Macro-Micro Interactions in Birth Rate Models The Importance of Husbands' and Men's Education

6. Other Effects of Aggregate Education

When a model for fertility desires was estimated, with the same control variables as in the birth rate models, a significant negative effect of average length of education appeared (Table 4) [Note 12]. Also breadth and depth of education were found to weaken fertility desires.

(Table 4 about here)

Post-partum susceptibility was found to be independent of aggregate education. The point estimate suggested that women in districts with many illiterate tend to breastfeed longer or more intensely or to resume sexual activity later than women at the same educational level in other districts. This accords with the commonly found impact of individual education (which does not appear clearly in these data), but the effect was far from significant.

When the probability of using modern contraception was estimated for women who did not want another child within the next two years, and who were sexually active, not amenorrhoeic and not infecund, the effect of average education did not reach significant, while that of average education among the literate was significant at the 0.01 level. In accordance with the results for fertility desires, an unconditional model for contraceptive use gave more clearly significant effects of aggregate education. Both models for contraceptive use suggested sharper effects of depth of education than breadth.

Apparently, the bias introduced by considering individuals within a district or enumeration area as independent is minor (and certainly compared with other problems that hamper this and other similar analyses). An ordinary single-level logistic model for fertility desires gave a point estimate of -0.26 for the effect of average education (Table 4), with a standard error of 0.070 (not shown). This result was replicated in a two-level MLwiN model (based on the 1. order MQL algorithm) by using a constant variable as a level-two identifier and constraining the variance at this fictitious level to be 0. When enumeration area or district were instead defined as level two, the variance of the aggregate-education effect estimate only increased to 0.084-0.087 (both according to the 1. order MQL and the 2. order PQL algorithm; not shown). Several similar attempts were made to estimated MLwiN models for contraceptive use, but convergence to plausible estimates was never achieved.

The interactions between average and individual length of education were not significant in these demand, supply and regulation models. However, the estimates suggested that the impact of aggregate education on fertility desires is strongest among women who themselves have little education.

Closely similar patterns appeared in models that were estimated only for women in rural enumeration areas (not shown). With this restriction, however, effects of average length of education were significant at the 0.10 level in models for contraceptive use among women not wanting another child soon. Besides, the interaction between average and individual education was significantly positive (at the 0.05 level) in models for unconditional contraceptive use.

The results were only marginally changed when a more detailed categorization of the current place of residence was employed by grouping rural areas according to distance from a city.


Macro-Micro Interactions in Birth Rate Models The Importance of Husbands' and Men's Education

A Search for Aggregate-Level Effects of Education on Fertility, Using Data from Zimbabwe
Øystein Kravdal
© 2000 Max-Planck-Gesellschaft ISSN 1435-9871