The University of Southampton has a long history of working with census agencies, particularly in the UK, to develop best practice in conducting censuses and to design coverage surveys and estimation procedures that improve the quality of the census outputs. The direct and indirect benefits of a high quality national census are substantial and wide-ranging.
The periodic population census is a crucial component of a nation’s statistical system, providing an opportunity to produce detailed statistical estimates nationally and locally.
These statistics have many uses, including for local planning and monitoring, and for formation and evaluation of government policy.
Traditional census data collection suffers from selective non-response by participants, and it is often groups of policy interest that are underrepresented. Appropriate designs to collect relevant data are therefore needed to generate high quality estimates.
Professor Paul Smith worked on the UK census and a framework based on capture-recapture methods, with robust ratio estimation to extrapolate from sample areas to the population.
Professor Peter van der Heijden explored the robustness of a population census based on administrative data sources – a so-called “virtual census” – as used in the Netherlands and elsewhere to estimate population totals. He explored alternative approaches to estimating the population size in an administrative census, focusing particularly on the imputation of missing usual residence information.
Professor Li-Chun Zhang’s focus has been on the over-coverage of administrative data sources, often caused by the lack of requirement for someone to deregister when they move away. He proposed an innovative method using dual system estimation, in this context through progressive trimming of the datasets, leading to the “trimmed dual system estimator”.
The Southampton team worked with the Office for National Statistics (ONS) to design the coverage survey and the methods for estimating the population size in the UK’s censuses in 2011 and 2021.
The outputs of the 2011 census were published over 2012-2013. The ONS identified £500m of benefit from these outputs, including for the following:
Statistics Netherlands undertakes a “virtual census” using administrative data. In 2016, Statistics Netherlands implemented the methods developed by van der Heijden to estimate the unregistered usually-resident population.
As well as increasing confidence among direct users of the census outputs, the census results are submitted to the European Union (EU) and affect representation in the EU’s institutions.
As the number of usual residents is typically underestimated when using only the population register, the methodology has led to more seats in the European Parliament and therefore political influence in Europe for the Netherlands.
A range of countries including the UK, Ireland and New Zealand are seeking to move from periodic large-scale censuses to a virtual census similar to the Netherlands model, based on administrative datasets.
Southampton’s trimmed dual system estimator (TDSE) has been used in the statistics offices of all three of these countries to evaluate administrative sources for use in a virtual census.
Operational deficiencies in New Zealand’s 2018 census meant that it did not provide robust estimates of the Māori population. These are critical to meet crown obligations under the Treaty of Waitangi, and because the Māori are a disadvantaged group and therefore of key policy interest, for example in health outcomes.
Van der Heijden and Smith worked with Statistics New Zealand to explore the use of linked census and administrative data to estimate the size of the Māori population. The method makes the census robust against coverage challenges, and provide useful insights into the misclassification errors of the contributing sources.
| Staff Member | Primary Position |
|---|---|
| Paul Smith | Professor of Official Statistics, Director of Southampton Statistical Sciences Research Institute (S3RI) |
| Peter G M Van der Heijden | Professor of Social Statistics |
| Li-Chun Zhang | Professor of Social Statistics |