Main Effects of Aggregate Education on Birth Rates Other Effects of Aggregate education

5. Macro-Micro Interactions in Birth Rate Models

The fact that there are no main effects of aggregate education does not exclude the possibility of interaction effects, because the effect of a high average educational level might, in principle, be significantly positive for a woman with low education and significantly negative for one with high education, or vice versa. However, more flexible model specifications revealed no such pattern.

The simplest check was to include an interaction term between average length of education (or its deviation from the sample mean of about 7 years, in order to obtain individual effect estimates closer to those shown in Table 1) and length of individual education (transformed into a 0-1 scale to facilitate interpretation). This is a linear interaction, in the sense that each additional year of individual education increases or decreases the impact of aggregate education equally much. According to the point estimates, the impact of an extra year of average education on first-birth rates changes gradually from 0.11 at the lowest individual level to -0.01 (=0.11-0.12) at the highest level, but neither the main effect nor the interaction effect was significant (Model 1, Table 3). The estimates for higher-order births suggested an opposite pattern, but were also very far from attaining significance.

(Table 3 about here)

Some other specifications were also tried. First, aggregate effects were allowed to differ freely across three broad categories of individual education, rather than assuming a linear interaction. No significant effects appeared, and there was no U-shaped pattern indicating that the linear specification was highly inappropriate (Model 2).

Second, the proportion illiterate was included instead of the average education as a main effect and in interaction with individual education. As an alternative, average education among those with more than 3 years of schooling was included. No aggregate main or interaction effects were significant with these specifications (Model 3 and Model 4). Nor did the effects attain significance when the linearity assumption was removed, as in Model 2 (not shown).

Finally, it was checked whether women at low, medium or high education were influenced by the proportion at a higher (if relevant) or lower (if relevant) level. No significant effects were found in these models either (not shown).

Also interactions between urban/rural and individual education were included in these models, but without influencing the cross-level education interactions markedly. In fact, interaction effects were insignificant even when main effects of urban/rural and religion were left out.

Models estimated only for women living in rural enumeration areas (not shown) gave results very similar to those shown in Table 3. The only notable difference was that the main effect of depth of education was significant at the 0.10 level for higher-order births. The main effect was -0.14 and the interaction effect 0.15, indicating that an effect of depth only exists at low educational levels.


Main Effects of Aggregate Education on Birth Rates Other Effects of Aggregate education

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