U.S. Department of Justice Office of Justice Programs Presented by William Sabol/Ann Carson Director/Statistician Bureau of Justice Statistics NCRP Data Providers Meeting March 31, 2015 Proposal to expand linking NCRP data to other administrative records to improve understanding of correctional populations BUREAU OF JUSTICE STATISTICS
Utility of linking NCRP data demonstrated by BJS/Abt current efforts to link NCRP data Linking NCRP records within persons over time has led to new, offender based measures of recidivism William Rhodes, et al., Following Incarceration, Most Released Offenders Never Return to Prison, Crime & Delinquency, Sept. 29, 2014, pp. 1 23. Longitudinal analysis of each individual entering prison; follows the history of that individual across events, such as a single or several entries and exits from prison. Measures: Probability of 0, 1, 2,, N returns to prison within a period Burden placed on prison system 2
Recidivism: Prison returns (longitudinal) 0.8 0.7 Estimated Probability 0.6 0.5 0.4 0.3 0.2 0.1 Average North Carolina New York South Carolina California 0 0 1 2 3 4 5 6+ Prison returns over a twelve year period following release
Recidivism: Prison burden Of all the bed days that all prisoners spend during an observation window, what is the distribution for the amount of time that offenders spend in prison? Are there a few offenders who occupy a large proportion of all of the beddays over the observation window? Over a 12 year observation window: First 25% of prisoners accounted for 2% of prison bed days (prison utilization) Second 25% of prisoners accounted for 6 % of prison bed days Third 25% of prisoners accounted for 17% of prison bed days Last 25% of prisoners accounted for 72% of prison bed days
Utility of NCRP can be enhanced by linking to administrative data from other domains BJS proposal: Explore linking NCRP with other administrative datasets for the purposes of understanding income, employment, mortality, public assistance histories of prisoners: Pre and post prison income Post prison mortality Public assistance receipt (TANF) Social security benefits and potentially of their families: Effects of removal due to incarceration on family income, mobility, public assistance and relating these to recidivism as measured by return to prison. 5
Utility of NCRP can be enhanced by linking to administrative data from other domains BJS s prisoner release cohort recidivism studies: Linked to State criminal history records through Triple I (the Interstate Identification Index) Ex prisoner employment: Corrections records linked to unemployment insurance (UI) wage data Tracking post prison management of HIV/AIDS (Josiah Rich/Abt): Linked corrections to Ryan White clinical data (RI, NC, TX, MA). BJS s Deaths in Custody Reporting Program data and the National Death Index: Up to 10 ICD codes on cause of death; study of intoxication death, smoking related mortality 6
Sources of other administrative data: Each individual state s agencies that have responsibility for various domains; e.g., depts. of jobs/employment; health; various welfare departments; vital statistics; Requires individual MOUs Comparability and cost issues National data systems, such as SSA, IRS, HUD, CMS U.S. Census Bureau s Center for Administrative Records Research & Applications (CARRA) 7
U.S. Census Bureau s CARRA group Part of the U.S. Census Bureau s Center for Economic Studies Has obtained data from (and is seeking additional data): Supplemental Security Income recipient record (SSA) Medicare enrollee data (Center for Medicare and Medicaid Services) Change of address file (USPS) Temporary Assistance for Needy Families (TANF) Public housing and rental assistance history (HUD), HUD insured mortgage program Death Master File (SSA) Decennial Census, American Community Survey (Census) Patient registration file (Indian Health Service) Tax returns (IRS, requires additional permissions) Census Longitudinal Employer Household Dynamics (LEHD): State UI wage data: Separate data enclave 8
CARRA and the Numident file Contains all SSNs ever assigned by SSA Associated with each SSN: Individual s full name Date of birth Sex Race Parents first and last names State or country of birth Numident is updated when individuals change their information (name changes are most common) 9
CARRA s record linkage efforts Uniquely identifies persons in the various datasets; heavy reliance on the SSA Numident file; Assigns each uniquely identified person a Personal Identification Key (PIK) Associates individuals with other persons living at an address; allows for the capacity to track outcomes for the family members of an incarcerated person 2010 Census Match Study: Designed to study costs of enumeration and determine whether admin records use could reduce costs 10
CARRA s record linkage efforts: 2010 Census Match Study Matched persons and addresses in administrative data to evaluate the quality and coverage of administrative data Addresses: 131.7 million addresses in the 2010 Census; 151.3 million addresses in admin recs. Of the 2010 Census addresses, admin recs matched to 122.0 million or 92.6 percent; Persons 308.7 million persons in the 2010 Census, and 279.2 million were assigned a protected identification key. 312.2 million unique persons in administrative records that were assigned a protected identification key and were alive on Census Day, April 1, 2010. Of persons with a PIK, admin recs matched 98% of persons (274 MN) in the 2010 Census Of all persons in the 2010 Census universe, including records lacking protected identification keys, admin recs matched to 88.6%. 11
BJS s proposal Conduct a pilot study using data from NCRP to develop, test, and validate record linkage with Census CARRA data Conduct the study behind Census firewall, so that PII will be protected Generate test statistics such as pre and post prison income, use of Medicaid, receipt of SSA, and other measures. Report results back to NCRP providers If successful, create a system that allows researchers to access the linked data through Census AFTER PRIOR APPROVAL each State s DOC research review board. 12
Why we think this will work: NCRP contains data elements that allow for record linkage BJS began requesting inmates first and last names through NCRP in 2009 Term record construction and backfill submissions have produced named records for almost half of the states back to 2000 In 2013, 42 out of 44 states provided inmates names on the prison release ( B ) and yearend stock ( D ) files In 2013, BJS asked Abt Associates to survey NCRP respondents about the feasibility of obtaining 9 digit SSN Only ~15 states said that it was unlikely or impossible for them to provide SSNs because of state statutes/department rules BJS is asking for OMB clearance to request SSN through NCRP in 2016. In 2013, 4 states provided 9 digit SSN on B and D files 13
NCRP term records: Linked admits/releases within persons State 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming
How the linkages would be done Census matches NCRP records and assigns each person a PIK; the PIK transfers to NCRP Identification variables (names, SSN) on PIK ed NCRP file are destroyed Only approved staff from BJS and CARRA will have access to the PIK ed NCRP file PIK ed NCRP inmates can then be linked to the PIK in other datasets 15
Proposed pilot test NCRP data linkage BJS proposes to use the 4 states that currently provide SSNs to NCRP on B records as test cases for data linkage SSN generally considered to provide a better match than names BJS understands the data quality problems of inmates SSNs Pilot study will match the records using names alone, SSN alone, and the combination of names and SSN Depending upon the success in the use of names, etc., without SSN, other states data could enter the study. Match will be between NCRP and Social Security Administration s Numident file behind U.S. Census Bureau s firewall 16
Research questions to be addressed in the pilot study What is the distribution of inmates income, pre and post prison? If individual is a member of a family, what happens to family income upon an offender s removal through imprisonment? What is the distribution of inmates receipt of SSA benefits, pre and post incarceration? What is the death rate of former prisoners over time? What is the household structure of former prisoners before and after they are released? What does prison release do to household income and use of federal benefits? 17
General plan for the NCRP CARRA pilot Approach: Data to CARRA; Census performs match; matched records assessed for bias; BJS seeks permission from IRS to use tax return data; BJS supports analysts to create analytic files from available datasets; Analyses of outcomes: Report back to ICRN/NCRP data providers Opportunities for State DOC researchers BJS Visiting Fellowship Program (FY2015) Intergovernmental Personnel Act (IPA) Transfers 18
Other datasets that could potentially be linked to NCRP National Death Index (NCHS/CDC) Would allow for examination of cause of death of former prisoners CMS Virtual Research Data Center (VRDC) Would allow BJS to look at Medicaid enrollment by former prisoners, both before and after Medicaid expansion under the Affordable Care Act Claims data could provide an understanding of some of the health conditions experienced by former prisoners State specific datasets Models for this are the LINCS study (following HIV testing and treatment of former inmates in TX, NC, RI) and linkage of data from state mental health hospitals in Michigan 19
Other datasets that could potentially be linked to NCRP U.S. Department of Veterans Affairs Would allow for more frequent research on veterans in state prisons, data could be passed back to the states as validation of self reported veteran status State specific datasets Models for this are the LINCS study (following HIV testing and treatment of former inmates in TX, NC, RI) and linkage of data from state mental health hospitals in Michigan 20
Status of efforts to link NCRP with other data Data source Status CARRA BJS executed and funded an IAA with CARRA Seeking permission from states with SSNs to do pilot tests on linkage quality National Death Index BJS has applied and gained permission to link DCRP data with NDI Waiting on funding CMS Virtual Research Data Center BJS reviewing access and applicability Department of Veterans Affairs Ongoing discussions with VA on feasibility of non SSN match LINCS study Studies ongoing in TX, NC, RI Researchers need to get permission from other states DOCs prior to linkage Michigan MH hospital study Interest by NRI (Lynda Zeller) in linking NCRP to state mental health hospital data. 21