Models for Combining Aggregate-Level Administrative Data in the Absence of a Traditional Census

Abstract

Administrative data sources are an important component of population data collection and they have been used in census data production in the Nordic countries since the 1960s. A large amount of information about the population is already collected in administrative data sources by governments. However, there are some challenges to using administrative data sources to estimate population counts by age, sex, and geographical area as well as population characteristics. The main limitation with the administrative data sources is that they only collect information from a subset of the population about specific events, and this may result in either undercoverage or overcoverage of the population. Another issue with the administrative data sources is that the information may not have the same quality for all population groups. This research aims to correct an inaccurate administrative data source by combining aggregate-level administrative data with more accurate marginal distributions or two-way marginal information from an auxiliary data source and produce accurate population estimates in the absence of a traditional census. The methodology developed is applied to estimate population counts by age, sex, and local authority area in England and Wales. The administrative data source used is the Patient Register which suffers from overcoverage, particularly for people between the ages of 20 and 50.

Estimating Population Counts with Capture-Recapture Models in the Context of Erroneous Records in Linked Administrative Data

“…Accordingly, this section continues with explaining how the capture-recapture approach and the log-linear models with offsets are used in the context of estimating population counts from two linked registers which are subject to both underestimation as well as overestimation. Similar loglinear models with offsets are discussed in detail in Yildiz & Smith (2015). Hence, this section focuses more on the capture-recapture models.…”

“…To adjust the overestimation, we use the log-linear models with offsets to combine the linked Patient Register and CIS information with auxiliary marginal information. Yildiz & Smith (2015) combined auxiliary marginal information with population counts from the Patient Register. In the second step of the estimation, this paper follows their work, and uses the marginal age group, sex and age group-sex tables from the 2011 Census estimates as the auxiliary source.…”

“…However, in the future it is possible to take the marginal tables from another source, such as another administrative source, an annual survey, or a coverage survey. Moreover, in the first step of the estimation, we improve on Yildiz & Smith (2015) and combine auxiliary marginal information with population counts estimated by different capture-recapture models instead of using population counts from one administrative source.…”

Using Twitter data for demographic research

“…This methodology can be applied more broadly for demographic research using other nonrepresentative data sources thought to be characterized by measurement error. For example, Yildiz and Smith (2015) use a similar approach to reduce the bias in the England and Wales Patient Register data source, using marginal information from population census estimates as the auxiliary data source. Further applications could also include other digital populations, such as Facebook and LinkedIn, and nonrandom or convenience survey samples.…”

“…Using these estimates, combined with location information from the geo-located Tweets, we extended a calibration methodology developed by Yildiz and Smith (2015) to the framework proposed by Zagheni and Weber (2015). We compared the Twitter population by age group, sex, and location to the usual resident population of South East England, using the 2014 mid-year population estimates.…”