COVID-19 amplified racial disparities in the US criminal legal system

The criminal legal system in the USA drives an incarceration rate that is the highest on the planet, with disparities by class and race among its signature features1–3. During the first year of the coronavirus disease 2019 (COVID-19) pandemic, the number of incarcerated people in the USA decreased by at least 17%—the largest, fastest reduction in prison population in American history4. Here we ask how this reduction influenced the racial composition of US prisons and consider possible mechanisms for these dynamics. Using an original dataset curated from public sources on prison demographics across all 50 states and the District of Columbia, we show that incarcerated white people benefited disproportionately from the decrease in the US prison population and that the fraction of incarcerated Black and Latino people sharply increased. This pattern of increased racial disparity exists across prison systems in nearly every state and reverses a decade-long trend before 2020 and the onset of COVID-19, when the proportion of incarcerated white people was increasing amid declining numbers of incarcerated Black people5. Although a variety of factors underlie these trends, we find that racial inequities in average sentence length are a major contributor. Ultimately, this study reveals how disruptions caused by COVID-19 exacerbated racial inequalities in the criminal legal system, and highlights key forces that sustain mass incarceration. To advance opportunities for data-driven social science, we publicly released the data associated with this study at Zenodo6.


A.1 State-by-state breakdowns
In Figure A.1, we show the prison population over time for all 50 states, the District of Columbia, and the Federal Bureau of Prisons. In Table A.1, we give an overview of the scope of each state's data in our dataset. In Table A.2, we list every state in order of the maximum reduction in prison population, alongside the month that this decrease was observed. In Figure A.2, we reproduce state-specific versions of Figure 1B. For consistency, the data plotted have been up-or down-sampled to the quarterly level (most states in the dataset report monthly data, but several report twice-yearly and yearly).
Note that for several states in Figure A.2, we plot the percent incarcerated population who are not white, as opposed to Black. This is due to either the small number of incarcerated Black people in the state (as in Hawaii, Idaho, South Dakota, Vermont, etc.) or the absence of Black as a racial category in the state's reports (e.g. Michigan). With this state-by-state view, we see that the only state that unequivocally does not show a similar trend to Figure  1 (and in fact, shows the opposite) is Oregon; ongoing and future work will attempt to disentangle whether this is a reflection of the state's criminal justice policies, demographic patterns, reporting procedures, or any of the above. Each violin plot represents every state and federal prison system's slope of the percent incarcerated Black population curves. Violin plots depict distributions of data, such that the width of each violin corresponds to the density of points in a given range, and the lines at the top/bottom are the max/min of the underlying data. The horizontal lines in the middle of each violin plot corresponds to the means of each distribution. From 2020 to 2021, we observe a statistically significant difference in the slope from before the pandemic (two sided t-test, t(50) = 5.6251; p = 1.691162 ⇥ 10 7 ).

A.2 A nationwide trend or state-level heterogeneity?
In this section, we report raw and interpolated data for each state, the District of Columbia, and the Federal Bureau of Prisons, in order to compare time series of the percent Black population in prisons to the percent non-white population (see [4]). In some cases, the main effect observed in Figure 1B is clearly recapitulated; in others, we do not see the effect in the percent Black population, but rather we see it in the percent non-white population. The only states where we do not observe the main effect in both demographic categories are Maine, Oregon, and Wyoming.
In Figure A.3, we plot the distribution of the slopes of each state's percent incarcerated Black population data during different time windows (2017-2019, 2019-2020, 2020-2021, and 2021-2022). With few exceptions, this plot shows that the slope of the percent Black incarcerated curves increased during 2020, compared to the 2019 averages. After this, the same slopes decreased during 2021. Isolating these slopes is useful again for nationwide comparison for two reasons: 1) because it can allow us to ask questions about what we would expect to happen if we did not observe some sort of spike in the percent Black population in 2020-2021, and 2) because we can use the slope of 2019-2020 as a way of normalizing the entire time series, allowing for a simple test of whether or not we observe a spike in 2020.  Figure  3D (by slope of pre-pandemic curve), we compare the trends of the every state and federal prison system's percent of incarcerated Black population. Under this view, five states do not follow the broad nationwide trend as in Figure 1B: Maine, Maryland, Missouri, Oregon, and Wyoming. For every other state, we see the percent of incarcerated Black people increase for multiple months, starting after March 2020. The top four curves on the plot correspond to prison systems that are smaller in size (New Hampshire, North Dakota, Vermont, the District of Columbia), with demographic statistics that can be disrupted by small fluctuations in incarcerated populations. Figure A.4 normalizes each state's percent incarcerated Black population data by the best-fit line from its slope between 2019-2020. That is, we divide each state's time series of percent Black population by the corresponding value in the best-fit line from the slope from the year prior to March 2020. Visually, this bundles the states' curves to the 100% value before the pandemic (i.e., 100% of normalized pre-pandemic values). After March 2020, there are five states that do not at least briefly show a spike in their percent Black population: Maine, Maryland, Missouri, Oregon, and Wyoming.

A.2.1 Interpolating monthly population data
In [4], we have included tables of the population and demographic data for each prison system studied here. Also included in these tables is a tag about whether or not the data is raw data from the state or interpolated. For most (35) prison systems, we have raw monthly data for the entire duration of the study period. For some, we only have data at the quarterly (8) or bi-annually (4); for these 12 states, we simply do a linear interpolation on the raw demographic data and sum these columns together to arrive at a total estimate for the number of people incarcerated each month in between the quarterly/biannual dates.
As a validation, we do this same interpolation on states where we do have reliable monthly  Michigan, New Jersey, South Carolina, Tennessee, and Virginia, we only have demographic data at the yearly level. In each of these states, we have population totals at the monthly level; with these, the task becomes to estimate the counts of incarcerated people by race each month, given the population totals. In some ways, this is an easier task, since we know the overall trend in the prison population. Here, again, we do a linear interpolation between the dates without missing values, multiplied by a factor of (interpolated_sum / actual_sum). Doing this same validation on states with reliable, monthly data reporting gives us high alignment again. Lastly, we note that every combination of including or excluding states based on their reporting frequency and quality still produce the same qualitative results, which we would expect given the extensive discussion above.

A.3 Comparison across prison population datasets
Other organizations collect and report data about prison populations over time. In order to situate the data used here within a broader body of work studying U.S. prison population trends, we validate against data released by the Bureau of Justice Statistics [2] (BJS) and the Vera Institute for Justice [3] (Vera). In Figure A.10, we plot the BJS's yearly estimates of the number of people in state prisons across the United States from 2014 until 2020. We concatenate the BJS data with the Vera data to approximate a "ground truth" estimate for the prison population over time. We note several key points. First, starting in 2020, our dataset almost exactly matches the Vera dataset. Prior to 2020, our dataset reports a prison population that is approximately 1% smaller than the BJS data. After investigating what could have brought about these differences between the two datasets, we identified five states with the largest betweendataset differences (Montana, Florida, Texas, Virginia, and Ohio; see Table A.3). Because of these discrepancies, we took additional care to confirm that the data we had collected was exactly what was reported by states' Departments of Corrections websites (or sometimes, through Freedom of Information Act requests). In Section A.3.1, we describe the rationale for why we are confident in the data included in the present study, and we also directly link to the data sources used to offer transparency in the data collection process.

A.3.1 Comparison to Bureau of Justice Statistics data
To our knowledge, the scale of the data that we assembled in this work is unique among the available public datasets about states' prison populations over time. In Table A.3, we dive deeper into the discrepancies between the data used here and those that were released by the Bureau of Justice Statistics [2]. We offer explanations that reconcile why we may observe such differences, and we conclude that the data reported here is consistent with state prison population statistics reported by states' Department of Corrections.

A.3.2 Comparison with the National Corrections Reporting Program
We drew on another large, well-known dataset to validate the findings from the data we collected: the National Corrections Reporting Program (NCRP) [1]. The NCRP data we used contains detailed information about individuals incarcerated in almost every state, yearly, until December 31, 2019. In recent years, data from the NCRP has been subject to scrutiny both in terms of its coverage and completeness as well as how it reports data about the race of incarcerated people [5][6][7][8][9].
Nevertheless, these data are a cornerstone of legal and justice research in the United States, and as such, we sought to use it as a benchmark of pre-2020 data to see 1) the extent to which there is significant overlap between the two datasets and 2) if we analyzed the subset of our data that only had high overlap with the NCRP data (i.e., exclude states where there is significant disagreement between the two datasets), whether we would reproduce the main result in Figure 1B.
In Figure A.14, we plot a comparison between the NCRP data and our data for 12 states' total incarcerated population over time (NCRP data plotted in red). Immediately, we are reassured about the correspondence between these states in the two datasets. There are, however, several states where the overall trend in the prison population is similar between the two datasets, just shifted uniformly up or down (i.e. states with the same or similarshaped curves but shifted by a fixed amount). Lastly, there are several states where the NCRP data is clearly not capturing the same information that our dataset contains. These We compare the novel data presented here to data from the Bureau of Justice Statistics [2] and the Vera Institute for Justice [3], finding high data agreement during 2020 and early 2021. There are small differences between our dataset and the Bureau of Justice Statistics prior to 2020, but see the Data & Methods section and Table  A.3 for further explanation of these differences. are states that-we suggest-do not have high coverage or high data quality in the NCRP dataset or have otherwise changed their reporting protocol during the duration of their inclusion in the NCRP.
Ultimately, if we only analyze states with high overlap between NCRP and our data (a proxy for overall reporting quality: Washington, California, Nevada, Utah, Nebraska, Arizona, Colorado, Wyoming, Kansas, Texas, Iowa, Minnesota, Illinois, Indiana, Kentucky,  Tennessee, Mississippi, West Virginia, Ohio, Wisconsin, Georgia, Florida, New Jersey, New York, and South Carolina), we see the same qualitative result (see Figure A.15). We see this as a validation with multiple benefits: First, it grounds the data we have collected in a known to a well-studied companion dataset. Second, we see an opportunity to use our data to augment or help fill in states with known reporting irregularities or other issues in the NCRP dataset. While the insights from these comparisons between the two datasets were not the intended contribution of our paper, we are encouraged nonetheless that these analyses offer a roadmap for future work improving data quality in the NCRP. States with light grey labels (e.g. Alaska, Oregon, Hawaii, Idaho, etc.) are not included in the "term file" (states categorized as having more reliable data). Washington, California, Nevada, Utah, Nebraska, Arizona, Colorado, Wyoming, Kansas, Texas, Iowa, Minnesota, Illinois, Indiana, Kentucky, Tennessee, Mississippi, West Virginia, Ohio, Wisconsin, Georgia, Florida, New Jersey, New York, and South Carolina are the states with the highest degree of correspondence between the two datasets.

A.3.3 Survey of states' race reporting procedure
As briefly mentioned in the previous section, there remain several known challenges in relying on administrative data to study racial disparities. In particular, one common issue is the heterogeneity between how states and other government agencies report data on race. Broadly, there are two main approaches to collecting information about an individual's race and/or  Figure 1, based only on data from states with high overlap with NCRP data. If we reproduce our main result using only states with high overlap between the NCRP data and our own (a proxy for overall reporting quality: Washington, California, Nevada, Utah, Nebraska, Arizona, Colorado, Wyoming, Kansas, Texas, Iowa, Minnesota, Illinois, Indiana, Kentucky, Tennessee, Mississippi, West Virginia, Ohio, Wisconsin, Georgia, Florida, New Jersey, New York, and South Carolina), we see the same qualitative result. ethnicity: self-reported from the individual in question or "visual inspection" by either a clerk, law enforcement official, or other administrative staff. On the one hand, some scholars argue that external assignment of race more accurately reflects the scale at which discrimination or prejudice operates [10] and should therefore be relied upon for studying race and ethnicity. On the other hand, self-reported data about an individual's race is likely more accurate and therefore more useful for large-scale quantification [11,12]. Many researchers argue for an array of different survey questions in order to accommodate both approaches [13,14], while others have found ways of improving the accuracy of administrative race data through a variety of post hoc statistical corrections [15,16]. An Oregon Criminal Justice Commission report, for example, analyzed race/ethnicity data in their management system and discovered that the system regularly mis-labeled individuals' race, such Latino, American Indian, and Asian individuals were under-represented by up to 15% [17]-a number that can be corrected through statistical re-weighting. Table 5 of the BJS Prisoners in 2020 report emphasizes this issue and describes the steps taken to statistically adjust the data in order to estimate the underlying racial distribution in state and federal prisons [18]: providers' information systems categorize race and ethnicity in this way. In addition, these data are administrative in nature and may not reflect a prisoner's self-identification of race and ethnicity... For state prisoners, BJS calculated the ratio of the distribution of state prisoners by race and ethnicity in BJS's selfreported prisoner surveys, which use OMB categories for race and ethnicity, to the distribution of prisoners by race and ethnicity in NPS data for the year closest to the fielding of the survey. BJS then multiplied this ratio by the distribution of state prisoners' race and ethnicity using the current year's NPS. The percentage of persons self-reporting to the NPS as non-Hispanic and as two or more races was assumed to be equal to that of the self-reported prisoner survey. The final percentage distribution of race and ethnicity was multiplied by the total of sentenced state prisoners to obtain counts for each category." In other words, the BJS performs a statistical correction that attempts to correct for heterogeneity in how race is reported across states. A natural question that arises here is whether we think that the same heterogeneities in administrative race reporting are present in the dataset we introduce here. If there were systematic differences in how race is reported by state (e.g. some states may report the race of a newly incarcerated person via a self-reporting procedure during intake; other states may record this data via a staff member assigning a race during intake based on visual features), this could potentially be problematic for the main results in this work. That is, we may be systematically mis-estimating the magnitude of the observed trends in Figure 1B. Alternatively, it may be even more problematic if there are non-systematic differences in how race is reported; in this case, it could potentially require a different statistical correction performed for each state, opening the dataset up to a deserved scrutiny. Fortunately, we do not think the dataset used in this work is subject to these concerns-or if it is, the impact is minimal. The reason for this is based on newly collected data from almost every state's Department of Correction, displayed in Table A.4 1 .
In nearly every state, it is the stated policy to collect self-reported race data during admissions into prison. There are a few states with either ambiguous policy language (e.g. Minnesota, which explicitly writes "Race information may be self-identified or classified by an observer.") or policy language that is suggestive of self-report but not entirely (e.g. Massachusetts: "It is mostly self-reported, however, if the county sends a face sheet the Booking Officer will use that."). One state (Texas) explicitly referred to staff members visually assigning someone's race "...during intake, the [Texas Department of Criminal Justice] will visually determine the race of the individual." While Texas was the only state that uses this procedure for collecting race statistics, we must stress that it is the largest prison system in the country, and trends in data from Texas strongly influence national averages.
The other states' policies are based on self-reported data; we include the precise language from the policy in Table A.4. We do not know of other research with this kind of detailed state-by-state data. Note: here, we do not assume perfect compliance with the self-reporting policy (e.g. either on the administrators or the people being admitted into prison), but at A.4 Case studies: Court system, policing, and inmate release data In addition to data about states' prison populations and prison policy, we also used statespecific data about outcomes of court proceedings, crime/offense type and severity, traffic stops, and inmate releases in order to tell a broader story about the structural effects of mass incarceration during the COVID-19 pandemic.

A.4.1 Florida trial statistics data
Using Florida as a case study, we show how changes to typical court proceedings can potentially lead to new racial disparities in the prison population [19][20][21][22]. In Figure A.18, we plot monthly trial statistics from circuit criminal defendants in Florida. Prior to March 2020, an average of 14,000 defendants were disposed each month (i.e., pass through the court system and have their charges dropped, agree to a guilty plea, go to a jury trial, or go to a non-jury trial; Figure A.18A). Starting in March 2020, the number of defendants decreased sharply, reaching nearly 4,000 in May 2020, resulting in a backlog of cases ( Figure A.18B).
Between March 2020 and June 2020, more than 99% of cases did not go to trial (up from an average of approximately 97% prior to 2020). An increase in the proportion of cases that get resolved pre-trial means that a greater percent of all defendants passing through the Florida courts system will either agree to a guilty plea or see their charges dropped Here, we highlight the evolution of the ratio between r x /N x , where r x is the percent of releases who are race x and N x is the percent of the incarcerated population who are race x (in this case, we use Black and white incarcerated people for x). If this ratio is 1.0, there is a proportional number of releases as one would expect, given the demographic composition of the prison population. In the three states included here, white incarcerated people account for a larger share of releases than one would expect, given the demographic distribution of the prison population; conversely, incarcerated Black people are released at lower-than-expected rates.
entirely. Importantly, both of these can be sources of statistical bias in the resulting prison population.
First, prior research has demonstrated that Black defendants are almost 70% more likely than white defendants to receive a plea deal that involves spending time in prison [23]. Second, we show here that the percent of cases that were dismissed entirely increased from an average of 10% before 2020 up to 15% in June 2020 ( Figure A.18C), and this increase in dismissed cases is strongly correlated with the percent of non-white incarcerated individuals (lagged one month to account for time delays in sentencing; Figure A.18D). This correlation did not simply arise after March 2020-we see these same correlations between percent of dismissed cases and percent of non-white incarcerated people prior to the COVID-19 pandemic, suggesting that this relationship is potentially more general and not merely an anomaly due to the pandemic. This case study highlights multiple potential disparities that can stem from disruptions in the court system, and it emphasizes the need for more states to make similar sentencing data available to the public. In this vein, we report preliminary analyses about the race of defendants whose cases were dismissed in Figure A.19, using data compiled by the Florida Office of the State Courts Administrator (collected by court clerks) to show a relative increase in the number of white defendants among those who had cases dismissed in the early months of the pandemic. See Section A.4.1 for further discussion of these data, including limitations and future work. Having access to standardized data about the race of defendants across every state will further allow us to connect case-dismissals to prison demographic distributions. To collect the court proceedings data used in Section 2.3 [19], we manually downloaded monthly data about the statewide data on the outcomes of Circuit Criminal Defendants between January 2018 and June 2021. We summarize this dataset in Table A. 5.
In addition to summarized monthly statistics about the nature and outcomes of circuit criminal court defendants, we also requested individual -level data about the defendants that make up this aggregate data. The Office of State Court Administrator (OSCA) in Florida provided us with data from the Criminal Transaction System for 2018, 2019, 2020, and 2021; these data contain a column for defendants' race, the action taken by the court, and the date each case was decided (among many other variables) 2 . This dataset allows us to run a simple analysis: Among defendants with cases that were dismissed, between 2018-2021, what percent are recorded as white? We plot this in Figure A.19B (A.19A is a reproduction of Figure A.18C, and A.19C shows the two curves atop one another, rescaled using min-max scaling in order to highlight the timing and relative increase that both measures show after March 2020).

A.4.2 Texas offense severity data
In recent years, White/Black/Latino people have accounted a for similar percent of the total incarcerated population in Texas state prisons (33.7%, 32.6%, 33.1%, respectively); mirroring the nationwide trend in Figure 1B, the percent of incarcerated Black people increased sharply in Texas after March 2020 ( Figure A.20A). While it is difficult to point to any single cause behind this abrupt disparity, some have argued that more careful consideration must be given to racial differences in the severity of the crimes for which people are incarcerated.  That is, without more extensive data about the incarcerated population in Texas, we cannot rule out the possibility that the observed spike in the relative number of incarcerated Black people is due to a relative increase in the severity of crimes committed, by race. We show, however, that this is not the case.
To do this, we merge data from two sources. The first is monthly data about every incarcerated person in Texas from July 2019 until November 2021 (the Texas Department of Criminal Justice "High Value Dataset" series, from https://www.tdcj.texas.gov/kss_ inside.html); each row in this dataset corresponds to an incarcerated person and includes details about the individual's race and sex, as well as sentencing information. The second dataset is a table that maps every offense to one of four severity levels: low, moderate, high, or highest (https://www.tdcj.texas.gov/bpp/parole_guidelines/Offense_Severity_ Class.pdf).
We assign an offense-severity category to each incarcerated person by merging the two datasets along the National Crime Information Center (NCIC) code for the sentenced offense. In Figure A.21, we compare these offense-severity categories by race. In Figure A.21A-D, we plot the race distribution within each offense-severity category (i.e., we plot the percent of incarcerated persons with a given offense-severity category who are White/Black/Latino). To accentuate the relative changes in these trends, in Figure A.21E-H, we plot the same curves standardized (i.e., divided by) by their pre-pandemic averages; in these subplots, 100% would indicate no difference from before the pandemic. After March 2020, we see abrupt increases in the relative number of incarcerated Black people in the "low" and "moderate" offense-severity groups (dark green curve, Figure A.21E-F). This is exactly counter to the suggestion that the nationwide trends in Figure 1B are due to Black people committing more severe crimes during the early months of the pandemic. Without these same datasets for every state, we cannot yet say that trends seen in Texas are universal across the United States, but following these analyses, we urge every state to make these types of data available.

A.4.3 Arkansas eligible release data
In Figure A.23, we compare the demographics of the prison population in Arkansas in May 2020 to the demographics of incarcerated people eligible for early release under Governor Hutchinson's authorization [24]; despite the fact that 57.2% of the Arkansas prison population was white, over 72% of the incarcerated people eligible for early releases were white-a disparity that we would not expect to see in a prison system absent of release policies that favored incarcerated white people. These outcomes may manifest in multiple ways: sentencing patterns that create longer sentences for incarcerated Black and Latino individuals, different classifications (e.g., violent or nonviolent) and other categorizations that may drive a disparity in those released.
In Figure A.17 we show that while, on average, incarcerated Black people are released at disproportionately low rates (and incarcerated white people are released at disproportionately high rates), the effect of COVID-19 based releases can temporarily bring about release patterns that are less statistically skewed by race. We see this in Massachusetts especially in May and November of 2020, a sign that in at least one state, it is possible to decarcerate in a way that does not exacerbate existing racial inequalities. A key insight into why this might produce more equitable releases-to-incarcerated ratios in general has to deal with differences in the average length of prison sentence, by race (sentence length, here, we consider to be a proxy-albeit an imperfect one, see [25]-for "time served" in prison). That is, if there are systematic differences in the length of prison sentences by race (e.g. if, on average, Black people in prisons are more likely to be sentenced to longer sentences than white people, which we show in Figure A.22 for Texas), a relative increase in the number of monthly releases would decrease the overall average sentence length, provided that the release criteria is based in part on people who are close to the end of their sentence. While there are not particularly noticeable changes after March 2020 in the trends of both curves in Figure A.17C, here we directly quantify racial disparities among the inmates who were eligible for release [24]. Using data released by the Arkansas Department of Corrections, we joined inmate identification numbers to their listed race and sex according to the Arkansas Department of Corrections Inmate Search tool (https://apps.ark.org/ inmate_info/index.php).

A.5 Citation diversity statement
Recent work has quantified bias in citation practices across various scientific fields; namely, women and other minority scientists are often cited at a rate that is not proportional to their contributions to the field [26][27][28][29][30][31][32][33]. In this work, we aim to be proactive about the research we reference in a way that corresponds to the diversity of scholarship in this field. To evaluate gender bias in the references used here, we obtained the gender of the first/last authors of the papers cited here through either 1) the gender pronouns used to refer to them in articles or biographies or 2) if none were available, we used a database of common namegender combinations across a variety of languages and ethnicities. By this measure (excluding citations to datasets/organizations, citations included in this section, and self-citations to the first/last authors of this manuscript), our references contain 28% woman(first)-woman(last), 22% woman-man, 12% man-woman, 12% man-man, 0% nonbinary, 12% man solo-author, and 14% woman solo-author. This method is limited in that an author's pronouns may not be consistent across time or environment, and no database of common name-gender pairings is complete or fully accurate.  Because there were slight discrepancies between the data collected in the current study and the yearly data released by the BJS, we include this table, which links to the data sources used. These data were collected from states' Departments of Corrections websites (or obtained through Freedom of Information Act requests). For both of these states, the BJS includes a disclaimer. Montana: "Data for 2019 are not comparable to data for previous years." Ohio: "Includes a small number of incarcerated individuals sentenced to one year or less." For Texas and Virginia, we include the link to directly access data reported by the state for the time period in question, and in each case, the BJS data does not correspond to data reported by the state. The discrepancies in Florida's numbers are relatively small compared to the overall number of people in prison. As stated above, we are confident in the data collection here, which involved making successful public records requests to the Department of Corrections.

State Method
Source Relevant language AL ⇥ ⇥ ⇥ AK self-report emailed Race is self-identified by the inmate. AZ self-report emailed The data is self-reported/self-identifying. AR self-report emailed During the intake process, individuals self-report their race to staff. CA self-report document Ethnicity is self-reported by offenders who choose from a list of 28 ethnicity types. CO self-report document Data from the DOC were used... because it contained self-reported ethnicity for offenders. CT self-report phone Returned phone call to confirm the policy is self-report. DE self-report ‡ document e.g. "Delaware's Level IV population identified as..." DC ⇥ ⇥ ⇥ FBOP self-report document Race/ethnic categories are presented as separate... individuals can select multiple racial categories. FL self-report phone Returned phone call to confirm the policy is self-report. GA self-report emailed The data is self-reported at admission. HA self-report emailed Ethnicity is self-reported; PSD does not verify the accuracy of the information given. ID self-report emailed It is self-report within the pre-sentence investigation process. IL self-report emailed Self-report. IDOC does not verify the information. IN self-report emailed The offenders self-identify at the time of entry into our pre-determined categories (not free form). IA self-report emailed We collect from inmate self-identifying and court documents. KS self-report emailed Demographic info, including race, is collected as part of the intake process. Individuals self-identify. KY self-report phone Returned phone call to confirm the policy is self-report. LA ⇥ ⇥ ⇥ ME self-report emailed This information is self-reported by the resident during intake. MD self-report* emailed Both self-report and staff-assigned. MA self-report* phone It is mostly self-reported, but if the county sends a face sheet the Booking Officer will use that. MI self-report emailed Each inmate self-identifies to staff for the sake of collecting data that is used in statistical reports. MN self-report* document Race information may be self-identified or classified by an observer. MS self-report emailed Race is self-reported during the intake process. MO self-report emailed The offender is self-identifying. MT self-report emailed Self-reported, unless the individual refuses...
[then] race would be estimated by staff during intake. NE self-report emailed The race data that we use for our statistical reports is self-reported by the inmate at intake. NV self-report ‡ document Demographic information is collected during intake... data are gathered from inmates progressively. NH self-report emailed The residents do self report upon intake. Thank you for your inquiry and have a great day. NJ self-report emailed Race and ethnicity are verified through self-report at intake. NM self-report emailed The inmates self-identify. NY self-report document The information about racial and ethnic origin is self-reported. NC self-report emailed Race is self-identified via a questionnaire that is administered during the entrance interview. ND self-report emailed Race information is self-reported. OH self-report ‡ document Information collected from incarcerated individuals: Race/ethnic origin, nationality, age... OK self-report emailed Yes, it's self-report by the inmate. OR self-report ‡ document HB 3064, passed August 2019, effective January 2020 PA self-report emailed The PA DOC racial data is self-reported by the inmate at intake. RI self-report document ...we cannot determine whether inmates identifying themselves as Hispanic are white or black. SC self-report document Based on inmate self-reported information at intake. SD self-report document ...collect specified demographic information (religious preference, age, race, height, weight, etc.). TN self-report emailed The information is self-reported. TX visual emailed ...during intake, the TDCJ will visually determine the race of the individual. UT self-report emailed Race is inmate self-identifying. VT self-report document Some categories are combined due to the low number of individuals who identify with these races... VA self-report emailed Virginia collects race information from several different sources, as well as being self-reported. WA self-report document Race and ethnicity are self-reported and grouped to include the following... WV self-report emailed No problem, in WV race is based on inmates self identification at intake. WI self-report emailed When an individual enters prison they are asked which race designation best describes them. WY self-report phone Returned phone call to confirm the policy is self-report. ⇤ = Nonspecific policy; ‡ = Documentation is suggestive of self-report; ⇥ = could not confirm data.