Administrative Records for Survey Methodology 1 |
|
| Chair | Dr Asaph Young Chun (US Census Bureau ) |
| Coordinator 1 | Professor Mike Larsen (George Washington University) |
| Coordinator 2 | Dr Ingegerd Jansson (Statistics Sweden) |
| Coordinator 3 | Dr Manfred Antoni ( Institute for Employment Research) |
| Coordinator 4 | Dr Daniel Fuss (Leibniz Institute for Educational Trajectories) |
| Coordinator 5 | Dr Corinna Kleinert (Leibniz Institute for Educational Trajectories) |
Since the last census based on a complete enumeration was held in 1971, the willingness of the population in the Netherlands to participate has decreased tremendously. Statistics Netherlands found an alternative in a Virtual Census, by using available registers and surveys as alternative data sources. Advantages of a Virtual Census are that it is cheaper and more socially acceptable. The combined use of registers and surveys for composing a census, however, also leads to several methodological challenges. One of them is determining the effect of the quality of the sources. For registers, for instance, the collection and maintenance is beyond the control of the statistical institute. It is therefore important that these institutes are able to determine the quality of the sources used. Insight into the quality of the sources used enables a well thought-out comparison between comparable information in various sources.
Surveys in both Europe and the United States are facing pressure to reduce the burden imposed on respondents. One way to achieve this is to shorten the length of the survey instrument, sometimes by removing items that may be critical to the analysis. However, if the information addressed by these items can be obtained through administrative sources with minimal measurement differences (i.e., the values provided from the two sources are similar) then survey methodologists can reduce burden and minimize information loss.
Substituting administrative data for survey data is an ideal solution, so long as two criteria are met. First, the administrative data need to be consistently defined and accurately maintained across all data sources. Second, the manner in which the administrative data attributes are defined needs to be consistent (i.e., with minimal measurement error) with how the data would be defined through self-reported responses.
For prison inmates, the agency with custodial jurisdiction – a state or the federal government – maintains a large amount of administrative data on each inmate including demographics, criminal history, and sentencing. These types of data are often collected prior to the final stage of sampling and used for stratification and nonresponse models. However, if an element meets the two criteria, then the element also could be used in lieu of a survey item.
The 2016 Survey of Prison Inmates (SPI) obtained a wide range of administrative information on the inmates housed in the sample of prisons. SPI, a periodic nationally representative survey of U.S. prison inmates sponsored by the Bureau of Justice Statistics (BJS) and conducted by RTI International in 2016, collected information from inmates on a variety of topics including the basic demographics and their sentence length and date of admission for their current offense. The administrative data were collected during the rostering process prior to selecting a sample of inmates.
BJS is interested in expanding the use of administrative data in conjunction with self-reported information in the analysis of the inmate population. As such, the administrative data initially collected to assess nonresponse bias also can be used to evaluate how well administrative data may perform as a substitute for self-reported data. While some characteristics, such as age, have little likelihood of measurement error, other characteristics such as race/ethnicity and sentence length may not be consistently defined across states or by respondents.
In this presentation, we assess 1) both of the stated criteria for using administrative data in place of self-reports and 2) the consistency and quality of administrative data as they relate to inmates in prison. If these basic inmate characteristics are consistently defined and have good agreement with the self-reported data then it is possible that administrative data on other characteristics such as economic or criminal history data also may be viable substitutes for self-reported data.
Administrative records are data collected primarily for fulfilling administrative purposes by non-statistical agencies such as by Internal Revenue Service, Social Security Administration, and Selective Service System. The utility of AR has been widely discussed for decades to supplement complex surveys and censuses across the Atlantic (Chun and Scheuren, 2011). The quality of administrative records is central to making decisions about their use over the survey life cycle, from frame construction to adaptive data collection to estimation, including imputation. This paper presents a framework of Total Administrative Data Errors, discusses approaches to assessing administrative data quality and the linked data quality, respectively, and demonstrates with case studies the utility of the Total Linked Data Errors approach, including pros and cons. We discuss practical implications of using administrative records for survey methodology.