Methodology

This project, funded by Arnold Ventures, focuses on people who are in prison past their parole eligibility. By analyzing trends over time and projecting data to 2023, the project provides key findings and state-specific metrics related to parole eligibility and release timing, which impact prison population sizes. These insights aim to inform policy decisions and corrections strategies.

Estimation Methodology

We began by limiting our population to individuals incarcerated in prison, excluding people who were returned to prison from parole and those serving a life sentence or a sentence of less than one year. Parole eligibility year (PEY) estimates for 2020 and earlier were calculated in three steps:

1. We started with the PEY variable in the NCRP for the state and year of interest. 
2. Next, we looked at the PEY provided in previous years’ reports to identify the earliest PEY recorded for a person. This can help capture PEY that is missing for people who should have one and likely had their PEY deleted or replaced with a new date after a failed hearing.
3. When PEY was missing, we used the state’s parole eligibility rules to create a conservative estimate of the PEY. We also replaced the reported PEY if our rules-based estimate was earlier than the PEY reported in the NCRP data. Because the estimate is conservative, the results may undercount the number of people in prison past their PEY.

To match people to their PEY in previous years’ reports without an NCRP identifier available across reports, we created our own identifier based on the variables that should be constant across reports: state, sex, race, admission year, admission type, age at time of admission, sentence length, maximum release year, offense (detailed), offense (general), terms count, sentences count, most serious prior sentence, last serious prior offense, and second-to-last serious prior offense. If multiple people had the same identifier, the latest PEY among them was selected, with missing PEY treated as later than any other year. If a previous report returned a PEY earlier than the one from the year of interest, that one was selected. The process was repeated for each report back until 2000.

Rules-based imputation was based on rules identified by Reitz and colleagues.¹ It is a conservative estimate computed as follows: First, each person was assigned a minimum sentence that was the highest possible percentage of the maximum sentence or a minimum number of years, based on each state’s laws. A separate calculation was made for people whose most serious offense was nonviolent, for those whose most serious offense was violent, and for those with an offense category with special eligibility restrictions. For some states that have additional special rules based on previous sentences, the imputed minimum sentence took into consideration previous sentences. For states without data about previous sentences, the calculation assumed everyone with the relevant offense type had previous sentences. For states that have special restrictions for most serious offenses that always involve sentences above certain thresholds—e.g., a felony A comes with a sentence of at least 20 years—the restricted minimum sentence was calculated for people with a maximum sentence equal to or above that threshold. All fractions were rounded up. If the rules-based imputation was earlier than the one found in the previous steps, it was used as the best estimate.

Results were then checked against available data from states about people in prison past parole eligibility and parole hearing approval rates. We also checked whether estimates and the percent of missing PEY were consistent with the state’s eligibility rules and for the potential influence of any missing data in the imputation. Data was excluded for any year in a state when there were more than 33 percent of the cases missing their admission year or the maximum sentence length variables, as well as for years when more than 40 percent were missing PEY after imputation.² Even if missing data was under this threshold, if the missingness appeared to follow a pattern that might bias the estimate, that state’s yearly data was excluded. Conversely, if a state was missing more than the abovementioned thresholds any year but due to other available data that missingness was not problematic, the year was not excluded.

Because there was a change in the trend of percentage of people past PEY in 2020 in many states, likely due to the COVID-19 pandemic, the state reports use data from 2019 if available. If a state did not have reliable data for any year after 2017, it was excluded from state reports and national analysis of disparities. However, reliable data from previous years was used for 2023 projections in the national analyses.

Our models were restricted to a subgroup of the prison population, excluding all individuals who were returned to prison from parole and people with life sentences or sentences of less than one year. Estimates for the percentage of people who had passed their PEY in 2023 were derived from a regression model based on data from 2010 to 2020. This regression accounted for state-specific trends in the annual changes to the percentage of people past their PEY over the decade, as well as the impact of special COVID-19 policies in 2020. (Technically, a mixed-effects model with random slopes for both factors was used.) The total number of people past their PEY in 2023 was calculated by multiplying the 2023 restricted subgroup of the population by the estimated percentage of people past their PEY.  

The size of the restricted subgroup of the population in 2023 was estimated using a regression model based on the percentage of the total population represented by this subgroup between 2010 and 2020. For data from 2010 to 2022, the custody population reported in the NPS was used if it matched the type of population reported in the state’s 2023 data. If the state reported the jurisdiction population instead of the custody population, the jurisdiction population was used. When the type of population reported in 2023 was unclear, it was assumed to be the one closest to the data reported in the 2022 NPS.   

Race, Ethnicity, and Sex Disparities in Incarceration Past Parole Eligibility

To estimate disparities in incarceration past parole eligibility, we utilized data from the NCRP and estimates developed by The Council of State Governments (CSG) Justice Center. These analyses focus exclusively on individuals who remained incarcerated beyond their parole eligibility date. Due to differences in how states record race and ethnicity, the analysis was limited to the following categories: White, Black, Hispanic, and Other. Data for Hispanic populations should be interpreted with caution because the original question and coding for ethnicity in state systems were unavailable and likely varied across states. This inconsistency may impact how individuals identifying as Hispanic are recorded in prison data.

We used the RRI to measure racial and ethnic disparities for incarceration past parole eligibility. The RRI compares the rates of these outcomes across different racial and ethnic groups relative to their representation in the overall population. We calculated the RRI by comparing the rates of an outcome—such as remaining in prison past parole eligibility—experienced by Black and Hispanic populations to those experienced by the White population. For example, an RRI of 10-to-1 between Black and White populations would mean that Black individuals are 10 times more likely to experience that outcome than White individuals. Conversely, an RRI of less than 1 would indicate that Black individuals experience the outcome at a lower rate than White individuals.