarceration are
  difficult to synthesize because of differences in
  definitions, data, and methodology. In recent
  years, better data and better methodology have
  refined results, but have not provided
  significantly better answers.
• Common findings are that age, race, gender, and
  prior criminal history have statistically
  significant impacts on recidivism, but the
  estimated impacts vary greatly across studies.
• Most analyses ultimately conclude that the impact
  of incarceration depends critically on inmate
  characteristics.

12
Gendreau, Goggin, and Cullen (2002)

• The authors conducted a comprehensive literature
  review and meta-analysis of hundreds of studies
  on the impacts of sanction type and incarceration
  length on recidivism.
• Key findings
• Tentative indications that increasing sentence
  length slightly increases recidivism
• No evidence that impacts vary by gender, minority
  status, or risk level
• Only limited evidence that higher-quality study
  designs produce better results
• Many studies lacked critical information about
  methods or omitted important variables from the
  analysis
• Other reviews reach similarly ambiguous
  conclusions. For example, Song and Lieb (1993)
  concluded that the impact of time served on
  recidivism may be offender specific and
  influenced by characteristics such as age,
  offense type, and criminal history.

13
Orsagh and Chen (1988)

• The researchers sought to identify the sentence
  length that minimizes recidivism (i.e., the
  optimal sentence length) using a sample of North
  Carolina prison releases in 1980.
• Recidivism was defined as re-arrest within two
  years of release.
• The innovation looking for a U-shaped
  relationship between time served and recidivism.
• Key findings The optimal length of time served
  was 1.2 years on average, but the optimum varied
  significantly by offender characteristics such as
  age and type of conviction.

14
Kuziemko (2007)

• Kuziemko investigated the impacts of limiting the
  discretion of parole boards on the social costs
  of crime. The analysis included estimation of the
  impact of time served on recidivism.
• Recidivism was defined as a return to prison
  within three years of release.
• The innovation Kuziemko used a natural
  experiment a mass-release of prisoners in
  Georgia during 1981 to isolate the impact of
  time served.
• Key findings
• Age, race, and criminal history had significant
  impacts on recidivism.
• The author concluded that increasing time in
  prison reduced recidivism, but at a diminishing
  rate as time in prison increased.
• Results from less rigorous alternative analyses
  did not conflict with results from the mass
  release analysis.

15
Bierens and Carvalho (2007)

• The researchers conducted a sophisticated,
  multi-state analysis of recidivism using Bureau
  of Justice Statistics data.
• Recidivism was defined as felony or misdemeanor
  re-arrest within up to five years (felony and
  misdemeanor were considered separately).
• The innovation constructed state-by-state
  estimates for the impact of time served on
  recidivism.
• Key findings Age, race, gender, and sentence
  length all impact recidivism, but effects vary
  greatly across states. In Oregon, increased
  sentence length corresponds with increased time
  to recidivate (i.e., lower rates of recidivism
  during a given period post-release).

16
Multivariate analysis of Oregon DOC data

• The literature consistently identifies a handful
  of factors that predict recidivism, but remains
  unclear on the magnitude of the impacts.
  Well-designed local studies may provide the best
  guidance for decision makers.
• Our analysis included nearly 65,000 prison
  releases for inmates convicted of one or more
  felonies dating back to 1990.
• Variables of primary interest
• Offender demographics age, race, gender,
  criminal history, type of crime, others
• Incarceration characteristics length of stay and
  earned time
• Type of recidivism reconviction of a felony
  versus re-entry into prison for non-felony
  offenses
• Numerous model specifications all confirm the
  same fundamental results, but the DOC data were
  not sufficient to answer counterfactuals What
  would happen if an individual had spent more or
  less time in prison?

17
Findings Offender characteristics predict
recidivism

• Understanding these and other risk-factors can
  aid sentencing decisions and improve the
  allocation of prison resources.
• Estimated impacts do not necessarily indicate
  causation.
• Other variables had significant impacts on
  recidivism risk, including crime seriousness,
  criminal history classification, and county of
  adjudication.

Source ECONorthwest analysis of Oregon DOC data.
18
Findings Length of stay predicts recidivism

• Recidivism risk increases with time served for
  relatively short incarcerations
• The reduction in recidivism risk peaks for
  incarcerations of about six years.
• The empirical length-of-stay/ recidivism curve
  could be used to identify optimal sentence
  lengths
• but the results only identify correlations. They
  do not necessarily imply the effect of changing
  sentence lengths for an individual offender or
  group of offenders.

Source ECONorthwest analysis of Oregon DOC data.
19
Conclusions

• Our analysis of DOC data suggest the impacts of
  offender characteristics and length of
  incarceration on recidivism but do not prove
  causation.
• Analyses that incorporate data from additional
  sources (e.g., arrest data, judicial data) would
  improve the analysis and provide better, more
  precise input to cost-benefit models.
• More data could provide improved recidivism
  definitions, better controls for offender
  characteristics, and could allow analysis using
  natural experiments (e.g., arising from
  systematic differences across counties in
  sentencing or charging).
• Known predictors of future criminal activity
  provide valuable, actionable information in the
  absence of better information about the impact of
  incarceration length. This information would
  allow sentencing decisions that better allocate
  scarce prison resources regardless of the
  incarceration length.