Photo by UNICEF Ethiopia (color adjusted). Source: Flickr (Nutrition-CBN-Deder East Harerghe).
You might have observed a theme around sex-disaggregated data emerging in our posts on the GNIE. Why? We ask Agnes Quisumbing, Senior Research Fellow in the Poverty, Health, and Nutrition division, at the International Food Policy Research Institute, to explain how this type of data are critical in agriculture research which explores the linkages between gender and nutrition. In addition, Agnes shares some experiences from the field, and gives some helpful tips for analysis. Catch up on the other posts on sex-disaggregated data, here and here.
Q: How can sex-disaggregated variables be incorporated in econometrics analysis to improve our understanding of gender-nutrition linkages?
Econometric analysis is a good tool to use when analysing sex-disaggregated data to understand gender-nutrition linkages better. One advantage of using econometric analysis is that one is able to control for multiple factors that might also affect the variable you’re interested in. The type of analysis that is possible depends on the questions you want to ask, and the level of disaggregation at which your data are available. Do you want to look at the determinants of household level outcomes, such as household calorie availability or dietary diversity, or do you want to look at the determinants of individual nutritional status? Are you interested in looking at differences in nutrition behaviour that would affect these outcomes? Most importantly, what is your “gender indicator?” Ideally, you would have more information aside from “sex of the household head,” the most common indicator used in many studies, and that you would have information at the individual level for all household members (available from the household roster) as well as the relevant decisionmaker for the outcomes under study.
Q. The data set that I have access to only has the sex of the household head as a “gender indicator”. Can I still do some meaningful gender analysis?
Although the sex of the household head is a very blunt instrument for understanding how gender relations can affect food security and nutrition, you can still do some meaningful and interesting analysis. Let’s say that you want to examine whether households headed by women in Ethiopia are more likely to lower their food consumption if they experience an unexpected negative event (a “shock”). Your dependent variable would be per capita consumption, and your independent variables would include a measure of the shock (say, whether the household experienced drought), the sex of the household head (a dummy variable that equals one if the head is female), controls for other household characteristics (size of land owned, age and sex composition of the household, livestock owned) and village characteristics. To find out whether the drought had a larger detrimental impact on female-headed households, you interact (multiply) the dummy variable for the sex of the household head with the “drought shock” dummy variable. If this variable is negative and significant, then you could conclude that the drought had a bigger negative impact on female headed households. An example of this analysis can be found in Chapter 5 of this book from CAPRi (2010).
Another example from Kumar and Quisumbing (2013), also from Ethiopia, examines whether female-headed households are more likely to report experiencing negative consequences from the global food price increases in 2007-2008, and whether they are more likely to use coping mechanisms that might have detrimental impact on children. Some of our work does this by using a dummy variable for the female-headed household, and then seeing whether the coefficient on this variable is significant. You can also interact the female headship dummy with other variables (say, size of land owned, asset ownership, educational attainment) to see whether the effect of female headship varies with other characteristics of the household (e.g., in many data sets size of land owned and livestock ownership is only at the household level) or the individual (e.g., Does it matter if the female head is older? More educated?)
Q. Suppose I have data from the household roster on the age, sex, and education of everyone in the household, their relationship to the household head, and also anthropometric measures for everyone in the household. What else can I do with these data?
A. You can do a lot more interesting stuff with these data! Let’s say you want to see whether boys or girls have a higher probability of being stunted, controlling for the characteristics of the child (child sex, age, birth order, relationship to the household head), maternal characteristics (mother’s height, age, schooling), household characteristics (size of land owned, livestock holdings), and village characteristics. A simple way to do this would be to regress a dummy variable for whether the child is stunted on the above mentioned variables. The dummy for child sex would tell you whether boys or girls have a higher probability of being stunted. But then you want to know whether it is mother’s education or father’s education that has a greater impact on child stunting. If your household roster has been completed, and you have information on paternal and maternal schooling, you can include both the father’s and mother’s characteristics in the regression (so, age, schooling, for both parents). You could also test whether paternal schooling or maternal schooling has differential effects on boys and girls by interacting (multiplying) the father’s schooling and mother’s schooling with the dummy variable for child sex. If this interaction term is significant, this means that there is a differential effect of parental schooling by child gender. You could also test whether paternal schooling has a different effect from maternal schooling.
You can examine the differential impact of all sorts of parental characteristics by child gender. In this paper on Ghana, we examine whether boys or girls whose mothers are more empowered (in various dimensions) have better nutritional outcomes. The results may surprise you! We interpret these results by appealing to qualitative work—illustrating that econometric analysis using sex-disaggregated variables benefits when we can also draw on work by other social scientists.
Q. But what if richer households just have children who are less likely to be stunted, regardless of child sex? How would I be able to control for household wealth?
A. The answer is by doing household fixed effects analysis, or within-household analysis. Most regression packages will allow you to specify a household specific dummy variable. Then you can examine the effect of variables that vary within the household, such as child sex and birth order. For example, if you still are interested in the effect of maternal characteristics, you can interact maternal characteristics with the dummy for child sex and see if the interaction term is significant. To be able to do this, you need to have observations on more than one child, one of each sex, from each household. There are many, many things you can do with sex-disaggregated data using regression analysis!
This post is part of a blog, the Gender-Nutrition Idea Exchange, maintained by the CRP on Agriculture for Nutrition and Health. To add your comments below, please register with Disqus or log-in using your Facebook, Twitter, or Google accounts. You must be signed-in or registered in order to leave a comment.
