Better population estimates with the Census

Buried amongst the numerous post-census data releases is a low profile yet critical release – the post enumeration survey (PES). From the PES, Stats NZ revises their historical population estimates. This article looks at the PES to find out who doesn’t complete their census form (and who completes it twice!) and how much your population has been revised.

After every Census, Stats NZ conducts the lower-profile but still very-important post-enumeration survey (PES), surveying a large sample of 16,500 households to understand the number and characteristics of people who don’t complete their census form. To understand the PES, we can consider three groups:

1. People who fill out their census form. Stats NZ estimate that in 2023, 87.6% of the population completed their census form, whether online or on paper. This group is fully counted across the various census variables.
2. People who don’t fill out their census form, but were counted from administrative data sources (such as tax returns or health enrolments). Stats NZ estimate that in 2023, 11.3% of the population were counted through administrative data only. This group is counted in core statistics such as population counts and age structure, but some variables are likely to be missing or imputed.
3. People who don’t fill out their census form and weren’t counted from administrative data sources. From the PES, Stats NZ estimate that this group amounts to 1.1% of the population. This group isn’t counted in census outputs at all, but Stats NZ tries to account for them in population estimates.

The decline in the census response rate has been well documented – falling from 95.1% in 2006 to 87.6% in 2023. However, growing use of administrative data means that the undercount has actually reduced – from 2.0% in 2006 to 1.1% in 2023. Reduced undercount means Stats NZ is doing better at counting everyone, but reduced response rate means the quality and coverage of the data collected has reduced. Stats NZ can only glean so much information from administrative sources - in many cases a survey like the census is the only way to gather information such as Māori descent or occupation.

Let’s focus on the third group – those who don’t complete the Census and aren’t counted through administrative data. This group makes up 1.1% of the population or 57,200 people. Focusing on this group tells us a lot about the challenges of surveying the population, and where Stats NZ needs to work hardest.

Chart 1 shows that census undercount roughly decreases with age – in other words, we got a better Census response the older you were. Children aged 0-14-years-old are undercounted by a net 2.0%, with the undercount down to 0.7% for 45-64-year-olds.

People identifying with European ethnicity have the lowest undercount, at just 0.7% of this group missed. Undercount for Māori and Pacific peoples reached 1.6%, and up to 2.3% for Asian ethnicity. The PES shows that undercount is nearly double for people born outside of New Zealand, which makes sense – people born overseas may not have been in New Zealand very long, and will have less administrative data about them than someone that is New Zealand-born.

Surprisingly, undercounting the population isn’t the only census coverage problem to consider. The 2023 census overcounted 65-74-year-olds by 0.3% on net, and 75-years-and-older by 0.2%. This means that more people were counted by the Census and administrative data collection than actually existed in New Zealand. Overcount could come about through people completing the census form multiple times, or with significantly different details to what is recorded in administrative data. For example, if someone puts a different birth date on their census form to what they give to their GP, then Stats NZ might not be able to connect the census form with administrative data on health enrolments.

Stats NZ use these insights from the PES to develop population estimates, with the aim of counting everyone. However, undercount is clearly systemic, and shows the challenge for organisations looking to serve, fund or simply understand communities with higher rate of undercount.

With the Census and PES in hand, Stats NZ revised their annual subnational population estimates in April 2025. Until Census and the PES is finalised, Stats NZ estimates the population using birth and death records, and administrative data that indicates migration of the population. For areas with a higher rate of undercount, or a lot of recent migrants (whether from overseas or the rest of New Zealand), potentially large population revisions can arrive after the Census.

Chart 2 shows that four areas – Mackenzie, Ashburton, Kaipara, and Waimate districts – had their 2023 population revised down by at least 4%. For Ashburton district, this means that they have 1,600 fewer residents than previously thought. Wellington City’s population was revised down by 2.4%, meaning that the city had 5,100 fewer residents than previously thought.

Fortunately, upward revisions weren’t quite as large as the downward revisions, though some areas still received a shock. In the most extreme case, South Wairarapa’s population was revised up by 2.1% (up 250 people).

These revisions are largely unavoidable, unless we are willing to wait five or more years for population estimates after each census. We think this is a fair trade off, as annual estimates are immensely valuable for timely recognition of shifting trends, especially after sudden shifts like natural disasters or global pandemic.

Looking at the PES and the subsequent revisions to population estimates underscores the importance of selecting the right measure.

To understand how many people live in an area, you should use population estimates, which adjust for net undercount. Population counts from the census – such as the census usually resident population – don’t account adjust for undercount. Population counts from the census are useful for understanding the characteristics of the population – such as the percentage with a post-school qualification.

Whatever you use, ensure that you are making an apples-with-apples comparison. If comparing growth over time, this should always be estimate-to-estimate (ideally) or census count-to-census count. If comparing estimated population growth against past projections, ensure that you adjust for any revisions to population estimates that have happened since the projections were made.

Stats NZ has committed to a future without a full enumeration census, using a much smaller (60,000 household) annual census attribute survey (CAS) and administrative data to provide census-like data. It’s not clear exactly how Stats NZ will replicate the PES process under this new model. The CAS may be used to understand how many people are missing in administrative data sources.

Population estimates are still likely to face revisions in subsequent years, though we may not have large infrequent revisions like we previously had from the Census and PES – these revisions will instead be spread more evenly over time as the CAS is collected every year.

If you have any questions about Census data or population estimates for your area, please get in touch.