State prison populations over time
Time series data about states’ prison populations over time were collected manually through scraping Departments of Corrections websites, as well as direct requests to state officials through public record requests (for example, Freedom of Information Act requests and so on). For every state in our dataset, we sought the most temporally resolved data as possible. We collected population data at either weekly, monthly, quarterly or, for some states, yearly levels. The most common form of data we were able to collect is the number of people incarcerated at a given time in a given state, on a monthly timescale. In Supplementary Table 1, we link to the data source for every state in our dataset, and in Supplementary Section 1, we show how the prison population of every state has changed over time.
We compared the data collected here to data from other organizations that report statistics about the US prison population—the Bureau of Justice Statistics and the Vera Institute for Justice44—and find high overlap between all three of the datasets. In Supplementary Section 3, we identify areas in which our data differ from those of the Bureau of Justice Statistics, and we offer an explanation for why we are confident in accuracy of our approach (for example, in several cases, we received the data directly from the states’ Departments of Corrections, through public records requests).
For every state in this dataset, the total prison population includes both male and female incarcerated people (something that is not always the case in studies about the US carceral system, which so often focuses on male incarcerated people). In Alaska, New Mexico, Vermont and California, “Transgender”, “Other” or “Non-Binary” are also listed as gender categories, although this practice is not widely adopted in reporting statistics about the incarcerated population. In 27 states, incarcerated race statistics are separated by “male”, “female” and “total”, and further characterizing the interaction between race and sex in biases in admissions and releases during the COVID-19 pandemic remains future work.
State policy data
Court closures and reduced admissions
Qualitative data on the closure and reopening of all 50 state court systems were collected primarily through the administrative orders and/or press releases of each state system’s Supreme or Superior Court or chief judicial officer as well as through local news coverage. Most states suspended all in-person proceedings with the exception of limited emergency matters between 12 March and 20 March 2020. Several states that adopted policies early in this period issued increasingly strict guidance as the pandemic worsened. New Jersey, for example, suspended new trials on 12 March and issued a 2-week suspension on municipal court proceedings on 14 March before finally suspending all proceedings (with emergency exceptions) on 15 March. In addition to closing judicial buildings and suspending proceedings, most court closures also extended statute of limitations and filing deadlines owing to pandemic disruption. A handful of states, Pennsylvania and Texas among them, permitted or encouraged courts to begin conducting remote proceedings in their initial closure orders, although the adoption of remote proceedings was not widespread in this initial lockdown stage.
Court reopening policies were more heterogeneous than the initial closures, although trials remained suspended in most states through at least early summer 2020 (and in most cases substantially later). The earliest such policies appeared at the beginning of April 2020, with most aimed at giving regional and local judges discretion to begin hearing proceedings remotely (for example, Louisiana, Massachusetts, Florida and Iowa, among others). A substantially larger group of states adopted reopening guidelines between late April and mid May, many of which allowed essential judicial staff to return to offices following new public health guidance while also maintaining remote proceedings and expanding the number of non-trial proceedings that courts could conduct remotely. Further reopenings and the resumption of limited in-person proceedings took place in many states throughout June, July and August 2020, although trial proceedings remained suspended. Notably, several states, especially those that adopted phased reopening plans, restricted in-person proceedings and further delayed trial resumption with the autumn–winter 2020–2021 COVID-19 surge. In many states, most administrative orders restricting court operations have at the time of publishing been rescinded, although others, California notably among them, still retain certain accommodations including the option for remote proceedings.
Release policy data
Data on COVID-19 release policies, when they existed, were collected from states’ individual corrections and prison bureau systems, governors’ executive orders and local news coverage. Fifteen states did not adopt any official release policy, although our data nevertheless show that there were still reductions in the overall prison population during the pandemic in all of these states. The remaining 35 states adopted policies with varying degrees of specificity and effectiveness, although many overlapped in their broadest contours, allowing consideration for early release to be granted to incarcerated people at increased public health risk (either due to age or underlying health condition) and for those nearing parole and/or the end of their prison sentences.
Almost all states with such policies did, however, adopt a restriction preventing the release of those incarcerated for violent crimes or sex offences. North Dakota was an outlier in this regard. Of the 120 people the state initially released from prison in March 2020, 14 were serving time for violent crime convictions and 11 were convicted of sex offences. New York’s release policy was notably more restrictive (on paper at least) than that of many other states—only those incarcerated for “non-criminal technical parole violations” were eligible for COVID release. As an example of one state’s release policy, we include below an excerpt from the Virginia Department of Corrections’ policy on releases45, from 24 April 2020.
“The Director of the Department of Corrections is authorized to consider early release for individuals with less than one year left to serve while the COVID-19 emergency declaration is in effect. Offenders convicted of a Class 1 felony or a sexually violent offense are not eligible for consideration. The exact number of individuals eligible for early release consideration will change depending on the length of the emergency declaration order. The [Department of Corrections] will identify those that are eligible for consideration using the procedures it has developed to ensure public safety and will notify offenders who are to be released under the early release plan. A diagnosis of COVID-19 is not a release factor.
The following Early Release Criteria will be utilized in considering an incarcerated person for early release pursuant to legislation:
Release Date: The inmate’s Good Time Release Date must be calculated and verified in order for the incarcerated to be considered.
Inmate Medical Condition: The inmate’s medical condition will be considered.
Offense History: By legislative mandate, early release does not apply to inmates convicted of a Class 1 felony or a sexually violent offense. Consideration for early release will be based on the seriousness of the current offense, in descending order as follows: Non-violent Offense, Felony Weapons Offenses, Involuntary Manslaughter, Voluntary Manslaughter, Robbery, Felony Assault, Abduction, Murder, Sex Offense.
Viable Home Plan: The incarcerated person must have a documented approved home plan to be considered.
Good Time Earning Level: The inmate’s current good time earning level must be I or II to be considered.
No Active Detainers: Inmates must have no active detainer to be considered.
No Sexually Violent Predator Predicate Offenses: Inmates convicted of one or more sexually violent offenses established in §37.2-903 of the Code of Virginia are not eligible pursuant to legislation.
Recidivism Risk: Inmates must have a risk of recidivism of medium (5-7) or low (1-4), as identified by the validated COMPAS instrument, to be considered.”
Note especially the inclusion of the COMPAS risk assessment tool, which is used in court systems across the USA as a way of quantifying an offender’s likelihood of reoffending (recidivism). Over the past several years, we have seen a growing body of scholarly work devoted to identifying problematic and harmful racial and economic biases that arise when algorithmic risk assessment tools are used in practice46,47,48,49,50,51. COMPAS, in particular, has been the subject of a number of studies that take a critical look at the effectiveness—and ethics—of these risk assessment tools in the justice system47,52; in one study, COMPAS was found to predict recidivism 61% of the time, but at the same time, Black people were almost twice as likely to be labelled as high risk for reoffending but not actually reoffend52.
Further research is needed to quantify demographic patterns in the incarcerated individuals who were released across different states, and because there was such high heterogeneity in different states’ policies, it remains an open question whether we will see the same broad, systematic racial differences among the people who were released. However, as has been the case throughout the COVID-19 pandemic, the heterogeneity of policy responses across localities has typically had detrimental effects on our collective response to the pandemic53.
Study definitions of race and ethnicity
The data that we collected for the study used definitions of racial and ethnic groups that were determined by the agencies that collected the data. When the authors are discussing race and ethnicity in their interpretations, they are referring to the historical categories that have social, cultural and political consequences. We use the term Latino to describe people who are otherwise described as Hispanic in many settings. We have used the term non-white in select locations, as not all states had data disaggregated into the same set of categories. Thus, for some analyses, the term non-white directly describes the available data. For a list of the race categories reported by every state in our dataset, see Supplementary Table 7.
Recent advances in medical conventions have prompted discipline-wide introspection about the ways that race and ethnicity are discussed and used in research54. This is of critical importance to health equity and racial justice, and although in this work we rely on race statistics reported by states’ Departments of Correction, future work will critically examine the differences in approaches for reporting race and ethnicity statistics of incarcerated populations. Notably, it is important to know whether a state’s statistical reports use race categories that have been self-reported by the incarcerated person or whether it is interviewer-observed, which is often the case in administrative databases. These approaches are quite different and often result in inaccuracies in measurement of racial disparities55. Last, in Supplementary Section 3.3, we introduce a dataset that contains policies from 48 states and the Federal Bureau of Prisons about whether race data of incarcerated individuals are obtained through self-report or visual-assignment from administrators or staff.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.