An Overview of Reallocation
States tread many different paths to reallocate inmate data for redistricting. Although these differences are noted in subsequent sections, it is possible to discern a general pattern for those who do the reallocation:
- The redistricting entity receives notice that it will be reallocating inmate data for redistricting purposes and begins planning for its implementation.
Four states—California, Delaware, Maryland and New York—had at least a decade’s warning that they would be required to reallocate inmate data. The remaining nine states adopted their policies sometime between 2018 and 2022; in those states, planning was constrained by the short timeline.
- The state receives the data needed to complete the reallocation.
Most states sourced this directly from the Census Bureau’s redistricting data file, often referred to as the P.L. 94-171 data, and from the data submitted by their state corrections department to the Census Bureau itself. Corrections data was often considered more authoritative than the Census Bureau’s, possibly because the bureau added statistical noise to its data to protect privacy. This disclosure avoidance system, also known as differential privacy or the “TopDown Algorithm,” is how the bureau protected the confidentiality of responses in the P.L. 94-171 data for 2020.
- The state applies its chosen methodology to conduct the reallocation.
The methods differed by state, though most sorted addresses by the relative difficulty of geocoding—or locating them to a particular point in two-dimensional space within redistricting software—on a first pass. In this process, all addresses would be run through geocoding software, then sorted based on confidence in the accuracy of the results. Addresses with less than 100% confidence would be subject to “cleaning,” a process of fixing errors like misspellings or discernible omissions like ZIP codes. To clean addresses, states used proprietary tools from redistricting software vendors, publicly available tools such as Google, or access-restricted databases like E-911. After cleaning the addresses, states set thresholds for the minimum acceptable level of confidence. Addresses below that threshold were handled in various ways according to state law. Some states randomly allocated the remaining inmate data records to the lowest possible level of geography (e.g., a city or a county); others left the population at the prison while still others excluded inmate records that could not be reallocated from their state’s redistricting data set.
- The reallocated data set is transferred to the legislature or commission for use in redistricting.
The following sections will address in more detail how states tackled each of these steps and the difficulties they faced. Included are staff recommendations on how to improve the process for future redistricting cycles.
How the Reallocation Process Transpired
- Steps Taken Before the 2020 Cycle
The key differentiator in how states prepared to complete inmate data reallocation was time: The more notice states had, the more they prepared. While the states with a full decade’s notice generally did more, staff in one of these states noted they “took for granted” the time they had to prepare and procrastinated, doing nothing until redistricting was imminent—something the staff regretted. Steps commonly taken among the three states that worked well in advance were:
- Connecting with staff who previously performed the reallocation to retain institutional knowledge gained in the previous cycle.
- Asking the state’s corrections department for an inmate data file in advance to do a mock reallocation, which provided insight into the quality of the department’s data.
- Determining whether state law would prohibit certain types of data from being shared and, if so, which clearances would be needed to access them. For example, only people with certain clearances are permitted to access E-911.
- Geocoding throughout the decade, if possible. Staff noted this depended on having resources appropriated to redistricting and a political consensus about the importance of data accuracy.
The remaining nine states had significantly less preparation time, with all laws or resolutions adopting reallocation policies passing within two years of April 1, 2020, or Census Day, as the bureau refers to the date on which enumeration begins. Four states said they lacked time to do any preparation prior to the start of redistricting. Of the five states that were able to do some preparation, two were able to complete at least one or two of the steps listed above. Other steps these states took included:
- Meeting with fellow staff and vendors to run through hypothetical challenges the states would face and what to do with them.
- Communicating with the state’s corrections department about ways to improve the quality of “last known address” data.
- Modifying existing contracts with redistricting vendors to include reallocation data, or quickly contacting other vendors if those under contract lacked that capacity. This was particularly important to staff because of their belief that the number of vendors/contractors with the ability to reallocate inmate data was small.
- Determining from other branches of government whether the staff or agency tasked with completing the reallocation had the technical capacity to carry out the policy. Staff in these states typically had strong interagency relationships that gave them access to peers’ time at little to no cost.
Two states noted they tentatively began taking these steps after the relevant laws or resolutions had been introduced but before they were enacted or adopted. Because of tense politics around reallocation, staff in these states recommended that their peers in other states avoid this strategy in the future. Their concern may be most helpful to staff in states where reallocating inmate data for redistricting is politically divisive.
- Sourcing and Receipt of Data
All 13 states sourced their data from their corrections departments. One state’s law also required the reallocation of different wards of the states, such as those housed in mental health treatment facilities, so that state sourced data from additional state agencies. Staff offered three reasons for not sourcing data from outside sources: the additional cost of using an outside data provider; concerns about the accuracy of outside data sets; and state laws mandating that only corrections department data be used.
Three states also requested data from wardens at individual prisons to compare with data received from their corrections departments. One state that received low-quality data from its corrections department reached out to the health agency that provided medical care to inmates because it had its own records, including medical contact data.
NCSL identified three conditions that could make data sourcing difficult:
- If states contracted with other states to house some of their inmates.
- If states used private prisons to house some of their inmates.
- If states contracted with county or regional jails within the state to house some of their inmates.
Only two states contracted with other states to house some of their inmates, and because their laws were clear that the reallocation policy applied only to inmates in the state on Census Day, the inmates housed out of state were not included in the inmate data reallocations. One state housed inmates from other states, but its policy applied only to its own inmates, so the out-of-state populations were excluded. In the two states that contracted with jails to house some inmates, the reallocation policies applied to those populations; in both states, data on the state populations housed in the jails on Census Day were reported for reallocation.
Several states had contracts with private prisons on Census Day 2020. NCSL identified this as a possible concern for policymakers because data gathered by private prisons might be collected via different methods than in public facilities; in addition, because inmates can be moved between public and private facilities while serving their sentences, there is a risk that some people might be counted twice. One state avoided this problem by collecting all inmate data at a single intake facility regardless of whether the person would be housed in a public or private prison. Another state asked its corrections department to cross-compare private prison data with its own to remove duplicate records.
Several staff noted that contracting with other states to house inmates could pose unique challenges to data reallocation, particularly if both states had adopted such policies. The staff cited this as a reason to build and maintain relationships with peers in other states: As more states adopt reallocation policies, the issue is likely to play a bigger role in future redistricting cycles. This is especially true if two states adopt reallocation policies, but one sends some of its prison population to the other; if both count only the state’s own inmates who are housed in-state on Census Day, the inmates housed on contract would be excluded from reallocation.
Lastly, NCSL asked states if their corrections departments did any geocoding of last known addresses on their own. If they did, the quality of data sent to staff tasked with reallocation could be of higher quality. Only three states’ corrections departments said they did any geocoding, though the type and quality differed greatly among the agencies. One state said it performed geocoding on different types of data but had not done any on last known addresses and therefore could not offer any geocoded data to the redistricting staff. The second state attempted to geocode latitude and longitude coordinates at the request of redistricting staff, but the data had significant errors and could not be used. The third state attempted geocoding in years past, but the data was so old and—in the corrections agency’s opinion—of such poor quality that it was not shared with redistricting staff.
- Completing the Reallocation
Once states received last known address data from the appropriate sources, they typically had a short time frame to complete the reallocation process before sending the data to the legislature or commission. Usually measured in days or weeks, this period is when the ability to prepare in advance bears fruit: Across the board, the states that had years to prepare reported fewer difficulties with timelines than the states that adopted reallocation policies just before redistricting began.
In the nearly three dozen interviews with policy staff, it became clear there are three key components to successfully completing the reallocation of inmate data: selecting a geocoding tool or tools; developing a process for cleaning addresses; and developing a methodology (or, as one state put it, order of operations) for completing the reallocation.
A handful of states had a fourth component: ascertaining the earliest possible date by which a prisoner could be released. These states’ reallocation policies required that inmates serving life sentences, and those guaranteed to be incarcerated beyond a certain date, be excluded from reallocation. In these states, ascertaining “time earned,” or reductions in the time before someone is eligible for parole, is critical to determining whether a minimum sentence is beyond the cutoff date in the reallocation policy.
The following subsections address these elements in greater detail.
Selecting a Geocoder
More than half a dozen geocoders are in use among the states. The most common was the tool created by the Census Bureau, though staff in several states reported difficulty using it; in fact, of the seven states that told NCSL they attempted to use the bureau’s geocoder, only one reported no issues with its user interface or concerns with its accuracy.
Other geocoders used included products from redistricting software vendors Esri and Caliper Corp.; TIGER/Line files from the Census Bureau; QGIS; Bing and Google geocoders; and geocoders developed by state demographers offices or other state agencies. Many of these geocoders had costs associated with using them, while others could process only a limited number of addresses. While most states used multiple geocoders as part of their methodology (more on this below), the Census Bureau’s tool was the only one that elicited staff concerns.
Staff sought two qualities from their geocoders: the ability to geocode an address with a high level of confidence despite having imperfect address data, and the ability to break ties between two potential geocodes of equivalent confidence. Typically, staff would gravitate toward whichever tool they could afford that could geocode with high levels of confidence and could correct minor errors in addresses, including missing ZIP codes or city names. Other geocoders would be used to check this primary geocoder’s accuracy.
Processes for Cleaning Addresses
In a perfect world, there would be no need to correct errors in, or “clean,” last known addresses received from other sources. But the addresses received in all 13 states were flawed, and staff needed to develop a system to fix these errors so the addresses could be geocoded. Three states contracted with vendors to do this for them and could not provide feedback to NCSL. In the other 10 states, staff generally followed the same pattern: They would run all the addresses received from the state’s corrections department through a geocoder, yielding “scores,” or a measure of confidence the geocoder had in the accuracy of its result. Then all results below a certain level would be set aside for staff to fix errors.
One state declined to share its cleaning process with NCSL, characterizing it as proprietary. In the other 12 states, staff used both automated and manual methods to clean addresses—frequently both. To limit the number of addresses to be cleaned manually, staff used automated systems first. While most geocoders can make use of imperfect data, the ones staff pointed out as being particularly helpful were Google, Esri and Caliper. Remained addresses would then be cleaned manually by entering them into databases that could fill in the blanks; tools used for this step included the U.S. Postal Service’s ZIP Code Lookup tool and Google Maps. One state cited concerns with the accuracy of Google Maps for this purpose and did not use it; others voiced no such concerns.
Inevitably, some addresses could not be cleaned, either because they were so incomplete there was no way to fill in the blanks with even medium confidence, or because the address was listed as “homeless” or “transient.” Those addresses were then handled according to relevant reallocation policy or law.
Developing a Methodology
While geocoding and address cleaning are part of a state’s methodology for carrying out a reallocation policy, staff also had to develop rules to make the process work.
While some states were vague about their methodology, most generally fell into two camps that we will call “confidence” and “formulaic.” Please note that NCSL is using these terms for descriptive purposes only.
The “confidence” states sorted addresses according to their geocoder’s accuracy scores and would reallocate people directly from the facility where they were incarcerated to their last known address. These states would start with their highest accuracy addresses and proceed to addresses with lower accuracy scores. Then they took all addresses below a certain accuracy threshold to policymakers or counsel to determine what to do with them. Data for inmates who had addresses determined to be impossible to geocode with any confidence would then be left at the prison location or eliminated from the redistricting data set entirely, depending on the state’s statutory or commission resolution text.
The “formulaic” states followed roughly the same process but began by subtracting all inmate data records from the redistricting data set, then returned those that could be reallocated. The states that reported using this process also were required by law to exclude from the final data set prisoners with last known addresses that could not be geocoded.
Difficulties Faced by Staff and Recommendations for the Future
It is no surprise that with a process as complex as reallocating inmate data, staff identified several problems they perceived to be “fixable.” Several of the nine problems listed below would require legislation to fix and are included for policymakers to consider as they revise existing reallocation statutes or consider the pros and cons of adopting this policy.
- Census Bureau policies complicate the process.
The bureau does not reallocate inmate data on states’ behalf; instead, it considers this to be a policy option states can pursue on their own after the bureau’s role in redistricting is finished. This is in keeping with the concept of “usual residence” as established by the Census Act of 1790. In the last decade, many comments were submitted to the Census Bureau in favor of counting prisoners at some previous address prior to incarceration, but Bureau staff could not find clear evidence that counting them at one of the many possible pre-incarceration addresses would be an improvement and better represent the concept of usual residence than current practice.
Many staff noted that the bureau’s decision to use differential privacy to protect the confidentiality of responses had consequences for prison population data, which was, by design, noisy.
Staff recommendations include:
- Requiring the Census Bureau to perform the reallocation on states’ behalf—either automatically or on request.
- Requiring the bureau to refrain from applying differential privacy to group quarters population numbers (in other words, keeping group quarters population numbers “invariant”).
- Refusal of the federal Bureau of Prisons to comply with states’ requests for inmate data.
Nine of the 13 states’ reallocation policies encompassed people incarcerated at federal prisons, yet none of those states was able to receive inmate data from the Bureau of Prisons. All nine states requested data multiple times through official and unofficial channels to no avail. While some states reported receiving no response to their requests, a few recommendations include:
- Coordinating with the executive branch in power leading up to 2030 to force the Bureau of Prisons to comply with states’ requests for data.
- Lobbying Congress to mandate that the Bureau of Prisons comply with states’ requests for data.
- Circumventing Bureau of Prisons bureaucracy and getting data directly from the federal prisons by building a rapport with wardens at individual facilities within a state.
- Amending state policies to clarify how to handle federal inmates if the Bureau of Prisons continues to withhold requested data from the states.
- No quality control on last known addresses.
Staff tasked with reapportioning inmate data reported dissatisfaction with the quality of last known address data supplied to them by their respective corrections departments. In addition to the previously discussed questions around intake at private prisons versus state-run facilities, staff noted that corrections divisions used inconsistent methods to gather and maintain last known address data. The corrections staff NCSL spoke with gave two reasons for the inconsistencies: They do not track address information on their own, apart from having a location where someone on parole might be found; and the additional labor and computer system upgrades needed to improve data gathering practices at prisons would require additional appropriations from state legislatures.
Staff recommendations include:
- Appropriating additional funds to state corrections departments to improve data gathering practices.
- Passing state laws directing corrections departments to develop a uniform system of gathering inmate data.
- Using an address predictor, like Google Maps uses, when typing in addresses during intake at prisons or other incarceration facilities.
- Prevalence of incomplete or transient inmate data
Twelve of 13 states cited this as a major issue. Transient or incomplete data almost always occurred when the incarcerated person was experiencing homelessness before his or her arrest. Addresses for these inmates may be entered as the intersection where they were arrested or the street or park where they usually stayed. Other times the address field may be completed with “transient” or “unknown,” and sometimes it is left blank. One state reported it successfully geocoded some of these addresses if there was a reference to a park or an intersection and that place was wholly within a single census block. No other states indicated they attempted to do this, with staff in most of these states indicating that addresses for people without permanent residences made up the largest portion of the data they could not geocode.
Staff recommendations include:
- Sourcing last known address data from multiple sources, such as police agencies, court records and prison records. That would give states a hierarchy in code or regulation for determining which address should be used for geocoding.
- Adopting a policy that allows for random placement of an inmate in the “best guess” location if a specific census block cannot be identified. For example, using software to randomly geocode people somewhere in a city or county where they are known to have lived. Notably, other staff recommended that inmate data be left in the census block of the facility if their data could not be geocoded, rather than requiring a “best guess” or random geocode or excluding them from the redistricting data file altogether.
- Corrections department race categories do not match the Census Bureau’s.
The Census Bureau currently uses race categories promulgated by the Office of Management and Budget. For the 2020 decennial census, these included White, Black or African American, American Indian or Alaska Native, Asian, and Native Hawaiian or Other Pacific Islander in the 2020 cycle. Additionally, a Hispanic ethnicity tag can apply to any of these categories. Only one state’s corrections department used race categories that matched the Census Bureau’s. The other 12 had significant differences. For example, several states list Hispanic as a race category. Other states did away with race altogether and asked for self-reported “ethnicity”; but, as staff in that state noted, most people confused ethnicity for nationality and simply wrote the name of a country from which they or their family had immigrated. One state reported that its race categories were tailored to a national corrections organization’s recommendations rather than the Census Bureau’s. When these race categories were combined with the statistical noise created by differential privacy, the data from corrections and the Census Bureau were so different that race data was usually disregarded for reallocation purposes. In states that specifically required the use of race data, legal workarounds were created out of necessity.
Staff had one recommendation for fixing this problem: asking policymakers to consider passing laws mandating that state agencies use Census Bureau race categories. This may require recanvassing a state’s entire prison population.
- Delays in receipt of census data hindered state efforts.
Due to the COVID-19 pandemic, the Census Bureau did not release the P.L. 94-171 redistricting data set by the April 1, 2021, deadline set in federal law. This created problems for many states because a majority have statutory or constitutional deadlines of their own to complete redistricting. Because reallocating inmate data takes time and cannot be completed until the P.L. 94-171 file is received, and because reallocation must be completed before maps can be drawn, staff in the 13 states that reallocated in this cycle operated on a particularly acute timeline.
For the states that adopted reallocation policies late in the process, staff reported the delays were a blessing in disguise; in fact, staff in three states said they believed they could not have taken the steps necessary to prepare for reallocation without these delays. This perspective was outweighed by complaints from other states that the delays impeded their ability to comply with other areas of law, such as codified dates for soliciting public input in redistricting.
Staff did not recommend solutions to this problem because the delays were a product of unique circumstances that are (hopefully) unlikely to occur in future redistricting cycles.
- Lack of clarity in statute or commission policy.
Staff in states that either reallocated inmate data for redistricting a decade ago or had nearly a decade’s notice that they would need to reallocate inmate data reported no issues with the clarity of their laws. But staff in many of the states that adopted policies in 2019-21 told NCSL their laws lacked the specificity staff needed to resolve difficulties they faced during reallocation. For example, many staff expressed discomfort with statutes or commission policies that did not specify how to handle transient or homeless addresses, or what minimum level of certainty in a geocoder’s accuracy was acceptable; staff believed policymakers should determine the acceptable level of certainty for reallocation. Other staff noted the added cost in both time and expertise required to reallocate inmate data and felt that they could have achieved a higher reallocation rate had their policy been accompanied by additional appropriations.
Staff recommendations include:
- Detailing specifically how staff should address the many issues relating to reallocation either in code or in regulations.
- Creating a statutory framework so that all problems are resolved in a similar manner.
- Revising state statutes to provide thresholds for what level of uncertainty in geocoding is acceptable to policymakers.
- Prisons in the wrong census blocks or spread across multiple blocks.
Staff in six states told NCSL there were errors with the location of prisons in the redistricting data file. Errors included prisons being in the wrong location (e.g., across the street from where they actually are), prisons being spread across multiple census blocks, and the combination of buildings for inmates and buildings for corrections staff within a single block. Such errors slowed staff down because the data they received from corrections did not always differentiate between buildings, and staff in some states reported having to randomly divide inmate populations among the different blocks to complete the reallocation—an added wrinkle they felt was easily avoidable.
Staff recommendations include:
- Participating in the Census Bureau’s block boundary suggestion project, through which a state can influence choices the bureau makes on census block boundaries and encourage the project’s bureau liaison to consult corrections staff on prison locations.
- Encouraging states to participate fully with the Census Count Review Operation, with a specific emphasis on both correctional facility locations and coordination with corrections staff.
- Gathering data on the number of beds in each building of each correctional facility in advance to know the proportions if dividing populations among buildings is once again necessary.
- Developing a positive relationship with wardens or other leaders at correctional facilities so that policy staff can go to them with questions.
- Encouraging the Census Bureau to not apply differential privacy to prison populations.
- Privacy concerns.
Most states only need three things to reallocate inmate data for redistricting: a last known address; race categories and a Hispanic ethnicity tag; and a unique identifier to avoid confusing individual records. Most states reported receiving exactly this information, though staff in four states reported having privacy concerns. One state reported they received a file from their corrections department with first and last names. A staffer reported particular concern about receiving an inmate data file with Social Security numbers and other unique identifiers. This staffer immediately deleted all records in the file and raised the issue of why such data was shared with corrections. This problem might not have been common in this redistricting cycle, but as data privacy becomes a larger issue to state policymakers, staff believe the intersection with inmate data reallocation may need to be finessed.
Policy and corrections staff recommendations include:
- Clarifying in state law or in commission resolutions what data fields must be included in the files sent by corrections to the legislature or commission.
- Having legislative counsel analyze all reallocation and data privacy statutes for potential conflicts of law.
- General Recommendations.
- Many staff recommendations concerned the reallocation process itself and were not attempts to resolve problems staff addressed during implementation. These are the recommendations staff specifically asked NCSL to include in the report:
- Policymakers should give corrections staff more input in the process of drafting reallocation legislation because their perspective may help avoid unseen problems.
- Staff should document every step in the reallocation process because states are almost guaranteed to be sued during redistricting; having a well-documented process to justify actions staff took can be invaluable.
- If policymakers are interested in adopting new policies for the 2030 cycle, they should do so now; waiting will lead to many states repeating the problems staff encountered in 2020.
- Policymakers in states with commissions should consider hiring staff earlier in the decade to begin doing preparatory work for reallocation (and everything else in the redistricting process).
- Policymakers should consider revising their laws to allow staff to begin the inmate data reallocation process before receiving the P.L. 94-171 data file.
- Staff should consider building connections with one another before redistricting to create a network for troubleshooting issues during the crunch time of reallocation.
- Staff should test different geocoders in advance to see which one works best for their state; not all states will come to the same conclusions.
- Staff and policymakers should consider the advantages and disadvantages of outsourcing geocoding to a contractor, keeping in mind there may be privilege benefits. Ask your state’s counsel.