Overall rearrest rates: the cohort, the event, and the follow-up period
Overall rearrest rates were lower for people released after serving sentences for rape or sexual assault than for the other offense categories combined in the Bureau of Justice Statistics' 2005 release cohort. Over nine years, the comparison was about 67 percent against 84 percent. Yet BJS reported that this group was three times as likely as other released prisoners to be rearrested for a sexual offense. A lower rate of arrest for any offense and a higher rate for a particular offense can coexist. They measure different outcomes within the same follow-up.
The National Institute of Justice's August 2019 TECHBeat publication notice identifies the underlying study by its full title, Recidivism of Sex Offenders Released from State Prison: A 9-Year Follow-Up (2005-14). Its comparison concerns the most serious commitment offense: rape or sexual assault versus other offenses. That is a prison-release classification, rather than a count of everyone on a registry or everyone who has ever received a sexual-offense conviction. The distinction fixes the population to which these overall rearrest rates apply.
Wendy Sawyer's June 2019 analysis for the Prison Policy Initiative reproduces several results from the BJS tables. Within the rape-or-sexual-assault group, 7.7 percent were rearrested for rape or sexual assault; rearrest rates for property, drug, and public-order offenses were higher. Half had a new arrest leading to a conviction for any offense, compared with 69 percent across the release cohort in the states with those data. The group also had lower five-year returns to prison. Sawyer's criticism of the government's emphasis is commentary; the separate percentages describe separate measured events.
The Center for Sex Offender Management's 2001 technical paper, preserved by the Government Publishing Office, explains the consequences of choosing those events. An arrest measure includes people who are never convicted. A conviction measure is narrower because it requires a finding of guilt. A return to prison can mean imprisonment for a new offense, but it can also mean a technical breach of release conditions. The paper gives consuming alcohol or prohibited contact with children as examples of such conditions. A return-to-custody percentage can therefore include conduct that did not generate a new criminal conviction.
That paper also warns against treating the studied population as homogeneous. Sexual offenses encompass different acts and victim relationships, and grouping them together can conceal differences associated with later offending. Observation time creates another problem. People within a nominal ten-year study may have spent very different periods in the community. The paper describes survival analysis as a way of incorporating each person's time at risk rather than treating a simple percentage as though everyone had an equal opportunity to offend. Comparisons across studies also require comparable periods in the community.
A 2015 research brief from the Justice Department's SMART Office supplies an earlier comparison. It summarizes a BJS study of 9,691 men released from prisons in 15 states in 1994. During three years, 43 percent were rearrested for any crime, against 68 percent among the comparison group without a sexual commitment offense. Arrest for a sexual crime ran in the opposite direction: 5.3 percent against 1.3 percent. The pattern resembles the later BJS comparison, but the shorter observation period and different release cohort prevent those percentages from being interchangeable.
The SMART brief also documents a limitation in official records. Many sexual offenses never reach police, and only some reported offenses lead to an arrest. Recorded sexual recidivism consequently understates actual reoffending. At the same time, comparisons become unreliable when studies use different definitions, populations, or follow-up lengths. Lower overall rearrest rates describe an observed group comparison. They do not establish that sexual violence is fully captured by arrest statistics, or that every person within the lower-rate group has the same future risk.
An original longitudinal study by R. Karl Hanson, Seung C. Lee, and David Thornton makes the population issue concrete. Using Canadian criminal-history records and supplementary searches, they obtained follow-up information for 377 adult men with sexual-offense histories. The sample had been selected for above-average risk, and its median observation period was 20.5 years. A subgroup of 190 classified as high risk and high need had a 40 percent observed sexual-recidivism rate across the extended follow-up. These were deliberately selected men followed for much longer than the 2005 BJS cohort, rather than another estimate for that cohort.
Hanson and colleagues excluded time incarcerated for nonsexual offenses from time at risk before sexual recurrence. Their minimum qualifying event was a charge, although most recorded events were convictions. Among 181 men who had remained free of recorded sexual offending for more than 20 years, only two subsequent sexual-recidivism events were observed. That conditional result describes a different stage of follow-up from the cumulative rate starting at release. Overall rearrest rates and sexual-recidivism rates change meaning when the starting point, observation time, qualifying event, or selected population changes.
- Prison Policy Initiative, Wendy Sawyer, analysis of the 2019 BJS sexual-offense follow-up, June 6, 2019 — www.prisonpolicy.org
- National Institute of Justice, TECHBeat, July/August 2019, official publication notice describing the BJS comparison — nij.ojp.gov
- Center for Sex Offender Management, Recidivism of Sex Offenders, May 2001, measurement definitions — www.govinfo.gov
- U.S. Department of Justice SMART Office, Roger Przybylski, Recidivism of Adult Sexual Offenders, July 2015 — smart.ojp.gov
- R. Karl Hanson, Seung C. Lee, and David Thornton, Long Term Recidivism Rates Among Individuals at High Risk to Sexually Reoffend, original longitudinal study — pmc.ncbi.nlm.nih.gov