Page 319 - Special Topic Session (STS) - Volume 3
P. 319
STS547 John D. et al.
2015; Statistics New Zealand, 2017). As part of this effort both countries are
exploring and evaluating new methods for the compilation of population
estimates. While undertaken separately, there are similarities in the respective
approaches. Projects in both countries involve first creating a Statistical
Population Dataset (SPD) to represent the population and then adjusting for
coverage errors.
In the Irish PECADO project (Population Estimates Compiled from
Administrative Data Only), the SPD is built using a Signs of Life (SoL) approach
and as such by design does not suffer from overcoverage. Ireland does have
an official person identification number (PIN) used across Public Services, and
linkage based on this number eliminates linkage error. DSE methods are used
to adjust for undercoverage using a separate administrative data list.
Reassurance with respect to the non existence of overcoverage is further
provided through an application of an extension of DSE methods called
Trimmed Dual System Estimation (TDSE). The TDSE approach was first
proposed by Zhang and Dunne (2018) and we will refer to this method as the
Zhang and Dunne method.
Statistics New Zealand build their SPD using a rules based approach to
determine if someone belongs to the population or not. The rules are typically
based on activity on administrative sources in the two year period up to and
including the reference period. In this context the SPD is expected to contain
both undercoverage and overcoverage errors. Statistics New Zealand have
developed and are evaluating a methodology to adjust SPD counts to obtain
population estimates. Their approach is based on a Bayesian model and
requires a second sample to be undertaken in the field. The approach is
proposed by Graham and Lin (2019) and we will call the Graham and Lin
method.
As different conditions and assumptions underpin each approach the
methods are not directly comparable. However, we consider their different
strengths and weaknesses. This short paper has two more sections. The next
section describes the two methods and in the final section we consider their
strengths and weaknesses.
2. Methods Illustrated
2.1 The Irish PECADO Project - Zhang and Dunne Method
We take as our starting point the DSE model proposed by Zhang and
Dunne (2018), and following this approach we develop a DSE estimator for the
population size as = / where is the size of list , is the size of list
̂
and is the size of the match between the two lists. The assumptions
underpinning this DSE model are i) no erroneous records in either list or list
; ii) no linkage error when matching records between list and list and iii)
308 | I S I W S C 2 0 1 9