Page 315 - Special Topic Session (STS) - Volume 3
P. 315
STS547 Owen A.
susceptible to failures in other assumptions (e.g. homogeneity bias) as in some
extreme cases the under-coverage rate can be quite large. Work is ongoing to
explore this further in Ireland and other countries. ONS have done some early
work on such ideas for both Census and Administrative data contexts (ONS,
2014).
2.5 Bayesian Approach
Statistics New Zealand (SNZ) have taken a Bayesian approach to this as
described by Bryant and Graham (2015). They are developing a Bayesian
hierarchical model, where the observed data comes from either administrative
data, or from a coverage survey where the two are linked at individual level.
This model seeks to infer from the observed data people’s “true” location
in a population-administrative data union, allowing for both under and over-
coverage. The key input is the use of prior information that provides
reasonable bounds on the total population size N (i.e. it’s likely to be similar
to previous size estimates), and to the coverage probabilities of the
administrative data (e.g. we might expect that some age groups are more likely
to interact with admin data than others). ONS are currently experimenting with
this approach (jointly with SNZ) to see how strong the prior information needs
to be in order for the estimates to be unbiased, and how robust the approach
is to failure in other assumptions.
3. Results
To highlight the different performance of the versions of DSE, specifically
the classic DSE and the weighting class estimator, ONS (2017) showed the
results of applying these methods to a linked administrative data based
‘statistical population dataset’. Figure 11 (reproduced from ONS (2017)) shows
the results for males by five-year age group when compared to the gold-
standard 2011 Census population estimates. It shows that the DSE suffers from
failures in the over-coverage assumption, and the Weighting class estimate
(WCE) is negatively biased due to within household non-response.
304 | I S I W S C 2 0 1 9