Smallest Geographic Unit:
Date of Collection:
Unit of Observation:
Individuals who were under active supervision by the parole office (on the day of roster preparation) after being released from state prison; were 18 years of age or older; were not re-incarcerated, absconded, in a half way house or community treatment center; had not been transferred to other parole offices; did not have warrants issued for arrest; and were not in violator status.
All individuals who were under active supervision by state parole offices (on the day of roster preparation) after being released from state prison, were 18 years of age or older, and were not re-incarcerated, absconded, in a half way house or community treatment center, did not have warrants issued for arrest, were not in violator status, and/or had not been transferred to other parole offices. All state parole offices were eligible for selection except those excluded due to uniqueness of supervision type, or due to geographic and/or size limitations (please see sampling description).
Due to the possibility of individual disclosure, all interviews from offices participating in the pre-test were eliminated from the final data file prepared for ICPSR. Because of multiple changes in the data instrument after the pre-test, the risk of identification of participating offices and individual respondents in the pre-test was determined to be too great.
The Prison Rape Elimination Act of 2003 (P.L. 108-79) requires the Bureau of Justice Statistics (BJS) to carry out a comprehensive statistical review and analysis of the incidence and effects of prison rape for each calendar year.
A multi-stage stratified systematic probability-proportional-to-size (PPS) sample design was used to select a random sample of over 250 parole (field or district) offices across the United States, followed by a random sample of parolees supervised at those offices.
The parole office sampling frames were developed using the 2006 Census of Parole Offices conducted by the United States Census Bureau on behalf of BJS. The original frame consisted of two sub-frames: a field office group (FOG) level frame of 800 field office groups and a district level frame of 101 district (or regional administrative) offices. Parole field offices were not included in the office sampling frame if they: (1) were administrative offices that did not actively supervise parolees, (2) were specialty offices (e.g., those supervising only sex offenders or only individuals committed to treatment/confinement facilities), (3) were located in difficult to reach geographic regions, (4) had a population of less than 40 and could not be combined with another office within 75 miles for sampling purposes. A separate frame for district offices was created because field offices within these districts (or regions) were not enumerated in the 2006 census. All district offices were included in this frame, with the intent of specifying and enumerating all field offices under their authority once selected. Field offices within these districts, once specified, were subject to the same criteria as the field office frame, with the exception that field offices were grouped if within 120 miles.
Each of these sub-frames was sampled from separately to produce the final sample. A two-stage stratified systematic PPS sample design was used for selection from the office sub-frame, while a three-stage design was used for the district sub-frame.
Field Office Groups (FOGs) were selected from the field office frame using PPS sampling procedure. A total of 151 FOGs (seven involving combined offices) were selected from the office sampling frame. At the time of sample selection, an additional 37 reserve offices were selected. The first 10 reserve offices on the list were eventually activated. The first four offices were activated as a result of four sampled offices having no eligible respondents supervised at those sites. The remaining sites were activated to offset the possibility of refusals in states for which approvals were slow to materialize; however, those activated were not eventually required for replacement and served only to enhance the sample size.
District offices were selected from the district office sub-frame using PPS sampling procedures. A total of 50 districts were selected for inclusion. Following selection of districts, project staff contacted states requesting an enumeration of field offices and office sizes within each selected district. Offices with less than 40 parolees were eliminated if there was no other office within 120 miles for grouping. FOGs within each district sampled were then sampled using PPS sampling procedures. A total of 71 FOGs were selected from these districts (21 involving combined offices) after enumeration of offices within the selected districts. An additional twenty reserve districts were selected from this frame. Eleven of the first twelve districts on the reserve list were activated. One was activated as a result of a sampled district having no eligible supervised parolees. The remaining districts were activated to offset the possible loss of states for which access was difficult to obtain. However, because approval was eventually obtained for all sampled states, the extra reserve districts were not required for replacement and served only to enhance the sample size. Once enumerated and selected, all chosen offices (or FOGs) agreed to participate, except one, and another office was randomly selected from the same district.
After approvals from states and offices were received, a roster of eligible former prisoners was requested from each parole office within each FOG to be developed at approximately two and half to three weeks prior to scheduled interviewing period at the office. Eligible former prisoners were defined as:
Individuals who are under active supervision by the parole office (on the day of roster preparation) after being released from state prison and were 18 years of age or older. Individuals who had absconded, were re-incarcerated or committed, were in a half way house or community treatment center, had a warrant issued for arrest, were in violator status, or had been transferred to another parole office were ineligible.
A sample of eligible former prisoners was randomly selected from these rosters for inclusion in the study.
The target sample size was 260 (196 males; 64 females) for the larger offices and 130 (98 males; 32 females) for the smaller offices. Female former prisoners were sampled at a higher rate than males to insure that sufficient numbers were included to provide precise estimates at the national level. For sites with a roster population of 130 or less; all eligible respondents on the roster were selected to participate in the survey. In total, 34,782 former prisoners were selected to participate in the study.
As a final step in data preparation, weights were attached to the data file to provide accurate national estimates.
The sample design of the FPS used a multi-stage stratified systematic probability proportional-to-size (PPS) sample design. In the district sub-frame, three stages of selection were required: (1) the selection of districts, (2) selection of offices (or FOGs) within districts, and (3) selection of parolees within offices (or FOGs). Selection from the field office sub-frame included only two stages; (1) selection of offices (or FOGs), and (2) selection of parolees within offices (or FOGs). Each of these sub-frames were sampled from separately to produce the sample of offices (or FOGs). The construction of the base weights considers the probability of selection at each stage. Sampling at the office and district level included the breakout of non-certainty and certainty districts and a calculation based on the probability of inclusion from each type.
In order to complete the non-response analysis and compute final weights using records data, it was necessary that all considered factors be non-missing for all sampled parolees. Missing information for the analysis and final weights was imputed using nearest neighbor hot-deck imputation for Race (0.13 percent), Age (0.03 percent), Years in Prison (0.06 percent), Years since Release (1.36 percent), and Number of Monthly Parole Contacts (9.03 percent). In order to produce the final weights, the weights given to sampled parolees who did not respond to the survey were distributed to responding cases. A response propensity model was constructed and used to form adjustment cells within which the weights were reallocated.
A sampled individual's response propensity is the predicted probability that the parolee will complete an interview, calculated from a logistic regression model. That model is based on the data from all eligible parolees and its outcome variable is whether the individual completed the interview. The final list of explanatory variables was selected after reviewing a list of variables for possible association with response. The variables used in the non-response adjustments were race, age, years in prison, years since release, and required office contacts per month.
The predicted probabilities obtained from the model were used to construct five non-response groupings, based on the quintiles of the probability distribution. Computing the non-response adjustment using groups rather than the estimated propensities avoids the creation of extremely large or small weights. The subsequent weight adjustment was done within each group, so that the weights of non-responding individuals within a range of propensities were allocated to like individuals.
As final steps, the weights were trimmed and re-calibrated. An examination of the non-response adjusted weights revealed a small number of weights that were undesirably large compared to the overall distribution of the weights. Weights were subsequently trimmed, treating the median of the weights plus six times the inter-quartile range as a reasonable threshold for trimming. The final step in the weighting process adjusted the weights to align with known control totals. All states in the sample frame were divided into seven strata, and control totals for each stratum were produced by BJS based on mid-year 2008 parolee counts.
Mode of Data Collection:
audio computer-assisted self interview (ACASI),
computer-assisted personal interview (CAPI)
After the elimination of ineligible cases, the overall response rate was 60.6 percent.
Extent of Processing: ICPSR data undergo a confidentiality review and are altered when necessary to limit the risk of
disclosure. ICPSR also routinely creates ready-to-go data files along with setups in the major
statistical software formats as well as standard codebooks to accompany the data. In addition to
these procedures, ICPSR performed the following processing steps for this data collection:
Standardized missing values.
Checked for undocumented or out-of-range codes.