A THEORETICAL AND EMPIRICAL CRITIQUE OF THE EASTERLIN-CRIMMINS SYNTHESIS MODEL OF FERTILITY--WITH APPLICATIONS TO COLOMBIA, COSTA RICA, NEPAL, SRI LANKA, AND THE UNITED STATES

This dissertation provides a theoretical and empirical critique of the Easterlin-Crimmins synthesis approach to fertility analysis. Five data sets are examined--Colombia (1976-WFS), Costa Rica (1976-WFS), Nepal (1976-WFS), Sri Lanka (1975-WFS), and the United States (1973-NSFG). Several "improvements" to the synthesis model are suggested, particularly in the areas of model specification and estimation. Among the more important findings, sans cavaets, are the following.

The vast majority of coefficients derived from a synthesis approach are correctly signed, of reasonable magnitudes, and are robust to changes in model specification or estimation. Two problem areas--the estimate of couple-level potential fertility and the estimate of the couple's numbers of birth averted by contraceptive use--are found to be interdependent, statistically. Estimation is improved when potential fertility is modeled as a latent variable. Appropriate estimators include Tobit 2SLS techniques, or the FIML methods developed and advocated herein.

Complications arise if potential fertility is heterogenous. Suggestions for dealing with heterogeneity are discussed, but remain unimplemented.

Several unexpected findings arise. There is evidence that a rationalization bias in reported numbers of desired children is stronger in South Asia than in Latin America. The data confirm the notion that the number of methods of contraception known by respondents is biased upwards by prior exposure to contraceptive use. Wives who work on a farm before marriage have abnormally high levels of infant and child mortality even after controlling for other pertinent socioeconomic status variables such as current rural residence.

The dissertation concludes with suggestions for "new and improved" fertility surveys; a call for a sequential/dynamic version of the synthesis model; and reference to a strategy for integrating micro-level synthesis output in macro-level analyses. Continued synthesis research should shed new light on the causes and consequences of the fertility transition in third world (and developed) countries.