A comparative analysis of determinants of birth intervals

The life-table approach is particularly appropriate to the analysis of birth intervals because a major pitfall of measurement can be avoided. If only intervals beginning with the birth of 1 child and ending with the next birth are considered, there is a major bias in any results because all information is omitted for women who stopped childbearing within that interval -- those who never had a subsequent birth or, in the case of data obtained from surveys, those whose reproductive histories may be incomplete at the time of the interview. New methods have become available that permit simultaneous analysis of life tables according to several covariates and produce more readily interpretable summary and comparative statistics. It is important to develop efficient ways of determining which hazard models are most likely to describe a particular data set well, so that exact estimation procedures are applied to only a small number of candidates for the final model. Thus, this discussion of the birth interval analysis is conducted in 2 parts: screening and final estimation. This is preceded by a brief discussion of the information used in the analysis. The decision was made to work with the 9 countries used by Hobcraft and Casterline (1983) and again by Gilks (1982) in his work on birth intervals. These include: Bangladesh, Colombia, Indonesia, Jamaica, Jordan, Kenya, Korea, Mexico, and Sri Lanka. The advantages of comparability with other work outweighed the marginal benefits of choosing a set of countries with optimal variability in terms of patterns of birth interval distributions. Even after controlling for all the factors in the final model, the adjusted quintums showed large variations between subgroups. For age at start of the interval, the differences in the adjusted quintums between the youngest and oldest groups was in excess of 0.2 for all countries except Kenya and, marginally, Jordan and was around 0.3 for Korea, Mexico, and Sri Lanka. At the other extreme differences between adjusted quintums by birth order exceeded 0.05 only for Korea, which had a large range of 0.20. These findings suggest that age is of far greater importance in determining fertility behavior than is birth order, even where there is known to be substantial fertility control, although Korea is a striking exception. A partial explanation of the unusual pattern of Korea is the impact of sex preference