Reviews of Deterrence Research
Since 1978, numerous detailed reviews have utilized deterrence theory to analyze the relationship between criminal sanctions and offending behavior These reviews encompass a wide range of empirical research, exploring various hypotheses about the impact of contemporary criminal sanctions on subsequent criminal activities Although no single review encompasses all elements of deterrence theory, they contribute to a growing interest in understanding its complexities and nuances in the context of crime prevention.
26 Johannes Andenaes, Does Punishment Deter Crime?, 11 C RIM L.Q 76, 78–79 (1968)
27 Mark C Stafford & Mark Warr, A Reconceptualization of General and Specific
28 See Robert Apel, Sanctions, Perceptions, and Crime: Implications for Criminal Deterrence, 29 J Q UANTITATIVE C RIMINOLOGY 67, 93 (2013); Robert Apel & Daniel S Nagin,
General Deterrence, in T HE O XFORD H ANDBOOK OF C RIME AND C RIMINAL J USTICE 179, 179–
206 (Michael Tonry, ed., 2011); Aaron Chalfin & Justin McCrary, Criminal Deterrence: A
Review of the Literature, 55 J E CON L ITERATURE 5, 5–6 (2017); Philip J Cook, Research in
Criminal Deterrence: Laying the Groundwork for the Second Decade, in 2 C RIME & J UST
211, 211–268 (1980); Steven N Durlauf & Daniel S Nagin, Imprisonment and Crime: Can
Both Be Reduced?, 10 C RIMINOLOGY & P UB P OL ’ Y 13, 13–54 (2011); Daniel S Nagin,
Criminal Deterrence Research at the Outset of the Twenty-First Century, 23 C RIME & J UST
1, 1–42 (1998) [hereinafter Nagin, Criminal Deterrence Research]; Daniel S Nagin, General
Deterrence: A Review of the Empirical Evidence, in D ETERRENCE AND I NCAPACITATION :
Estimating the effects of criminal sanctions on crime rates has been extensively reviewed in the literature, highlighting the relationship between deterrence and criminal behavior Nagin, Solow, and Lum emphasize that effective deterrence relies on the interplay between criminal opportunities and police presence Their findings suggest that enhancing law enforcement strategies can significantly impact crime rates by reducing opportunities for criminal activities.
C RIMINOLOGY 74, 74–100 (2015); Raymond Paternoster, The Deterrent Effect of the Perceived
Certainty and Severity of Punishment: A Review of the Evidence and Issues, 4 J UST Q 173 (1987); Raymond Paternoster, How Much Do We Really Know About Criminal Deterrence?,
100 J C RIM L & C RIMINOLOGY 765, (2010) [hereinafter Paternoster, How Much Do We
Really Know]; Travis C Pratt & Francis T Cullen, Assessing Macro-Level Predictors and Theories of Crime: A Meta-Analysis, 32 C RIME & J UST 373, 415–17, 427–28 (2005); Kirk R Williams & Richard Hawkins, Perceptual Research on General Deterrence: A Critical
Research on deterrent effects reveals that identifying and quantifying these effects is challenging Consistent themes in various reviews highlight the difficulties in detecting deterrence, as findings often show inconsistency and are hindered by methodological and measurement limitations.
Research on general deterrence has historically utilized official records of sanctions to analyze the relationship between arrests, police presence, prison populations, and executions against various crime types reported by the FBI This macro-level approach aggregates data at city, county, or state levels to compare jurisdictions over time However, challenges such as measurement errors and disputes over identification criteria have prompted researchers to explore the effects of sanctions in specific locations, focusing on treatment versus control groups, individual variations, and longitudinal studies In the 1970s, Franklin Zimring advocated for targeted policy interventions, including clinical trials, longitudinal surveys, and various analytical methods, to effectively test deterrence theory, leading to a significant shift towards evaluating policy interventions.
31 Nagin, General Deterrence: A Review of the Empirical Evidence, supra note 28, at 99–
32 See, e.g., Lan Shi, The Limit of Oversight in Policing: Evidence from the 2001 Cincinnati Riot, 93 J P UB E CON 99 (2009)
33 See, e.g., Lawrence W Sherman & David Weisburd, General Deterrent Effects of Police Patrol in Crime “Hot Spots”: A Randomized, Controlled Trial, 12 J UST Q 625 (1995)
34 See, e.g., John D Wooldredge & Amy Thistlethwaite, Reconsidering Domestic Violence Recidivism: Conditioned Effects of Legal Controls by Individual and Aggregate Levels of Stake in Conformity, 18 J Q UANTITATIVE C RIMINOLOGY 45 (2002)
35 See, e.g., Kirk R Williams & Richard Hawkins, Wife Assault, Costs of Arrest, and the Deterrence Process, 29 J R SCH & C RIME D ELINQ 292 (1992)
36 Franklin E Zimring, Policy Experiments in General Deterrence: 1970–75, in
D ETERRENCE AND I NCAPACITATION : E STIMATING THE E FFECTS OF C RIMINAL S ANCTIONS ON
Crime rates play a significant role in the contemporary literature on deterrence theory, with macro-level analyses primarily focusing on general deterrence In contrast, targeted policy interventions can evaluate either general or specific deterrence, depending on whether the relationship between sanctions and subsequent offenses is examined at the individual or aggregate level However, these interventions are often confined to one or two jurisdictions, limiting the generalizability of their findings without further replication in diverse locations Despite various methodological challenges in studying these interventions, Daniel Nagin emphasizes that well-conducted experimental and quasi-experimental studies offer the most compelling evidence regarding the effectiveness of deterrence in different contexts.
Research on deterrence includes both macro-level studies and evaluations of specific policy interventions, focusing not only on official records of punishment but also on how potential offenders perceive criminal sanctions While macro-level comparisons cannot accurately assess the individual deterrent effects of official sanctions, evaluations based solely on official records fail to capture how offenders make decisions However, these evaluations can reveal whether the relationship between sanctions and behavior aligns with deterrence theory predictions There is a consensus on the importance of understanding the connection between official sanctioning and offenders' perceptions, yet researchers differ significantly on how changes in the likelihood or severity of sanctions influence these perceptions.
37 Steven N Durlauf & Daniel S Nagin, The Deterrent Effects of Imprisonment, 43, 48 in
C ONTROLLING C RIME : S TRATEGIES AND T RADEOFFS (Philip J Cook, Jens Ludwig & Justin McCrary eds., 2011); Nagin, Criminal Deterrence Research, supra note 28, at 4
38 William Alex Pridemore, Matthew C Makel & Jonathan A Plucker, Replication in
Criminology and the Social Sciences, 1 A NN R EV C RIMINOLOGY 19, 21–24 (2018)
40 Paternoster, How Much Do We Really Know, supra note 28, at 780–86
Research indicates that understanding how police tactics influence offenders' perceptions of apprehension risk is crucial While traditional deterrence theory emphasizes official sanctions, it is essential to recognize that potential offenders also consider various factors beyond these sanctions when making decisions about criminal behavior Individual personality traits, emotions, and irrational considerations play significant roles, particularly in cases of intimate partner violence Incorporating insights from behavioral economics and rational choice theory can enhance our understanding of these decision-making processes Despite this, existing studies on the impact of prosecution and imprisonment on intimate partner violence have not gathered relevant information directly from potential offenders, limiting our knowledge of how post-arrest sanctions affect their behavior.
Limitations in Deterrence Reviews
Current reviews of deterrence research exhibit two major limitations They predominantly emphasize arrest and imprisonment while neglecting the potential effects of intermediate sanctions According to Steven Durlauf and Daniel Nagin, future evaluations of deterrence theory should consider that the effectiveness of deterrent measures will significantly hinge on the specific type of sanction policy under examination.
P UB P OL ’ Y 727, 729 (2016) (“All prior studies examining the correlation between objective and perceived arrest risk have yielded null results.”)
42 See Thomas A Loughran, Raymond Paternoster & Alex R Piquero, Individual Difference and Deterrence, in D ETERRENCE , C HOICE , AND C RIME 211, 215–19 (Daniel S Nagin, Francis T Cullen & Cheryl Lero Jonson eds., 2018)
43 Justin T Pickett & Shawn D Bushway, Dispositional Sources of Sanction Perceptions:
Emotionality, Cognitive Style, Intolerance of Ambiguity, and Self-Efficacy, 39 L AW & H UM
44 Jean-Louis van Gelder & Reinout E de Vries, Rational Misbehavior? Evaluating an
Integrated Dual-Process Model of Criminal Decision Making, 30 J Q UANTITATIVE
45 See generally D ANIEL K AHNEMAN , T HINKING , F AST AND S LOW (2011) (demonstrating dual-track decisionmaking processes in a wide range of circumstances)
Just as medications are assessed for their effectiveness in treating specific health issues, future research should focus on evaluating how different types of sanctions impact various crimes Additionally, current deterrence research reviews overlook the application of criminal sanctions for violent offenses against intimate partners, failing to justify the exclusion of this significant area of study.
This article aims to overcome existing limitations by offering a concise qualitative review of numerous studies on the specific deterrent effects of arrest in intimate partner violence cases Additionally, it includes three systematic quantitative meta-analyses that examine the deterrent impacts of prosecuting, convicting, and incarcerating offenders of intimate partner violence.
Deterrent Effects of Arrest
Research on the deterrent effects of arrest for intimate partner violence has been significantly influenced by a pivotal field experiment conducted in Minneapolis This study randomly assigned offenders to one of three groups: arrest, on-scene counseling, or physical separation Findings revealed that arrested offenders had a re-offending rate approximately half that of those not arrested The Minneapolis experiment garnered extensive media attention, leading to the adoption of arrest policies for intimate partner violence Its impact prompted five additional experiments aimed at replicating its findings, although results varied based on the data source, measures of re-offending, and jurisdiction.
46 Durlauf & Nagin, supra note 37, at 85–86
47 Sherman & Berk, supra note 20, at 267–68
48 E VE S B UZAWA & C ARL G B UZAWA , D OMESTIC V IOLENCE : T HE C RIMINAL J USTICE
49 A NTHONY P ATE , E DWIN E H AMILTON & S AMPSON A NNAN , M ETRO -D ADE S POUSE
The final report of the A BUSE R EPLICATION P ROJECT (1991) highlights significant deterrent effects on crime, as evidenced by both official records and victim interviews; however, only the findings from victim interviews reached statistical significance The study, conducted by Richard A Berk, Alec Campbell, Ruth Klap, and Bruce Western, underscores the importance of victim perspectives in understanding the impact of abuse deterrence.
A Bayesian Analysis of the Colorado Springs Spouse Abuse Experiment, 83 J C RIM L &
While none of the five studies replicated the strong deterrent effects observed in the Minneapolis study, a systematic assessment indicates some support for the deterrent effects of arrest A qualitative review highlighted that three of the studies leaned towards deterrent effects, whereas the other three suggested escalation effects, with arrest impacting different offender types variably An extensive summary of findings from the Minneapolis experiment and its replications identified thirty-five common tests of specific deterrence, revealing that 71% indicated deterrent effects, while 29% suggested escalation effects Notably, eight of the twenty-five deterrent effects were statistically significant, whereas none of the escalation effects reached statistical significance.
A meta-analysis of six arrest experiments, focusing solely on victim interview data, revealed a significant deterrent effect of arrest Additionally, an analysis of individual-level archived data from five replications, encompassing 4,032 incidents involving adult male suspects, supports these findings.
A study published in Criminology highlighted the deterrent effects of arrest in domestic assault cases, drawing on both official records and victim interviews; however, the findings were not statistically significant Researchers Franklyn W Dunford, David Huizinga, and Delbert S Elliot contributed to this examination of the impact of arrest on domestic violence incidents in Omaha.
A study published in 1990 in the journal Criminology examined the deterrent effects of victim interviews and the escalation effects from official records, though neither finding was statistically significant Researchers J David Hirschel, Ira W Hutchinson III, and Charles W Dean highlighted the limitations of these methods in understanding crime dynamics.
The study "Arrest to Deter Spouse Abuse" published in the Journal of Research in Crime and Delinquency highlights the potential deterrent effects of arrest on domestic violence, based on victim interviews and official records However, it notes that these effects were not statistically significant The research was conducted by Lawrence W Sherman and colleagues, emphasizing the complexity of addressing spouse abuse through legal interventions.
The Milwaukee Domestic Violence Experiment, published in the Journal of Criminal Law & Criminology, examines the varied impacts of arrest on criminal careers, revealing that while there were observed escalation effects in both official records and victim interviews, these findings lacked statistical significance.
50 L AWRENCE W S HERMAN , J ANELLE D S CHMDIT & D ENNIS P R OGAN , P OLICING
D OMESTIC V IOLENCE : E XPERIMENTS AND D ILEMMAS 16–18 (1992)
51 Franklyn W Dunford, System-Initiated Warrants for Suspects of Misdemeanor
Domestic Assault: A Pilot Study, 7 J UST Q 631, 641–50 (1990) (reporting consistent deterrent effects for arrest warrants)
52 Joel Garner, Jeffrey Fagan & Christopher Maxwell, Published Findings from the
Spouse Assault Replication Program: A Critical Review, 11 J Q UANTITATIVE C RIMINOLOGY
53 David B Sugarman & Sue Boney-McCoy, Research Synthesis in Family Violence: The
Art of Reviewing the Research, 4 J A GGRESSION , M ALTREATMENT & T RAUMA , 55, 66–69
Based on victim interviews, arrests led to a significant 25% reduction in the prevalence of new victimizations, with a notable 30% decrease in the frequency of these incidents However, official records indicated only an 8% reduction in the frequency of new victimizations following arrests, though this finding was not statistically significant.
Three research teams have analyzed non-experimental data from the National Crime Victimization Survey (NCVS) to investigate the specific deterrent effect of arrest Richard Felson and his colleagues examined 2,565 incidents from 1992 to 2002, discovering a correlation between arrest and a reduction in subsequent violence, although it was not statistically significant In a separate study, Hyunkag Cho and Dina Wilke expanded the sample size to 3,495 incidents from the NCVS.
From 1987 to 2003, research indicated a significant deterrent effect of arresting intimate partner violence offenders A study by Min Xie and James Lynch, analyzing data from 1,336 victims in the National Crime Victimization Survey (NCVS) between 1996 and 2012, also found a notable deterrent effect However, in a matched subsample of 688 victims, the results suggested a non-significant effect leaning towards deterrence.
A recent non-experimental study analyzing official police reports from 5,466 couples in Seattle revealed that arrests significantly reduced both the prevalence and frequency of future physical abuse incidents Additionally, a national-level panel study examined the effects of arrest on intimate partner violence and found a significant negative correlation between individuals' perceptions of arrest costs in 1986 and the prevalence of wife assault in 1987 Although these nonexperimental studies typically involve smaller sample sizes and less rigorous methodologies compared to SARP experiments, they provide valuable insights into the impact of arrest on domestic violence.
54 Christopher D Maxwell, Joel H Garner & Jeffrey A Fagan, The Preventive Effects of
Arrest on Intimate Partner Violence: Research, Policy and Theory, 2 C RIMINOLOGY & P UB
55 Richard B Felson, Jeffrey M Ackerman & Catherine A Gallagher, Police Intervention and the Repeat of Domestic Assault, 43 C RIMINOLOGY 563, 576–78 (2005)
56 Hyunkag Cho & Dina J Wilke, Does Police Intervention in Intimate Partner Violence
Work? Estimating the Impact of Batterer Arrest in Reducing Revictimization, 11 A DVANCES
57 Min Xie & James P Lynch, The Effects of Arrest, Reporting to the Police, and Victim
Services on Intimate Partner Violence, 54 J R SCH C RIME & D ELINQ 338, 353–55 (2017)
58 Vivian H Lyons, Mary A Kernic, Ali Rowhani-Rahbar, Victoria L Holt & Marco Carone, Use of Multiple Failure Models in Injury Epidemiology: A Case Study of Arrest and
Intimate Partner Violence Recidivism in Seattle, WA, 6 I NJ E PIDEMIOLOGY 1, 5–6 (2019)
Research consistently indicates that arrests lead to at least minor decreases in repeat offending However, previous reviews of deterrence research have not incorporated findings from either experimental or non-experimental studies on the deterrent effects of arrest.
Research on the impact of arrest for intimate partner violence is extensive and methodologically robust, revealing various conditions under which specific deterrent effects can be observed While findings on the deterrent effects of arrest are mixed, they indicate that intimate partner violence offenders may respond to harsher criminal sanctions, including prosecution, conviction, and incarceration, though this response remains uncertain.
Deterrent Effects of Post-Arrest Sanctions
DESIGN OF THIS RESEARCH
This research examines whether the characteristics of study designs influence the relationship between sanctions and subsequent behaviors Our main findings on deterrence effects are derived from the first question, while the insights gained from the second question aim to guide the design of future studies on the application of sanctions in cases of intimate partner violence.
Meta-Analytic Methods
Since its emergence as a distinct field in the statistical sciences during the 1980s, meta-analysis has evolved into a comprehensive method for integrating various aspects of individual studies within a unified research framework It is now extensively utilized in criminology to synthesize existing research on public policies and scientific hypotheses Meta-analytic techniques generate standardized effect sizes for reported findings, effectively summarizing results both within individual studies and across multiple studies.
76 F RANK L S CHMIDT & J OHN E H UNTER , M ETHODS OF M ETA -A NALYSIS : C ORRECTING
E RROR AND B IAS IN R ESEARCH F INDINGS 453–57 (3d ed 2015)
Meta-analysis in criminology serves two primary purposes: evaluating the empirical support for specific theories, policies, or programs, and assessing the quality of existing research literature This method involves making explicit decisions regarding the identification of relevant studies, computation of standardized effect sizes, selection among multiple analyses within individual studies, and weighing the contributions of these studies In contrast to qualitative literature reviews, which may rely on subjective judgments, meta-analytic methods provide a structured approach to research synthesis However, both qualitative and quantitative research synthesis methods are ultimately limited by the nature, quality, and volume of the available research on a given topic.
Scope of This Research
Studies Identified
This research employed a comprehensive keyword search across multiple online science publication indexes, including the National Criminal Justice Reference Service (NCJRS) and ProQuest, to identify relevant studies on intimate partner violence, criminal court disposition, and recidivism Utilizing a set of 41 targeted terms, the search strategy aimed to filter out irrelevant results, as initial searches for individual terms yielded excessive and unrelated hits To refine the search, composite commands incorporating logical operators like "and," "or," and "not" were implemented, and results were further narrowed by specific disciplines such as criminology, victimology, and sociology.
This review not only utilized specific search terms but also manually explored bibliographies of pertinent articles, books, and reviews to uncover additional studies After identifying potential studies, citations were further investigated using Google Scholar to find more relevant research.
79 See generally David Moher, Alessandro Liberati, Jennifer Tetzlaff, Douglas G Altman
& The PRISMA Group, Preferred Reporting Items for Systematic Reviews and Meta-
Analyses: The PRISMA Statement, 6 PL O S M ED 1 (2009) (describing guidelines for systematic reviews and meta-analyses)
80 See infra Appendix 1 for lists of the search terms used for the systematic literature search
81 See Jeffrey G Reed & Pam M Baxter, Using Reference Databases, in T HE H ANDBOOK
OF R ESEARCH S YNTHESIS AND M ETA -A NALYSIS 73, 89–90 (Harris Cooper, Larry V Hedges & Jeffrey C Valentine eds., 2d ed 2009) (identifying alternative ways to conduct citation searches and the assumptions underlying their use)
Figure 1: Identification of Research on Repeat Offending
Following Post-Arrest Sanctions for Intimate Partner Violence
Publications identified through keyword search of databases
Publications identified in references of known publications
Potentially relevant reports identified by keywords searches and reference reviews
Abstracts reviewed by one or more author
Publications with no eligible studies
Potentially eligible studies identified for full-text review by each author
Reviewed publications with no eligible studies
A comprehensive search of existing databases revealed 935 indexed publications related to repeat offending after post-arrest sanctions for intimate partner violence, with an additional 45 publications identified through bibliographic reviews After eliminating 263 duplicates, 717 abstracts were assessed, leading to the rejection of 538 documents that did not meet the inclusion criteria A full-text review of 179 potentially eligible studies involved multiple evaluations by the co-authors to ascertain the nature of findings, sample sizes, and effect directions Ten reports lacked necessary information for standardized effect calculations, prompting outreach to authors, with seven providing the required data Ultimately, the meta-analysis identified 237 tests of the specific deterrent effects of post-arrest sanctions across 29 documents, published between 1989 and 2013, analyzing over 127,000 criminal incidents from 33 jurisdictions in the U.S and Canada Notably, some publications featured multiple independent samples, such as Peterson’s 2003 study encompassing various tests from New York City and Bronx County, and subsequent findings from New York County in 2004.
82 R ICHARD R P ETERSON , N.Y.C C RIM J UST A GENCY , T HE I MPACT OF C ASE P ROCESSING
ON R E - ARRESTS A MONG D OMESTIC V IOLENCE O FFENDERS IN N EW Y ORK C ITY 43–49 (2003) (displaying the results of three bivariate tests and three multivariate tests in the New York City sample)
83 Id at 26–28; 30–34 (also displaying the results of three bi-variate and three multivariate tests for the Bronx sample but in a different configuration)
Peterson conducted a comprehensive analysis that included one test of conviction and one test of incarceration Our search yielded two documents, comprising four samples, five studies, and twelve tests focused on specific deterrence As shown in Table 1, the search results revealed ten studies supporting the prosecution hypothesis, twenty-six studies on the conviction hypothesis, and twenty-one studies related to the incarceration hypothesis.
84 R ICHARD R P ETERSON , N.Y.C C RIM J UST A GENCY , T HE I MPACT OF M ANHATTAN ’ S
S PECIALIZED D OMESTIC V IOLENCE C OURT 57–61 (2004) (reporting one multivariate test of conviction and one multivariate test of incarceration in each sample).
Multiple Tests within Studies
In criminology and other scientific fields, reporting multiple tests within a single study is encouraged to demonstrate the strength or generalizability of findings However, this practice complicates research synthesis, as it raises questions about which tests to prioritize and how to weigh their significance This meta-analysis highlights the issue, as shown in Table 1, which details the number of individual tests for specific deterrence across three sanction types: prosecution, conviction, and incarceration Out of thirty-three studies, twelve examine only one sanction type, fifteen explore two, and six assess all three Additionally, since some studies may use overlapping incidents to evaluate different sanction types, the tests are not entirely independent Therefore, we focus on presenting the distinct results for each sanction type.
Concerns about the independence of the 237 statistical tests presented in Table 1 are significant Multiple testing of a single hypothesis can arise from various methodologies, such as using repeat offending data from victim interviews alongside official records, or by reporting repeat offending at different time intervals post-sanction For example, Peterson's analysis of the conviction hypothesis in New York City includes three tests: a bivariate analysis of any subsequent offenses, a bivariate analysis of subsequent violent offenses against domestic partners, and a multivariate analysis of the same offense type Since all tests rely on the same individual case dispositions, employing all three violates the fundamental statistical principle that each test should independently estimate the relationship being studied This dependence among tests increases the risk of obtaining biased or imprecise estimates and may lead to the erroneous rejection of a true null hypothesis.
85 See N OEL A C ARD , A PPLIED M ETA -A NALYSIS FOR S OCIAL S CIENCE R ESEARCH 191–94
86 See Mariola Moeyaert, Maaike Ugille, S Natasha Beretvas, John Ferron, Rommel Bunuan & Wim Van den Noortgate, Methods for Dealing with Multiple Outcomes in Meta-
This analysis compares three statistical methods: averaging effect sizes, robust variance estimation, and multilevel meta-analysis The study highlights the challenges in research, noting that some studies conduct numerous tests while others may report only a few or just one, leading to variability in results.
C USING THE “BEST” TEST METHOD
This research design aims to enhance the independence of multiple tests assessing the same hypothesis by identifying the most methodologically robust test within each study For example, in Peterson's analysis of the conviction hypothesis, the strongest method is a multivariate analysis focused solely on violence offenses against intimate partners This approach not only allows for a precise estimation of the deterrent effect but also facilitates the calculation of standard errors from a single test The selection of the "best" test is based on clear methodological preferences that can be objectively applied to existing research, ensuring the identification of the optimal test in each case.
This research identified the optimal test among duplicate tests across fifty-seven studies by applying seven selection criteria: analysis type, treatment group size, follow-up duration, data source, offense type, victim type, and recidivism criterion (such as re-offense, re-arrest, or re-conviction).
In our analysis, we prioritized multivariate methods over bivariate results, specifically choosing count regression models over survival analyses and survival analyses over logistic regression Within bivariate models, we favored counts and means over correlations, and correlations over simple prevalence measures When faced with multiple tests, we opted for those with the largest treatment group or the longest follow-up period If these criteria did not yield a definitive "best" test, we preferred victim interviews over official records, outcome measures based on new offenses over those based on new arrests, and new arrests over new convictions.
J Becker, Multivariate Meta-Analysis, in H ANDBOOK OF A PPLIED M ULTIVARIATE S TATISTICS AND M ATHEMATICAL M ODELING 499, 502–03 (Howard E A Tinsley & Steven D Brown eds., 2000); M ICHAEL B ORENSTEIN , L ARRY V H EDGES , J ULIAN P T H IGGINS & H ANNAH R.
R OTHSTEIN , Multiple Outcomes or Time-Points within a Study, in I NTRODUCTION TO M ETA -
In their 2001 study, Lipsey and Wilson emphasized the preference for outcome measures related to violent offenses over non-violent ones, highlighting the importance of focusing on offenses against the same victim rather than any victim This approach underscores the challenges in identifying a unique test for specific hypotheses in the realm of meta-analysis.
The focus on deterrence of re-offending in intimate partner violence research emphasizes the importance of methodological rigor through factors such as statistical analysis, sample size, and follow-up periods Unlike general violent crime, our approach prioritizes tests involving violent offenses against similar victims, as arrest records often reflect police behavior rather than the actual conduct of offenders Criminologists argue that arrest and incarceration data serve as indicators of the official response to crime rather than true offense rates Additionally, the application of increased criminal sanctions is aimed at reducing future intimate partner violence specifically, acknowledging that a significant portion of such violence involves repeated incidents with the same victim Consequently, this research favors measures of offense frequency over prevalence or time to failure.
In a review of twenty-two studies on the specific deterrent effects of criminal sanctions, only one test was reported per study, which we deemed the "best." Our analysis identified twelve of these top tests as employing multivariate analyses, frequency measures, or extended follow-up periods Additionally, in eleven tests, the selection was primarily influenced by the sample size and follow-up duration.
In our analysis, we identified the "best" tests based on their criteria for recidivism, specifically focusing on re-offense rates We selected six tests that concentrated solely on violent offenses and one test that measured repeat offending against the same victim The implications of our selection criteria are illustrated in Table 2, which details the frequency of all 237 reported tests alongside the "best" tests across seven selection criteria Notably, only 55% of the reported tests were multivariate analyses, while 72% of the fifty-seven "best" tests utilized multivariate methods, highlighting a significant trend in effective assessment strategies.
88 See John I Kitsuse & Aaron V Cicourel, A Note on the Uses of Official Statistics, 11
S OC P ROBS 131, 132–34 (1963); D ELBERT S E LLIOTT , C TR FOR THE S TUDY & P REVENTION
OF V IOLENCE , L IES , D AMN L IES , AND A RREST S TATISTICS 3–5 (1995); Terence P Thornberry
& Marvin D Krohn, Comparison of Self-Report and Official Data for Measuring Crime, in
M EASURMENT P ROBLEMS IN C RIMINAL J USTICE R ESEARCH 43, 43–48 (John V Pepper & Carol
A recent analysis reveals that 77.2% of the "best" tests focus solely on violent repeat offenders, compared to 67.1% of all tests evaluated Despite these advancements, victim interviews are utilized only 22.8% of the time in the "best" tests, compared to 27.8% in all tests While the "best" tests exhibit preferred methodological traits, they only select from existing studies, highlighting a significant reliance on the independence of research observations across the three sanction hypotheses Notably, the criteria for determining the "best" tests were established before the study search and standard effect size calculations commenced.
In alignment with previous deterrence research, this study initially aimed to test three specific hypotheses regarding the deterrent effects of sanctions, utilizing a two-sided t test with a conventional significance level of p < 05 However, the limitations of depending exclusively on this arbitrary threshold prompted the American Statistical Association (ASA) to suggest alternative approaches.
Researchers must consider various contextual factors to make accurate scientific inferences Our study adheres to the recommended p-value threshold of 0.05, as suggested by John Ioannidis Additionally, we offer qualitative context for our findings and present results from multiple sensitivity tests, including basic counts of statistical tests and revised meta-analytic outcomes after excluding tests with the highest and lowest effect sizes.
90 Ronald L Wasserstein & Nicole A Lazar, The ASA Statement on p-Values: Context,
91 John P.A Ioannidis, The Importance of Predefined Rules and Prespecified Statistical
Analyses: Do Not Abandon Significance, 321 JAMA 2067, 2067–68 (2019)
D ALTERNATIVE APPROACHES TO ADDRESSING INDEPENDENCE
Prominent researchers have suggested alternative methods to address the challenges posed by multiple testing in prior research One approach involves calculating the average effect size and variance from all tests within a study, despite limited justification for averaging multiple variances Another method also averages effect sizes but employs statistical techniques to estimate "robust" standard errors, which are often larger than actual standard errors A third approach utilizes hierarchical linear modeling (HLM) to estimate effect sizes and standard deviations both within and across studies All these methods apply the traditional inverse-variance weighting, prioritizing larger studies for more precise effect size estimates While these approaches share similarities, they differ in their assumptions about research nature, ranking of methods, statistical independence, and meta-analysis objectives Ultimately, the compelling arguments for adopting the "best" test approach significantly influence our conclusions.
All approaches for estimating deterrence effects utilize a consistent method to standardize results from various statistical tests into a common metric, specifically the logged odds ratio Following recommendations from analysts, logged odds ratios were calculated for 237 tests by inputting bivariate proportions, mean differences, correlations, and regression coefficients, along with their variance measures and sample sizes, into Comprehensive Meta-Analysis (version #3) This process yielded standardized odds ratios and associated standard errors for each test, facilitating a coherent comparison of effect sizes and variance estimates.
92 Fulgencio Marín-Martínez & Julio Sánchez-Meca, Averaging Dependent Effect Sizes in Meta-Analysis: A Cautionary Note About Procedures, 2 S PANISH J P SYCH 32, 37 (1999)
93 Larry V Hedges, Elizabeth Tipton & Matthew C Johnson, Robust Variance Estimation in Meta-Regression with Dependent Effect Size Estimates, 1 R SCH S YNTHESIS M ETHODS 39,
94 See generally S TEPHEN W R AUDENBUSH & A NTHONY S B RYK , H IERARCHICAL L INEAR
M ODELS : A PPLICATIONS AND D ATA A NALYSIS M ETHOD s (2d ed 2002) (providing examples and explanations of the theory and use of hierarchical linear models)
95 Moeyaert, Ugille, Beretvas, Ferron, Bunuan & Van den Noortgate, supra note 86, at
561–62; Larry Hedges & Kimberly Maier, Meta-Analysis, in T HE SAGE H ANDBOOK OF
M ULTILEVEL M ODELING 487, 497 (Mark A Scott, Jeffrey S Simonoff & Brian D Marx eds.,
96 Joseph L Fleiss & Jesse A Berlin, Effect sizes for dichotomous data, in T HE
H ANDBOOK OF R ESEARCH S YNTHESIS AND M ETA -A NALYSIS , 237 (Harris Cooper, Larry V Hedges & Jeffrey C Valentine, eds., 2d ed 2009); T ERRI D P IGOTT , A DVANCES IN M ETA -
Alternative Approaches to Addressing Independence
Best Effects
Figure 2 illustrates a forest plot showcasing the results of each study's most significant test, arranged by estimated log odds ratio to highlight their support for deterrence The plot includes confidence intervals, two-sided z test values, p values, and the relative weight of each study calculated using the inverse-variance weighting method, which ranges from approximately three to nearly twenty percent Out of the ten studies examining the prosecution hypothesis, seven indicate a deterrent effect, with three of these surpassing the 1.96 standard threshold for statistical significance Collectively, these three studies account for about twenty percent of the total relative weight.
None of the three studies indicating an escalation effect showed statistically significant results Despite this, the cumulative weight of the studies suggesting escalation effects surpasses forty percent A traditional literature review might conclude a seven-to-three ratio in favor of deterrent effects; however, the limited number of studies and the significant weight assigned to those not supporting deterrence lead to a meta-analytic outcome suggesting deterrence (-0.196), although it remains statistically nonsignificant with a p value of 0.095.
In our sensitivity analysis, we examined the impact of excluding the prosecution study that most strongly supports deterrence, resulting in a mean odds ratio of -0.115, which was not statistically significant Conversely, when we removed the study least supportive of deterrence, the analysis of the remaining nine studies yielded a mean odds ratio of -0.246, demonstrating statistical significance with a p value of 041.
Figure 3 presents twenty-six studies examining the specific deterrent effects of conviction, with fifteen (58%) aligning with deterrence theory; however, only two findings are statistically significant Among the eleven studies indicating an escalation effect due to conviction, three show statistical significance The relative weights of these studies range from 1.55 to 5.78, indicating less dominance by any single study compared to the prosecution hypothesis The meta-analytic summary yields an effect size of 0.021, which is not only small but also contrary to the expected deterrent effect of convictions Notably, removing the most or least supportive studies does not alter the direction or statistical significance of the conviction effect.
A review of twenty-one studies on the deterrent effects of incarceration reveals that while nine indicate some level of deterrence, only two are statistically significant Conversely, among the twelve studies showing an escalation effect, seven are statistically significant The weights of these studies vary widely, with no clear trend favoring deterrence or escalation The overall findings suggest that incarceration is linked to increased repeat offending, as indicated by a statistically significant log odds ratio of 0.367 (p = 0.001), highlighting that incarceration for intimate partner violence offenders leads to a substantial rise in their offending rates compared to less severe sanctions like prosecution or conviction.
Among the twenty-one studies analyzed, one outlier exhibits an effect size five times larger than the next highest, leading to a significant average effect size of 0.269 after its removal, with a p value of 0.012 Conversely, excluding the study most indicative of a deterrent effect from incarceration raises the log odds of escalation to 0.405, accompanied by a p value of less than 0.000.
The analysis presented in Table 3 reveals distinct outcomes for the three hypotheses regarding post-arrest sanctions While the ten prosecution studies indicate a modest deterrent effect, they fall short of achieving statistical significance Confidence in these findings is tempered by the limited number of studies and considerable variation in results, particularly when considering the impact of outlier studies In contrast, the twenty-six studies related to the conviction hypothesis demonstrate consistent results, suggesting no deterrent or escalation effect from convictions Additionally, the twenty-one tests examining the incarceration hypothesis show a significant escalation effect that remains robust even after excluding extreme outliers.
Critics argue that the selection of "best" tests in meta-analysis may lack objectivity, suggesting the need for alternative meta-analytic methods to address the non-independence of research observations This concern, if valid, could compromise the primary aim of meta-analysis: to eliminate bias from qualitative literature reviews To tackle this issue, Table 3 presents findings from our meta-analyses alongside results derived from three alternative methods The first method averages effects within each study (Average), the second treats all tests as independent effects while employing robust standard errors for coefficient estimation and statistical testing (RSE), and the third utilizes hierarchical models that incorporate both study-level and effect-level tests (HLM) Table 3 summarizes the findings for each of the three sanction hypotheses across all four meta-analytic approaches.
Despite variations in log odds ratios and standard errors among the four methodologies, the core findings remain remarkably consistent All approaches indicate no effect for convictions and show escalation effects for incarceration While prosecution effects suggest deterrence, three alternative methods yield p values above the conventional 0.05 threshold, unlike the primary method which does not Notably, two of these alternative approaches are just slightly above the 0.05 mark by 0.004 and 0.006, raising questions about the rigidity of the fixed standard Consistent with our results on convictions and incarceration, other criminological meta-analyses have also demonstrated similar outcomes across various meta-analytical techniques applied to the same research data.
Based on the findings from the "best" test approach to meta-analysis, we conclude that the null hypothesis should be accepted, indicating that prosecution does not influence the occurrence of repeat intimate partner violence.
97 Hedges, Tipton & Johnson, supra note 93, at 41
98 R AUDENBUSH & B RYK , supra note 94, at 208–10
The outcomes from the initial two methods, which involved applying both the best and average effects, were computed using Comprehensive Meta-Analysis Version 3 In contrast, the results from the "RSE" method were derived using SPSS 25, while the HLM findings were generated with HLM 7.
Hierarchical linear and nonlinear modeling
100 Tammy Rinehart Kochel, David B Wilson & Stephen D Mastrofski, Effect of Suspect
In the study "Race on Officers’ Arrest Decisions," published in 49 Criminology 473, 489 (2011), significant odds ratios ranging from 1.53 to 1.74 were reported through various meta-analysis methods The research established explicit criteria for selecting the best tests from 29 studies, adhering to a p < 05 standard with two-tailed tests prior to data extraction for logged odds ratios and formal statistical analyses Despite ongoing debates in the research community regarding the validity of the 05 standard and the diverse methodologies for conducting meta-analyses, the authors advocate for presenting results from multiple approaches This approach allows for a comprehensive understanding of how findings may vary, emphasizing that methodological differences should not dictate preference for a single approach.
Our meta-analyses revealed that all four alternative approaches yielded consistent results regarding the conviction and incarceration hypotheses, indicating that our conclusions are robust across different methodologies Conversely, the three alternative methods used to test the prosecution hypothesis generated inconsistent findings, diverging from the results obtained through our primary approach.
The "best" test approach yielded findings in the deterrence direction, but did not achieve statistical significance at the p < 05 level Our analysis of three alternative approaches highlights that the lack of a statistically significant deterrent effect for prosecution is influenced by the methodology employed However, we do not assign equal importance to the findings from these alternative approaches when drawing conclusions about prosecution's impact Likewise, proponents of alternative methods need not regard our approach as equally valid in assessing prosecution's effects Ultimately, this paper emphasizes the findings from our original statistical tests to ascertain the presence of significant effects.
Meta-analyses often face challenges related to publication bias, which arises when completed studies are not reported and thus excluded from systematic reviews This issue highlights the importance of authors submitting their findings and journals ensuring comprehensive publication to enhance the integrity of research synthesis.
101 See Robert Rosenthal, The “File Drawer Problem” and Tolerance for Null Results, 86
Study Design Moderators of Deterrent Effects
DISCUSSION
104 Kochel, Wilson & Mastrofski, supra note 100, at 489–90
105 Travis C Pratt, Francis T Cullen, Kristie R Blevins, Leah E Daigle & Tamara D Madensen, The Empirical Status of Deterrence Theory: A Meta-Analysis, in T AKING S TOCK :
T HE S TATUS OF C RIMINOLOGICAL T HEORY 367, 384 (Francis T Cullen, John Paul Wright & Kristie R Blevins eds., 2008)
Mark W Lipsey's article, "Those Confounded Moderators in Meta-Analysis," published in the Annals of the American Academy of Political and Social Science, highlights the challenges and intricacies involved in interpreting moderator analyses within meta-analyses.
107 These sources were: Peter Jaffe, David A Wolfe, Anne Telford & Gary Austin, The
The impact of police charges on incidents of wife abuse has been examined in various studies, including a notable article in the Journal of Family Violence, which highlights the significant effects of law enforcement intervention Research by Lauren Marsland, Darryl Plecas, and Tim Segger also investigates the reluctance of domestic violence victims to report re-assault in jurisdictions that enforce pro-charge policies These findings underscore the complexities surrounding police involvement in domestic violence cases and the challenges faced by victims in seeking help.
J Crim L & Criminology); N ATHALIE Q UANN , C ANADA D EP ’ T OF J UST , O FFENDER P ROFILE AND R ECIDIVISM A MONG D OMESTIC V IOLENCE O FFENDERS IN O NTARIO (2006) may be sufficient to address this basic question, especially if they can address the problem of selecting high-risk offenders for more severe sanctions and measure repeat offending using victim interviews as well as official records Even if successful, those prosecution studies cannot explain why potential offenders do or do not repeat their offenses For that more important set of questions, future research on sanctions for intimate partner violence needs to study the decision-making of potential offenders and incorporate not just the characteristics of sanctions but the variety of factors—including emotions and fears—that can test alternative theories of the causes of intimate partner violence 108 To be especially useful for understanding intimate partner violence, such panel studies should interview both partners and capture information relevant to other theories of intimate partner violence, such as victim empowerment 109
This meta-analysis reveals that existing research has not sufficiently examined whether harsher criminal justice penalties are applied to individuals perceived as more likely to reoffend While some studies have attempted to mitigate selection bias through random case assignments or propensity scores, there is potential in creating precise prediction models for future intimate partner violence incidents Comparing predicted failures to actual failures among those arrested, prosecuted, convicted, or incarcerated could provide valuable insights into this issue.
In the realm of deterrence research, experts have not yet determined the optimal duration of risk necessary to effectively evaluate the influence of criminal sanctions on future offenses Consequently, our comprehension of deterrence remains inadequate to ascertain the immediacy and duration of the effects that specific sanctions may have on various types of criminal behavior.
108 See Greg Pogarsky, Sean Patrick Roche & Justin T Pickett, Offender Decision-
Making in Criminology: Contributions from Behavioral Economics, 1 A NN R EV
Cattaneo and Goodman highlight the crucial role of victim empowerment as active agents in mitigating intimate partner violence, while Mills underscores the importance of understanding how victims contribute to the relationship between criminal justice policies and recidivism.
A study by Mears et al (2016) in the Journal of Criminal Law and Criminology examines various methodological approaches to assess how imprisonment influences recidivism rates and future criminal behavior.
Research indicates that using domestic violence risk assessments can significantly reduce recidivism among high-risk perpetrators (Williams & Stansfield, 2017) To understand the lasting impact of criminal sanctions, it is essential to explore how long these deterrent effects endure, similar to the duration of relief provided by medications or vaccines Future studies on sanctions related to intimate partner violence should focus on offense-based frequency measures of violent incidents against the same victim and incorporate insights gained from victim interviews This approach will enhance the understanding of deterrent effects in intimate partner violence cases.
This research highlights the neglect of intimate partner violence and alternative sanctions beyond arrest in current deterrence studies The findings indicate that previous reviews have overlooked significant research demonstrating varying deterrent effects of different criminal sanctions on intimate partner violence Future deterrence research should consider Durlauf and Nagin's proposal that these effects may differ based on the type of sanction and the specific nature of the offenses.
Current federal programs encourage increased prosecution, conviction, and incarceration for intimate partner violence; however, research indicates that such measures may actually lead to higher rates of violence against intimate partners These findings challenge the effectiveness of incarceration as a solution to this issue While one study alone cannot inform new policies, the potential negative impacts of incarceration must be considered Advocates for harsher post-arrest penalties need to propose alternatives to incarceration or present compelling justifications that demonstrate significant social benefits to counteract the heightened violence linked to these punitive measures.
112 Durlauf & Nagin, supra note 37, at 85–86
113 Robert J Sampson, Christopher Winship & Carly Knight, Translating Causal Claims:
Principles and Strategies for Policy-Relevant Criminology, 12 C RIMINOLOGY & P UB P OL ’ Y
APPENDICES
Appendix 1 Keyword and Search Terms for Meta-Analysis
Intimate Partner Violence Court Disposition Recidivism
Intimate partner violence Criminal justice system Recharge
Intimate partner abuse Disposition Recidivism
Spousal violence Legal Intervention Repeat abuse
Wife assault Legal system Repeat offense
Violence against intimate partner Prosecution
Appendix 2: Documents Used in the Meta-Analyses
Margret E Bell, et al., Criminal case outcomes, incarceration, and subsequent intimate partner violence, 28 JOURNAL OF FAMILY VIOLENCE (2013)
Eve Buzawa, et al., Response to Domestic Violence in a Pro-Active Court Setting (University of Massachusetts-Lowell 1999)
Lauren Bennett Cattaneo & Lisa A Goodman, Through the lens of therapeutic jurisprudence: The relationship between empowerment in the court system and well-being for intimate partner violence victims,
Robert C Davis, et al., The deterrent effect of prosecuting domestic violence misdemeanors, 3 CRIME & DELINQUENCY (1998)
Jeffrey A Fagan, Cessation of family violence: deterrence and dissuasion, in FAMILY VIOLENCE (Lloyd Ohlin & Michael Tonry eds., 1989)
Mary A Finn, Evidence-Based and Victim-Centered Prosecutorial Policies: Examination of Deterrent and Therapeutic Jurisprudence Effects on Domestic Violence, 12 CRIMINOLOGY & PUBLIC POLICY (2013)
Durant Frantzen, et al., Predicting case conviction and domestic violence recidivism: measuring the deterrent effects of conviction and protection order violations, 26 VIOLENCE & VICTIMS (2011)
Lisa A Frisch, et al., Family Protection and Domestic Violence: Intervention Act of 1994: Evaluation of the Mandatory Arrest Provisions (Division of Criminal Justice Services, New York State
Thomas P George, Domestic Violence Sentencing Conditions and Recidivism (Washington State Center for Court Research 2012)
Melissa Gross, et al., The impact of sentencing options on recidivism among domestic violence offenders: A case study, 24 AMERICAN JOURNAL OF CRIMINAL JUSTICE (2000)
Adele Harrell, et al., The Impact of JOD in Dorchester and Washtenaw County (Justice Policy Center, The Urban Institute 2007)
Peter G Jaffe, et al., Impact of Police Laying Charges, in LEGAL RESPONSES TO WIFE ASSAULT: CURRENT TRENDS AND EVALUATION (N Zoe Hilton ed 1993)
Rodney Kingsnorth, Intimate partner violence: Predictors of recidivism in a sample of arrestees, 12 VIOLENCE AGAINST WOMEN (2006)
Andrew R Klein & Terri Tobin, A Longitudinal study of arrested batterers, 1995-2005, 14 VIOLENCE AGAINST OF VIOLENCE (2008)
Gilbert Stillman Macvaugh, III, Outcomes of court intervention and diversionary programs for domestically violent offenders (2004) (Dissertation, Antioch New England Graduate School)
In the report "Reticence and Re-assault among Victims of Domestic Violence in a Pro-charge Jurisdiction," authors Lauren Marsland, Darryl Plecas, and Tim Segger analyze the challenges faced by domestic violence victims in British Columbia The study highlights the issues of victim reluctance to engage with the legal system and the prevalence of re-assault in pro-charge jurisdictions Their findings underscore the need for improved support mechanisms and interventions to empower victims and enhance their safety The report, prepared for the British Columbia Ministry of Attorney General, aims to inform policy and practice in addressing domestic violence effectively.
Judith McFarlane, et al., Women filing assault charges on an intimate partner: Criminal justice outcomes and future violence experienced, 6 VIOLENCE AGAINST WOMEN (2000)
Christopher M Murphy, et al., Coordinated community intervention for domestic abusers: Intervention system involvement and criminal recidivism, 13 JOURNAL OF FAMILY VIOLENCE (1998)
Stan J Orchowsky, Evaluation of a Coordinated Community Response to Domestic Violence: The Alexandria Domestic Violence Intervention Project (Applied Research Associates 1999)
Richard R Peterson, The Impact of Case Processing on Re-arrest among Domestic Violence Offenders in New York City (New York City Criminal Justice Agency 2003)
Richard R Peterson, The Impact of Manhattan’s Specialized Domestic Violence Court (New York City Criminal Justice Agency ( 2004).