Abstract Theoretical Considerations

1. Introduction

Although women's education has been one of the most thoroughly studied determinants of fertility, with the perspective now often extended to include the closely related `women's position', the research area is still far from being exhausted. For example, several causal links seem plausible in light of existing empirical evidence, but we have a meagre knowledge about their relative importance. The differences in education effects across settings also deserve further exploration. Another, and not so widely recognized challenge, is to find out whether education at the aggregate level has any effect on a woman's fertility above and beyond that of her own education. The possible importance of `mass education' was discussed by Caldwell [Caldwell 1980] many years ago, and has occasionally been touched in more recent reviews (see e. g. [Cleland and Jejeebhoy 1996]), but little empirical evidence has so far been accumulated.

When the intention is to assess how expansion of education, which of course is a goal itself, influences fertility, one obviously needs to estimate effects of a woman's own education. In addition, there may be a `spill-over' from other people's education through, for example, social learning. Uneducated women who live in societies where a large proportion are literate, or where the average educational level is high, may have a fertility different from that of uneducated women elsewhere. Also the better-educated may be influenced by the educational distribution in the community. If aggregate education has, on the whole, a substantial depressing effect, fertility will decline more sharply in response to an increase in women's education than suggested by the estimates of individual-level effects.

It would clearly be important to try to quantify this aggregate-level contribution. This calls for the inclusion of both individual and aggregate measures of education in multilevel models (along with some control variables that are determinants of education, of course). As illustrated below, the individual-level models usually estimated will capture part of the aggregate-level effect, in addition to the purely individual one, but not the full effect.

Many researchers have included individual and structural characteristics simultaneously in regression models for reproductive behaviour, with or without the use of statistical tools specially developed for multilevel analysis. Not least the impact of family planning facilities in the community has attracted interest (see e.g. [Entwisle et al 1997]). However, very few have provided statistically well-founded assessments of how aggregate education influences fertility.

Some efforts were made already on the basis of community data that were collected as part of the World Fertility Surveys in a few countries, and with the focus largely on the number of schools in the village or nearby (reviewed in [Casterline 1985]). More importantly, Tienda et al. [Tienda, Diaz and Smith 1985] found effects of average length of education on cumulated fertility in Peru, net of the woman's own education. However, their models did not include a measure of urbanization, which is likely to be a particularly important determinant of education. In a Kenyan study, Lesthaeghe et al. [Lesthaeghe et al 1985] searched for effects of aggregate education, using both a measure of cumulated fertility and some proximate determinants as dependent variables. Some main or interaction effects showed up, but the close link between education and urbanization was not taken into account in this study either. Similarly, Hirschman and Guest [Hirschman and Guest 1990] estimated a significant fertility-inhibiting effect of the proportion of women with post-primary education in four Southeast Asian countries, but without an urban/rural control variable.

In a more recent study, Thomas [Thomas 1999] reported that mean educational level in the community had a significant depressing effect on the number of children ever born in South Africa. Unfortunately, he did not show the size of this effect. Moreover, a Tanzanian study by Kravdal [Kravdal 2000] provided indications that higher-order birth rates were relatively low in regions where many women were literate, whereas sharper such effects appeared in a model for contraceptive use. A similar result for contraceptive use was reported by Amin et al. [Amin, Diamond and Steel 1996], who estimated the influence of the proportion literate women in various districts of Bangladesh.

It is not unlikely that such aggregate effects are different for educated and uneducated women, one way or the other. This is, of course, equivalent to saying that the individual effect of education may depend on the general educational level. Very little is known about such context-dependence. Jejeebhoy [Jejeebhoy 1995] reviewed several studies of mixed character and quality from countries with different overall educational level, without arriving at a very clear conclusion, and only some of the original multilevel studies referred above included tests of cross-level education interactions.

The main objective of this analysis was to find out whether various measures of aggregate education have had an effect on fertility, net of the individual education, and with control for urbanization. As part of this, the differential impact across women's own education was also addressed. Contrary to some of the above-mentioned studies, where cumulated fertility was the dependent variable, the focus here has been on recent births in a hazard model approach. This means that the available cross-sectional data on aggregate education were more relevant in terms of place and time. Moreover, models were estimated for fertility desires, contraceptive use and post-partum insusceptibility due to lactational amenorrhoea or sexual abstinence in order to build up a more complete picture. These important determinants of fertility may respond differently (also) to aggregate education. As very common in studies of education and fertility, only women's education was included in the models. In a final section, however, the importance of husbands' and men's education will be briefly dealt with.

The data were from the Zimbabwe Demographic and Health Survey of 1994, combined with 1992 census data for 70 districts. This is a very suitable data source, because of the detailed information on education in the census, the fairly large number of districts, and the substantial variation across them. For example, women's illiteracy ranges from 4% to 42%, and the proportion with at least 4 years of secondary education from 4% to 38%. However, the data did not allow a check of how investments in schooling in a largely illiterate society may influence fertility. Zimbabwe is far beyond this stage, and is, also for other reasons (see e.g. [Guilkey and Jayne 1997]), one of the countries in sub-Saharan Africa with lowest fertility. According to the Demographic and Health Surveys of 1994 and 1988, total fertility rate was 4.3 in the early 1990s and 5.5 in the mid-1980s [Central Statistical Office 1993].

 

Abstract Theoretical Considerations

A Search for Aggregate-Level Effects of Education on Fertility, Using Data from Zimbabwe
Øystein Kravdal
© 2000 Max-Planck-Gesellschaft ISSN 1435-9871
http://www.demographic-research.org/Volumes/Vol3/3