Measurement and replication: evaluating the consistency of eight U. S. fertility surveys

This paper addresses the comparability of data from eight national fertility surveys conducted in the United States between 1955 and 1980. We examine the extent to which substantive conclusions about the levels and correlates of the timing of first birth vary across these data sources by comparing samples of women who were eligible to be included in more than one study. Results obtained from Ordinary Least Squares (OLS) and logistic regression analyses of several measures of timing indicate few study effects. Thus findings based on data from one survey can be replicated with comparable data from another. One notable exception is the relationship between contraceptive use prior to the first birth and the timing of that birth. The contraceptive variable appears to be poorly measured, leading to unstable estimates of the relationship. In general, our results suggest: (1) the feasibility of pooling data from the various surveys when modeling delayed childbearing; and (2) reasonable accuracy in assessing trends in the timing of first birth from multiple data sources.