Comparing Self-Report to Official Measures of Inmate Misconduct

Abstract

Studies have revealed systematic measurement errors in self-report data on crime and deviance resulting from poor recall and/or underreporting by certain groups of respondents. Official crime data have also been criticized, but for different reasons (e.g. gross underestimations of less serious offenses). Very similar observations have been made in studies of inmate crime (misconduct committed by prison inmates). Despite these criticisms, official data on inmate misconduct continue to be the most frequently used data in related studies. This study compared self-report and official data on inmate assaults, property thefts, and drug offenses for samples of inmates from 46 correctional institutions for adults in Ohio and Kentucky. Findings revealed that officially recorded misconduct underestimates the total volume of inmate crime. Analyses designed to uncover sources of the divergence between self-reported misconduct and officially recorded misconduct revealed far more consistencies than differences in the magnitude of inmate and facility effects on the different types of offenses. A few important differences did emerge in the magnitude of effects such as amount of time served (at the individual level) and facility population size (at the aggregate level).

Keywords:

• inmates
• prisons
• self-report data
• official data

Acknowledgments

This study was supported, in part, by grants from the National Institute of Justice (Award #2007-IJ-CX-0010), the National Science Foundation (Award #SES-07155515), and the Office of the Provost at the University of South Carolina. The opinions, findings, and conclusions expressed in this study are those of the author and do not necessarily reflect those of the Department of Justice, National Science Foundation, or the University of South Carolina. The authors also wish to thank Guy Harris, along with Brian Martin and Gayle Bickle with the Ohio Department of Rehabilitation and Correction for their assistance with the collection of the data for this study.

Notes

1. Kentucky also has three privately operated facilities for adult offenders. Those facilities were not included in the study per the wishes of KDOC. With two exceptions, inmates housed in correctional camps, mental health units, reception units, or youthful offender units were excluded due to practical constraints and unmeasured structural and managerial differences that exist between those units and the primary facilities in which these units existed. Inmates housed in the correctional camp at the Ohio State Penitentiary (Ohio’s supermax facility) were included for theoretical reasons dictated by the larger project. Inmates housed in the correctional camp for females at the Trumbull Correctional Institution were also included. At the time of the study, Ohio had three other facilities for women, but two of those facilities were pre-release centers, which typically do not house inmates longer than one year. The camp for females at Trumbull Correctional Institution was the most similar institution to the Ohio Reformatory for Women, which was the primary facility for women in Ohio. The camp for females at Trumbull Correctional Institution, which is physically separate from the main facility, was treated as a separate facility in all of the analyses.

2. The larger project included a longitudinal element (Ohio only) and so larger sample sizes were sought in 11 Ohio facilities, although the ODRC only granted our request for larger samples in seven of these facilities. The sampling frames in the 11 facilities selected for the longitudinal data collection were restricted to only those inmates who had at least six months of their sentence remaining at the time of the first survey (≈ 85% of the inmates in these facilities had six months remaining). The decisions to pursue larger number of inmates and restrict the samples to only inmates with at least six months of their sentence remaining to serve were made to reduce the effects of attrition in the longitudinal analysis. Our goal was to obtain usable information on at least 100 inmates per facility (or at least 200 inmates per facility in the facilities selected for the longitudinal data collection). The 30% oversample was included to compensate for refusals and incomplete surveys, based on the recommendations of research staff at the ODRC. In Kentucky, we targeted sample sizes of 200 inmates per facility, but these numbers were adjusted based on the inmate population and resource demands placed on individual facilities. Non-English speaking inmates were excluded from the study due to resource constraints.

3. Inmates were unavailable because they had been released or transferred (N = 125), posed a safety risk or were in the infirmary (N = 42), were receiving a visit (N = 44), or were not in the facilities on the date of data collection (e.g. out to court) (N = 86).

4. The analyses that are presented (and subsequently described) involved adjustments for the correlated error that may have existed across facilities due to differences in the methods of administering the survey. The analyses were also limited to the examination of within facility differences in misconduct, reducing the likelihood that differences in findings based on analyses of self-report vs. official misconduct could be attributed to differences in how the survey was administered across facilities.

5. Some inmates did not receive or respond to their pass. In most cases, we managed to locate these inmates and offer them the opportunity to participate. Some inmates could not be found and were treated as “refusals.”

6. We also considered examining the incidence of each type of misconduct, however, inspection of the distributions of incidence measures of the different types of misconduct revealed no substantive differences between the incidence and prevalence of misconduct. To illustrate, the least extreme example was the measure of drug misconduct. Examination of the distribution of the official measure of this outcome revealed that 2% of the sample engaged in misconduct, but less than one-fourth of a percent of the sample engaged in more than one drug offense during the study period. Examination of the distribution of the comparable self-report measure revealed that nearly 5% of the sample reported a drug offense, but just over 2% of the sample reported engaging in more than one incident of drug misconduct.

7. A potentially relevant facility-level variable could be whether a facility used direct supervision. However, all of the facilities included in this study used direct supervision in the majority of their housing units. Exceptions included a few of the housing units in maximum security facilities, which were supervised by officers in control booths designated for those housing units. A dichotomous measure of “direct supervision” was examined for possible inclusion in the analysis but was found to be highly correlated with close security.

8. The variable “male” was grand mean-centered because it was a constant within facilities.

9. We also estimated a hierarchical measurement model with items (self-report and official misconduct) nested within individuals, who were nested within facilities (see, e.g. Kamata, 2001; Kirk, 2006; Kuo, Mohler, Raudenbush, & Earls, 2000; Raudenbush & Bryk, 2002). For each types of misconduct (e.g. assault), these analyses revealed that “misconduct” did not vary significantly across individuals (p  .05), although significant variation did exist across facilities. These findings suggest that there was no item dependence across inmates, but only across facilities. Although these findings could be interpreted to mean that there were no differences in the inmate-level effects across the two data sources, we chose to proceed with the analyses described above because it is also possible that the findings from these hierarchical analyses could be attributed to the relatively few number of items measuring each type of misconduct. Nonetheless, the findings from these analyses do generally support the study findings which are subsequently discussed.

10. Multivariate analyses of some of the level-2 predictor variables and the misconduct measures could have revealed significant partial correlations even though the zero-order correlations reported in Table 3 were insignificant (Pedhazur, 1997).

11. Although the correlation between design capacity and population was strong (Pearson’s r = .83), including both measures in the same model did not appear to produce biased estimates or inflated standard errors. As such, this was considered nonessential collinearity (see, e.g. Marquardt, 1980; Pedhazur, 1997).

Additional information

Notes on contributors

Benjamin Steiner

Benjamin Steiner is an assistant professor in the School of Criminology and Criminal Justice at the University of Nebraska—Omaha. He holds a PhD from the University of Cincinnati.

John Wooldredge

John Wooldredge is a professor in the School of Criminal Justice at the University of Cincinnati. His research and publications focus on institutional corrections (crowding, inmate violence, inmate adaptation), and criminal case processing (sentencing and recidivism, extra-legal disparities in case processing and outcomes). He is currently involved in research on inter-judge variability in sentencing, short- versus long-term effects of sentencing guidelines on extra-legal disparities in prison sentences, and official responses to prison inmate rule violations.