‘Metrics Monday: Data Cleaning

I spent a few days last week attending the annual meetings of the Agricultural and Applied Economics Association (AAEA) in San Francisco as well as a one-day pre-conference on agricultural value chains in developing countries. At that pre-conference, I happened to be sitting with Ben Wood, who co-manages 3ie’s replication program, and in light of his experience with replication in economics, he suggested I write a post about data cleaning.

Why data cleaning? Because most students will typically have very little experience with that thankless task, which happens because most econometrics classes usually present students with nice, picture-perfect data sets for applied problem sets (and that’s when those classes actually have students estimate stuff instead of just memorizing the properties of various estimators). Continue reading

Food Deserts and Child Obesity

A very nice new article titled “The Effect of Food Deserts on the Body Mass Index of Elementary Schoolchildren” by Thomsen et al. in the American Journal of Agricultural Economics, with emphasis added:

Families in low-income neighborhoods sometimes lack access to supermarkets that provide a broad range of healthy foods. We investigate whether these so called “food deserts” play a role in childhood obesity using a statewide panel data set of Arkansas elementary schoolchildren. We use fixed-effects panel data regression models to estimate the average food desert effect. We next compare children who left (entered) food deserts to children who were always (never) in food deserts and homogenize samples for those whose food desert status changed as a result of a change in residence and those whose status changed only as a consequence of the entry or exit of a supermarket. We present evidence that exposure to food deserts is associated with higher z-scores for body mass index. On average, this is in the neighborhood of 0.04 standard deviations. The strongest evidence and largest association is among urban students and especially those that transition into food deserts from non-deserts. Our food desert estimates are similar in magnitude to findings reported in earlier work on diet and lifestyle interventions targeting similarly aged schoolchildren. That said, we are unable to conclude that the estimated food desert effect is causal because many of the transitions into or out of food deserts result from a change in residence, an event that is endogenous to the child’s household. However, there is evidence that food deserts are a risk indicator and that food desert areas may be obesogenic in ways that other low-income neighborhoods are not.

This is the kind of article I really like, and for my money, the AJAE–the flagship journal of my professional association, the Agricultural and Applied Economics Association–is at its best when it offers such topical, policy-relevant contributions.

All in the Family: Explaining the Persistence of Female Genital Cutting in West Africa


I’ve discussed this paper often on this blog (see here, here, here, and here), but here goes one last time: My article on female genital cutting in West Africa with Lindsey Novak and Tara Steinmetz  is now published on the Journal of Development Economics website, and will be published in the November 2015 issue of the journal.

Here is the abstract:

Why does female genital cutting (FGC) persist in certain places but has declined elsewhere? We study the persistence of FGC—proxied for by whether survey respondents are in favor of the practice continuing—in West Africa. We use 38 repeated cross-sectional country-year data sets covering 310,613 women aged 15 to 49 in 13 West African countries for the period 1995–2013. The data exhibit sufficient within-household variation to allow controlling for the unobserved heterogeneity between households, which in turn allows determining how much variation is due to factors at the levels of the individual, household, village, and beyond. Our results show that on average, 87% of the variation in FGC persistence can be attributed to household- and individual-level factors, with contributions from those levels of variation ranging from 71% in Nigeria in 2011 to 93% in Burkina Faso in 2006. Our results also suggest that once invariant factors across women aged 15 to 49 in the same household are accounted for, women who report having undergone FGC in West Africa are on average 16 percentage points more likely to be in favor of the practice.

Once again, here is the key figure from the paper, which shows where support for FGC comes from in West Africa. Note that in most cases, the bulk of support for the practice comes from observed and unobserved factors at the household level which do not vary from one individual to another within the household as well as from individual-level factors, i.e., from levels beyond those of the community: