Enumeration accuracy in a population census: an evaluation using latent class analysis

Abstract

To evaluate the coverage error in a population census enumeration, many countries conduct a’Post Enumeration Survey (PES) which is designed to identify individuals who were missed in the Census or individuals who were counted that should not have been. The quality of the PES evaluation of coverage error is only as good as the quality of the PES itself and much effort has been devoted world-wide to improving the PES enumeration methodology. In this article, we apply latent class analysis (LCA) to evaluate the quality of the PES and compare estimates of the PES classification error with the corresponding estimates from a traditional analysis of’these data. The primary basis for these evaluations is a reconciled reinterview survey of the PES respondents. The traditional analysis treats the reconciled reinterview survey results as infallible and attributes disagreements between the PES and the reconciled reinterview classifications to deficiencies in the PES. With LCA, the reinterview results are treated as fallible measures which simply produce another indicator of true residence status. LCA estimates of the error probabilities for all three classifiers (the Census, the PES, and the reconciled PES reinterview) are obtained by maximum likelihood estimation under an assumed latent class model. In this article, we demonstrate the use of LCA for evaluating post enumeration survey accuracy and summarize the key findings for PES evaluation studies.