On the care and handling of regression specifications in fertility research.

Results of a regression study in which samples of women whose ages spanned the entire childbearing period were used to determine the effects of couples' characteristics (education, income, age at marriage, duration of marriage, and infant and child mortality) on completed fertility were analyzed. It was suggested that the regression findings (no significant effect of income, infant mortality, education of a father, or age at marriage) were not valid behavioral measurements, but the result of specification errors. Specifically, it was suggested that 1) the relation between income and fertility is more complex than estimated in the regression specifications, and a rise in income at low-income levels may initially increase the number of children born; and 2) the relation between infant and child mortality and fertility cannot be measured in the type of study considered, only by complete birth histories and socioeconomic information on women past childbearing age. It is urged that significant findings in this type of research, which is on the increase, be examined very critically to eliminate spurious results.