Use of mark-recapture techniques to estimate the size of hard-to-reach populations.

Abstract

To review the main problems associated with mark-recapture methods of population estimation, and to indicate some practical strategies for addressing these problems, with illustrations from a study of drug-use prevalence across Wales. Unnamed identifier data were collected in 1994 on 2610 drug users who were in contact with various agencies across Wales: the police, drug treatment agencies, needle exchanges, probation services, and agencies reporting to the Welsh Drugs Misuse Database. Based on the dependency relationships between different agencies' datasets, different estimates of the 'hidden' populations (not in contact with agencies) were modelled for each county, for males and females, for injecting drug users and serious drug users, and for those under 25 and those over 25 years of age. Different models were also constructed for the same subpopulations, using different agency datasets and different criteria of overlap between them, yielding a total of 230 different models. The issues of sample heterogeneity and population definition are particularly intractable in mark-recapture studies. Sample heterogeneity may be partly addressed by separately modelling different subpopulations to check whether they show the same dependency relationships as the main population. Population definition may be partly addressed by restricting modelling to datasets thought to share roughly congruent population definitions.