This paper provides an update of our current understanding of how well administrative (admin) data can produce household and family statistics, compared with past censuses.
It forms part of Stats NZ's Census Transformation research programme. We also put the work in the context of the next census.
Our focus is on household and family statistics, which rely on two inputs: a list of individuals and their usual residential addresses, and a list of related individuals and the relationships between them. Because the quality of household and family statistics depends on both, we first assess the accuracy of admin addresses and households. We then outline the method used to derive relationships between individuals and estimate the consistency and coverage of these relationships by comparison with 2023 Census data, before producing household and family statistics from the combined address and relationship information.
The methodology has limitations and may introduce biases for some subpopulations: these are discussed in the section 'Biases and limitations in our methodology and data'. Finally, we describe improvements trialled for address selection and suggest directions for future research.
Download the document below, or read the key points online.
Key points
We have used address and relationship information from admin data to derive household and family information. Key points are:
- Compared with earlier research, the consistency of admin‑derived information with the census has improved since the first investigation in 2018, but is roughly unchanged from 2023, when it was used in the third iteration of the experimental administrative population census (APC) and the 2023 Census combined model.
- Households are formed at addresses, and errors accumulate. As a result, household consistency (68.3 percent) is substantially lower than address consistency (86.7 percent).
- Most familial relationships in admin data have high consistency with the 2023 Census, but partnerships and more distant familial relationships are more difficult to capture reliably.
- Families have high consistency with the 2023 Census where they can be formed, but they inherit quality issues from both household formation and partnership identification.
- The overall quality of household and family information from admin data may not reflect all experiences of our communities and data users. Some concepts, such as children not living with their biological parents, require deeper understanding and should be explored separately and in collaboration with researchers, subject matter experts, Te Tiriti partners, and others.
- For the next census, as a baseline solution, household and family data will likely need to be based on the new census annual sample survey. This will be a large survey and, when pooled over five years, will support a wide range of geographic and demographic analyses of households and families. However, this information will be less granular than what users have been accustomed to in the past.
- Future work will focus on how admin data can be used to improve survey estimates. However, this will require the development of new methods.
- Future admin data research will focus on improving the quality of the derived addresses and relationships (for example, by exploring additional data sources for relationships), and understanding their quality and limitations. There is methodological scope to improve the quality of household and family data, such as the possibility of using addresses and relationships simultaneously to form households.
- Any future work should continue to draw on engagement already occurring, while expanding and deepening collaboration with data users, subject matter experts, Te Tiriti partners, researchers, and other government agencies. Engagement is important both for identifying where current methods fail to reflect lived realities and for helping prioritise areas where additional data sourcing or methodological development would be most valuable.
ISBN 978-1-991431-30-1
