Page 324 - Special Topic Session (STS) - Volume 3

STS547 John D. et al.
erroneous records (overcoverage), the coverage list must be capable of being
linked in an error free way to the SPD and information with respect to the
inclusion probabilities of units of the population in the coverage list is
required. The Zhang and Dunne method operates under the assumption of
homogeneous capture and blocking can be deployed where there is a
suspicion or knowledge that inclusion probabilities will differ between
different population groupings. The coverage list requirements under the
Zhang and Dunne method allow for the use of a suitable administrative list
provided the conditions are satisfied and negate the need to undertake a
survey in the field. The Graham and Lin method requires knowledge of the
actual inclusion probabilities and where they are not known then high quality
estimates are required. The Graham and Lin method requires a greater
knowledge of inclusion probabilities. The coverage list requirements in the
Graham and Lin method can probably only be satisfied by conducting an
appropriate field survey that will satisfy requirements with respect to
knowledge of the inclusion probabilities. This is probably the critical difference
in the application of the two methods.
To date, the Zhang and Dunne method has been used to compile
population estimates as part of the PECADO project for years 2011 to 2016.
The PECADO estimates are considered research in nature. When comparing
the estimates with that of the 2016 Census of Population estimates there are
some differences, part of which can be explained by different population
concepts. However, more work is required to reconcile the differences
between the PECADO estimates and the Census of Population counts. So far,
the Graham and Lin method has been evaluated only on simulated data and
shows good statistical properties.
The Graham and Lin method will eliminate overcoverage errors from the
estimated target population. The Zhang and Dunne method depends on
deploying an effective trimming strategy to eliminate overcoverage errors if
they exist and the only guarantee with respect to completely eliminating
overcoverage is the belief that the trimming strategy has effectively evaluated
all possible sources of overcoverage.
In conclusion, the authors believe it is possible to compile population
estimates from administrative data sources without the requirement of a
public administration systems underpinned by a CPR. The work undertaken to
date shows this possibility, however, more work is required in developing the
respective methods.

References
1. Bishop, Y., Feinberg, S., and Holland, P. (1975). Discrete Multivariate
Analysis. Springer.

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