Page 311 - Special Topic Session (STS) - Volume 3

STS547 Owen A.
current population statistics system is underpinned by the decennial Census,
which is expensive and is arguably becoming increasingly unwieldy as a source
of data in a rapidly evolving society and with ever increasing demands for
more timely, relevant statistics. The system is also highly reliant on a port-
based survey to measure migrant flows to and from UK, with the result that
the intercensal population size estimates tend to have an increasing element
of bias. The ONS is therefore researching how it can transform its population
statistics system within that context (ONS, 2018), and the most important part
is the estimation of population size.
This paper will discuss the application of a number of approaches to
population size estimation, applied to an administrative based system. These
approaches are all based on capture-recapture methods which have been
used to measure undercoverage within the UK Census. The assumptions
underpinning such methods can be altered to suit different scenarios, for
instance when over-coverage becomes more of an issue. However, one does
not get something for nothing and whilst some options are less sensitive to
over-coverage, they are more sensitive to failures in other assumptions.
Firstly, the context within the UK statistical system is briefly described,
followed by a high level view of the current framework for transformation. The
presentation will discuss the different flavours of dual-system estimation in
use across National Statistical Institutes for estimating population size,
including the Bayesian approaches developed by Statistics New Zealand, and
where they differ in terms of the assumptions that underpin each application.
The research into exploring these methods in the UK context will be outlined,
with some early results.
Figure 1 shows the high level of view of transformation plans in the lead
up to and immediately after the 2021 Census. The principle is to try to produce
the best possible population statistics every year using all available sources,
whilst moving away from reliance on large scale data collection activities like
the Census. In 2022 and 2023, the quality of the statistical outputs will feed
into the decisions around whether a large-scale data collection (not
necessarily a census) will be required in 2031.

300 | I S I W S C 2 0 1 9