Testing for spatial diffusion of the onset of fertility decline accounting for structural variables

Renato M. Assuncao, Pontifícia Universidade Católica de Minas Gerais

Much of the debate in the fertility literature has concerned the presence and strength of spatial diffusion effects, as opposed to the role of development. Bocquet-Appel and Jakobi (1998) suggested the use of a space-time interaction test based on the Knox statistic to test for the evidence of spatial diffusion of contraception at the onset of fertility transition in Great Britain. This test has one major drawback: its failure to account for structural variables, which also change in space and time, induce biased results for the significance test increasing the real type I error level, even if the null hypothesis is true. We propose a Monte Carlo method for constructing unbiased space-time interaction tests. The procedure is illustrated with simulated data and with real fertility data from the last five Brazilian censuses.