Women's education may be even better than we thought: estimating the gains from education when schooling ain't learning
Women’s schooling has long been regarded as one of the best investments in development. Using two different cross-nationally comparable data sets which both contain measures of schooling, assessments of literacy, and life outcomes for more than 50 countries, we show the association of women’s education (defined as schooling and the acquisition of literacy) with four life outcomes (fertility, child mortality, empowerment and financial practices) is much larger than the standard estimates of the gains from schooling alone.
First, estimates of the association of outcomes with schooling alone cannot distinguish between the association of outcomes with schooling that actually produces increased learning and schooling that does not. Second, typical estimates do not address attenuation bias from measurement error. Using the new data on literacy to partially address these deficiencies, we find that the associations of women’s basic education (completing primary schooling and attaining literacy) with child mortality, fertility, women’s empowerment and the associations of men’s and women’s basic education with positive financial practices are three to five times larger than standard estimates. For instance, our country aggregated OLS estimate of the association of women’s empowerment with primary schooling versus no schooling is 0.15 of a standard deviation of the index, but the estimated association for women with primary schooling and literacy, using IV to correct for attenuation bias, is 0.68, 4.6 times bigger.
The paper's findings raise two conceptual points. First, if the causal pathway through which schooling affects life outcomes is, even partially, through learning then estimates of the impact of schooling will underestimate the impact of education. Second, decisions about how to invest to improve life outcomes necessarily depend on estimates of the relative impacts and relative costs of schooling (eg grade completion) versus learning (eg literacy) on life outcomes. The results do share the limitation of all previous observational results that the associations cannot be given causal interpretation and much more work will be needed to be able to make reliable claims about causal pathways.