Skip to main content

Modeling Social Diffusion Using Geostatistics: Finding the Causes of Fertility Decline

Explaining the decline of fertility in developing countries has been one of the main foci of demographic research in the past half-century. There are several leading explanations. One is traditional demographic transition, or "demand" theory, developed by Kingsley Davis in 1945, and presented in perhaps its most refined form by Richard Easterlin in the 1980s. This says that actual fertility is influenced in specific ways by the demand for children, the supply of children, and the costs of fertility regulation. These factors are in turn influenced by quantities such as the schooling of children and infant mortality. A second leading explanation is ideation theory, which identifies modern ideas and their diffusion as the key determinant. Of course, all these factors may be important, in which case identifying their relative contributions becomes the research task.

It is hard to adjudicate between these theories, because the relevant variables all tend to change together. In a previous study using data from Iran in 1950-1977, we were able to make this distinction, and found that the fertility decline in Iran was better explained by demand factors than by ideational ones.

The goal in this project is to extend these results to other countries, to see if they hold up on a wider scale. To that end, we have already collected a data set giving fertility rates and posited determinants for each of 35 countries that participated in the original World Fertility Survey for 1950 to 1995. A key part of ideation theory is the idea of diffusion of ideas, and this should reflect itself in spatial correlation in our data. We also intend to take account explicitly of missing data. We will build a Bayesian model for the data that incorporates geostatistical ideas from spatial statistics, and multiple imputation ideas for missing data, and estimate it using Markov chain Monte Carlo methods. We will compare the competing hypotheses by building statistical models corresponding to each of them, and comparing them using Bayes factors.