Using a mixture model to detect son preference in Vietnam.

ABSTRACT

Son preference is strong in Vietnam, according to attitudinal surveys and studies of contraceptive prevalence and birth hazards. These techniques assume a single model is valid for all families, but it is more plausible that son preference is found for some, but not all, families. Heterogeneous preferences may be addressed with a mixture model. This paper specifies and estimates a two-Weibull regression model, applied to the interval between the second and third births. The data come from the Vietnam Living Standards Survey of 1992-93. Applying information criteria, graphs, and martingale-based residuals, the two-Weibull model is found to fit better than a one-Weibull model. Roughly half of parents have son preference and, curiously, a propensity for fewer children. The other group has more children, no son preference, and is colourless in the sense that the birth interval is difficult to predict on the basis of the regressors used.

ABSTRACT

PIP Attitudinal surveys consistently indicate that parents in Vietnam have a strong desire to bear at least one son before they terminate their childbearing. However, stated preferences for sons are believed to be unreliable as a guide to actual fertility behavior. Contraceptive prevalence models and proportional hazards model have therefore been used to get better insight into the existence or absence of son preference in Vietnam. These latter studies have also found son preference to be strong in Vietnam. A central problem, however, is that these techniques assume a single model is valid for all families. The authors explain that the propensity to have another child may actually depend upon different variables, or depend upon the same variables but in a different way, for different families. One way to address this problem of heterogenous preferences is to fit a mixture model. In the context of son preference, it is plausible that the population is naturally divided into those who prefer sons and those who do not. Such a scenario calls for a two-distribution mixture model. In this paper, a Weibull regression model is applied to the interval between the second and third child, using data from the Vietnam Living Standards Survey of 1992-93. A single Weibull regression model is first estimated, and then a mixture of two Weibull regression models, to explore whether this represents an improvement over the single model. The mixture model estimates that preference for a son is strong for about half of the parents surveyed and absent from the other half.