The Unintended Effects of a Ban on Sex-Selective Abortion on Infant Mortality: Evidence from India

Abstract

India has long struggled with persistent problems of sex-selective abortions and the neglect of female children. In 1996, the Pre-Natal Diagnostics Techniques Act was implemented to stop the practices of prenatal sex determination and selective abortions. This paper examines whether the law has been effective in reducing gender imbalance, and in turn potentially exacerbated post-natal discrimination against newborn girls. Using retrospective birth history data from the Indian District Level Household Survey (2002–2004), we exploit a natural experiment involving a variation in the timing of the law across states. We analyse the differential impact of the law on newborn sex ratios and infant mortality rates. Our findings indicate that the law significantly increased the likelihood of a female birth, improving female-to-male sex ratios at birth. We also find that it was generally associated with no change in the relative mortality of infant girls.

Notes

¹ Hu & Schlosser (2012) show a negative relationship between prenatal and post-natal sex selection. The authors find that a rise in prenatal sex selection is associated with lower post-natal neglect of girls in India.

² For example, several studies (Hesketh & Zhu, 1997; Das Gupta, 2005; Hesketh et al., 2005; Qian, 2009; Zhu et al., 2009; Ebenstein, 2010) find that the “unintended consequence” of the Chinese “One Child Policy” was a strengthening of the preference for sons and, therefore, a male-biased sex ratio. Lin et al. (2010) find that after abortion was legalised in 1985–1986 in Taiwan, there was a significant rise in the share of male births. However, the neglect of living young girls also declined substantially, as exhibited by a 25% reduction in excess female child mortality.

³ However, we do find a small effect of the law in increasing relative female infant mortality in some of our additional analysis, as discussed later.

⁴ We choose an LPM specification in our analysis mainly because the wild bootstrap correction method has yet to be fully developed for non-linear models. We ran additional analyses with probit and logit model specifications (with clustered standard errors but without wild bootstrap correction) to find very similar results.

⁵ Although the R² values of our regression models are seemingly low, they are similar to those reported by other studies that use large cross-sectional data-sets. See, for example, Nandi & Deolalikar (2013). However, another reason for a low R² value could be that it is not a good measure of the goodness of fit of LPM models.