This paper assesses how well admin data in the Experimental administrative population census (APC) is currently able to provide ethnicity data from administrative (admin) sources compared with ethnicity data collected in the 2023 Census.
Download the paper below, or read the summary and key findings online.
Summary
Ethnicity is a core demographic variable.
Coverage of ethnicity across all levels in the APC is very high when compared with the 2023 Census.
Our evaluation compares the APC with the direct census responses (86 percent of all census values). It excludes responses from 2023 Census that were derived from historical censuses, admin data, and statistical imputation.
Individuals can have more than one ethnicity. Even though most of the population has at least one associated ethnicity in the APC, not all ethnicity combinations are captured. From this paper, it is evident that at the more detailed levels of ethnicity, there are undercounts in the APC, particularly for 'New Zealand European' at levels 2 to 4.
Ethnicity-specific data issues that affect multiple data providers are also apparent in the APC. In particular, the overcounting of 'not further defined' and 'other' group categories, along with the 'Fijian' and 'British and Irish' ethnicities is evident.
Key findings
- Coverage of ethnicity in the APC across all ethnicity levels is greater than 99 percent for level 1 and 2 ethnicities and 98 percent and 95 percent for level 3 and 4 ethnicities respectively.
- At the national level, proportions of level 1 and 2 ethnicities in the 2023 APC are similar to those in the 2023 Census.
- At levels 3 and 4, we see an undercount of many ethnic groups in admin sources compared with the 2023 Census, due to how ethnicity is being collected and classified.
- Due to high coverage, overall quality measures across all ethnicity levels are largely due to consistency between individual APC and 2023 Census responses. Consistency of admin data with 2023 Census data shows promise across all ethnicity levels, with quality ratings greater than 0.95 for all level 1 groups and overall quality ratings of 0.77 for level 2, 0.75 for level 3, and 0.71 for level 4. Key drivers of this reduced quality in lower-level ethnicities overall is the undercounting of 'New Zealand European' and corresponding overcount of 'European not further defined' categories.
This work will inform improvements to data sourcing and admin sources. Future work to improve ethnicity derivations include adding 2023 and 2018 Census responses to the APC ethnicity outputs and investigating alternative methods to derive ethnicity from multiple admin sources.
ISBN: 978-1-991307-82-8 (online)
