Disease Eradication, Infant Mortality, and Fertility Response: Evidence from Malaria Eradication in India

Abstract

Using a massive malaria eradication program in India during the 1950s as a natural experiment, we examine the effects of disease environment on child health outcomes and fertility. We harmonise a rich dataset on malaria endemicity with fertility histories of women to exploit the cohort level variation in exposure to the program. We find that the program leads to a significant decline in infant and neonatal mortality and leads to a significant increase in the probability of birth during adolescent years in high malaria endemic regions. We confirm that a fall in the mother’s age at first birth in the post-eradication period drives the fertility response.

Keywords:

• Malaria
• disease eradication
• infant mortality
• fertility
• selection bias

JEL Classification:

• I12 I
• 15 I
• 18 J13

Acknowledgements

We have used the replication dataset of Cutler et al. (2010) and we would like to thank the authors for making this publicly available. We would also like to thank Farzana Afridi, Sonia Bhalotra, Reyn van Ewijk, Tarun Jain, Jyotsna Jalan, Stephan Klasen, and Abhiroop Mukhopadhyay for their excellent comments and suggestions. We benefitted from suggestions by seminar participants at Ashoka University, Essen Health Conference, Indian Statistical Institute-New Delhi, Indian Institute of Management-Bangalore, Indian Institute of Management-Kolkata, Centre for Studies in Social Sciences, Indira Gandhi Institute of Development Research, Institute of Development Studies-Kolkata, and University of Göttingen. We will make our Stata codes available to researchers on request.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The malaria data come from the replication dataset of Cutler et al. (2010). The data are available at https://www.aeaweb.org/articles?id=10.1257/app.2.2.72. The individual-level data used in the paper come from the first round of the National Family Health Survey (1992–1993). It is freely available from the Demographic and Health Surveys website but requires their permission for download and circulation.

Notes

1 While a set of studies highlight the negative relationship between disease environment and growth (Ashraf et al., 2008; Bloom, Canning, Kotschy, Prettner, & Schünemann, 2019; Gallup & Sachs, 2001), an improvement in population health is not necessarily associated with an improvement in the rate of growth (Acemoglu & Johnson, 2007; Young, 2005).

2 The malaria endemicity data come from Cutler et al. (2010), who analyse the effects of malaria eradication on education and consumption in India.

3 Our main estimation strategy aligns with the difference in difference approach used in Bleakley (2010), Cutler et al. (2010), and Lucas (2010).

4 The increased probability of adolescent births might not necessarily translate into an increase in the overall number of children. To investigate the effects of the eradication program on the total number of children, we use data from the Rural Economic Development Survey (REDS) data conducted in 1982. This is explained in section A2 of the Supplementary Materials. We find an increase in the overall number of pregnancies and children for women in malarious districts, who have higher exposure in the post-period.

5 A recent set of papers (Barofsky et al., 2015; Cogneau & Rossi, 2019; Wilde et al., 2019) look at the impact of malaria intervention on infant and child mortality in the context of Africa.

6 Spleen rate is the proportion of children with an enlarged spleen. Spleen enlargement and infant parasite rate is a sign of malarial infection (Lucas, 2013).

7 Table A11 of the Supplementary Materials shows that our results remain unchanged if we define malaria endemicity as the modal value of the malaria index in a district. However, the sample size falls drastically due to the presence of bimodal or multimodal districts. Thus, some of the effects turn statistically insignificant, even though the magnitude of the effects do not change considerably.

8 Our results are robust to the inclusion of all years, and not just years after a woman moved into her current area of residence, as shown in Table A5 of Supplementary Materials.

9 Existing literature suggests that pregnancy and childbearing in adolescents may expose them to acute health risks during pregnancy and childbirth (Ganchimeg et al., 2014; Santhya & Jejeebhoy, 2003). Thus comparing adolescent births with all births might lead to biased results.

10 Births before the age of 12 are extremely rare. Only 0.01 per cent of all births take place before the age of 12 in the NFHS sample. There is also a discrete jump in the number of births at the age of 12.

11 Tables A1, A2, and A3 in the Supplementary Materials report the descriptive statistics separately for rural and urban areas.

12 The historical data on the spleen rate of children for 15 states in India come from Cutler et al. (2010).

13 One concern with the mortality specification is that we are selecting children based on the age of birth of the mother. We show that age at first birth fell in the post-eradication period. Table A4 of Supplementary Materials shows that the mortality results are robust to the inclusion of all children.

14 Our results are robust if we vary the age threshold as shown in Table A6 in Supplementary Materials

15 There can be concerns of age heaping which comes from the fact that ages are often rounded off to the nearest six months (Bhalotra, 2010). To address this concern, we have also estimated the infant mortality results excluding the 12th month. The results are presented in Table A9 of Supplementary Materials. We see that the results are unchanged.

16 We show that our results are robust to the inclusion of controls for parental education in Table A8 of Supplementary Materials. Given the potential effects of the malaria eradication program on educational outcomes as analysed in Cutler et al. (2010), we do not include these controls in our main specification.

17 One can argue that survival analysis is more accurate for fertility regressions since future births are not included in the data. We present the results from the estimation of a Cox hazard model in Table A15 of Supplementary Materials. Unlike the linear probability model, we cannot include district-specific linear time trends in the Cox model. The effect is still positive for rural areas (hazard ratio greater than 1). However, the effect is statistically insignificant.

18 Our results imply that the fertility change is likely to be driven by the direct biological impact on fecundity. In this case, we expect that the infant mortality and the fertility response to disease eradication should be around the same time. We verify that our results are robust to the exclusion of children born only in 1960 and 1961 (similar to the infant mortality regressions) in Table A10 of the Supplementary Materials. The table also shows that the results are robust to alternate choice of time window in fertility response.

19 We have done a series of robustness tests in the Supplementary Materials which shows that our results are robust to the sample criterion, choice of controls, age-heaping concerns in infant mortality regressions, choice of delay in fertility response and modal malaria endemicity. The results are presented in Tables A4–A12 of the Supplementary Materials.

20 There is no variation in the treatment timing in our analysis. Hence, our estimates are free from the bias generated when differential treatment timing with two-way fixed effects are used for the identification of treatment effects (Abraham & Sun, 2021; Goodman-Bacon, 2021).