N. K. Namboodiri deserves much credit for turning fertility analysis toward the study of the transition from one birth order to the next (Namboodiri, 1972). Economists are also modifying their approach to fertility in the same fashion (Heckman and Willis, n.d.; Michael and Willis, n.d.). In Namboodiri's recent attempt to put empirical flesh on his theoretical skeleton (Which Couples at Given Parities Expect to Have Additional Births? An Exercise in Discriminant Analysis, Demography, Vol. 11, February 1974, pp. 45-56), however, he arrives at conclusions which require further examination. Namboodiri's major finding, based on a discriminant analysis of the NFS data set, is that socioeconomic variables only discriminate among couples who have reached high parity levels with regard to birth expectations. We show, however, that this result is mainly the product of Namboodiri's experimental design and the misspecification of the discriminant functions with respect *to the socioeconomic variables. Based on a somewhat different analysis of the same data set, we obtain strong evidence that contradicts Namboodiri's conclusion that there exists a threshold below which economic and social characteristics have no bearing on birth decisions. The major reason for the low discriminatory power of the economic and sociological variables at low parities in Namboodiri's analysis appears to be the domination of the age variable at these levels. Since the relationship between age and birth expectations is not particularly informative-how much have we learned if we find that for women in parities 0, 1, and 2 a 43 year old is less likely to expect additional children than a 23 year old?-a better procedure to test the influence of the socioeconomic variables would have been to not stratify the sample by parity but to relate parity and age variables along with the social and economic proxies to birth expectations for the whole sample. This methodology would, in addition, have allowed a more precise testing with the use of interaction terms of the hypothesis that the sets of variables have different effects at each parity. Namboodiri's technique of comparing the R2's across parity levels (p. 51) is not informative in this regard because the variances in the independent variable (particularly age) differ by parity. Thus it is not unexpected that among higher-parity families, whose age variance is much less than that pertaining to low-parity families, Namboodiri finds that the predictive power of age is significantly lower. A more serious shortcoming in Namboodiri's analysis, which also would tend to reduce the significance of the set of socioeconomic variables, is his misspecification of these parameters. Since the deDendent variable is exDectational, it is
Read full abstract