Health Policy and Planning Advance Access originally published online on March 29, 2006
Health Policy and Planning 2006 21(3):231-240; doi:10.1093/heapol/czl007
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Do community factors have a differential impact on the health outcomes of boys and girls? Evidence from rural Pakistan
Department of Economics, Middlebury College, VT, USA
Correspondence: Jessica Holmes, Department of Economics, Middlebury College, Middleybury, VT 05753, USA. Tel: + 1 802–443–3439; E-mail: jholmes{at}middlebury.edu
In countries with large gender disparities in health status, can investments in local communities mitigate the gender bias observed in intra-household resource allocations? This paper explores the evidence for gender differences in the impact of community prices and infrastructure on child nutritional outcomes. Standardized heights and weights of rural Pakistani children are used as health indicators, while community factors include wheat prices, availability of piped water, accessibility of shops and government health clinics and the quality of the closest health facilities. The results suggest that food subsidies and programmes designed to improve the access and quality of local services may reduce the impact of intra-household gender bias on child nutrition, particularly in the long run. Specifically, by increasing the affordability of staple foods, improving the access to shops and government health centres and enhancing the quality of local care, particularly (gender-neutral) prenatal care, gender gaps in health outcomes are likely to diminish.
Key Words: anthropometrics, child health, community infrastructure, gender bias, Pakistan
1See Haddad et al. (1999) for an extensive review that includes other regions.
2In particular, they report a price elasticity of demand for physicians that is 58% higher for daughters than for sons in the lowest income group.
3Children with z-scores greater than +6 or lower than –6 have been excluded from the analysis since even in a country with a lot of malnutrition, a z-score outside of the –6<Z<+6 is very unlikely and probably indicates an error in measurement (Kostermans 1993).
4Many other studies report no gender differences or even a slight female advantage in anthropometric outcomes in South Asia (see Haddad et al. 1996; Haddad 1999; Hazarika 2000).
5In practice, however, Lee et al. (1997), using data from both Bangladesh and the Philippines, and Pitt (1997), using data from 14 sub-Saharan African nations, reveal that parameter estimates of the determinants of child health change very little when mortality selection is accounted for in the analysis.
6Medicine chests in the clinics were checked for the following medicines: chloroquine, fansidar, aspirin, paracetamol, ponstan, oral rehydration salts, lomotil, flagyl, anti-worm medicine, penicillin or other antibiotics, tetanus vaccination.
7UNICEF has also developed an important structural framework for understanding the determinants of malnutrition (1990). Unfortunately, individual-level data on caloric intake or food expenditure (one component of the UNICEF concept of food security) were not available in the PIHS. A reduced form model of the determinants of malnutrition is estimated in the following section.
8It is possible that birth order is endogenous since households may jointly determine quantity and quality of children, but note that the results presented in the next section do not qualitatively change with the exclusion of birth order.
9The intuition is that women with more schooling have better market opportunities and thus stronger bargaining power within the household. If women prefer daughters, then mother's education should have a larger impact on the health outcomes of daughters relative to sons.
10Budget figures are from Ahmad and Qureshi (1990). To approximate province-specific population levels for 1990–91, data from the last available census (1981) were adjusted by the population growth rate for each province from 1972–1981 (Government of Pakistan 1995; Government of Pakistan 1996).
11By pooling the males and females and then interacting each explanatory variable with a gender dummy, it is possible to test whether there are significant differences between the male and female coefficients. The interaction terms from this pooled model are found in the Appendix. An F-test that the interaction terms are jointly equal to zero is rejected at the 0.01 level for both models.
12An F-test verifies the joint significance of the community-gender interactions in the standardized height model (p<0.00).
13This was calculated as follows: a one standard deviation reduction in the distance to nearest government centre (4.5 km) will close the mean height-for-age gap by 0.23 (i.e. the standard deviation of the distance to nearest government centre multiplied by the female coefficient on distance to nearest government centre (0.05) is 0.23). Since the mean female height-for-age gap is 1.73, this represents a 13% reduction in the mean female gap in height-for-age (0.23/1.73 = 0.13).
14An F-test verifies the joint significance of the community-gender interactions in the standardized weight model (p < 0.00).
15See, for example, Behrman and Wolfe (1987); Barrera (1990); Strauss (1990); Thomas et al. (1991); Cebu Study Team (1992); Thomas and Strauss (1992); Bhuiya et al. (1995); Pebley et al. (1996); Thomas et al. (1996); Hazarika (2000); Paknawin-Mock et al. (2000).