Affiliations: Institute for Social Science Research, University of
Queensland, Brisbane, Australia | Southampton Statistical Sciences Research Institute,
University of Southampton, Southampton, UK
Note: [] Corresponding author: Bernard Baffour, Institute for Social
Science Research, University of Queensland, Brisbane, Australia.
E-mail: [email protected]
Abstract: The value of a census cannot be overstated, given that no other data
resource provides such detailed information about the population. Further,
censuses are often the only historical data source to map out change over time
due the consistency of questions asked. However, it is often the most expensive
undertaking – other than going to war – that a country embarks on.
Countries are thus seeking more cost-effective alternatives. This paper details
some exploratory research into one such alternative, based on capture-recapture
methods. Capture recapture methods have been used for population estimation for
decades, but the focus has been on dual system estimation. Dual system measurement of the population has been criticized for
its reliance on the independence assumption between the two systems. This
assumption is untestable, and failure introduces bias into the estimates of the
population. The most logical improvement of dual system estimation is triple
system estimation. In this paper, a simulation study is carried out to compare
the performance of different dual and triple system estimators of the
population size under various dependency scenarios. Performance is explored
through both the bias and variability. The study shows that the dual system
estimator copes well with dependence, provided the coverage of both lists are
reasonably high. In addition, although the triple system estimators yield less
biased estimates of the population, the dual system estimator is shown to be
robust enough to cope with low levels of dependence.
Keywords: Census estimation, capture-recapture, non-response, population coverage
