Category: Uncategorized

Simple regression analysis, the method of randomization, and the analysis of big data have been transforming development economics (Banerjee and Duflo 2009; Deaton 2010; Ray 2014; Varian 2014). This is truly welcome and has the potential to leave its mark on human well-being, growth, and development.

There is a risk, however, that this euphoria will once again have us carried away. We are seeing, especially in policy circles, these new empirical findings being quickly waved in front of our noses and treated as ground for doing whatever the policy maker wants to do. What is important to realize is that when we say that policy should be evidence-based, both words are important—“evidence” and “based.”We must not fall into the trap of evidence-waved policy. To see this mistake, consider the commonly heard policy refrain: “Recent data show 90% of jobs were created by the private sector. Therefore, we have to rely on the private sector for creating jobs.” The “therefore” is wrong. If it were not wrong, we would also have to go along with the Soviet economist who having studied Russian data in the 1980 s wrote: “Recent data show 90% of all jobs were created by the state. Therefore, we have to rely on the state for creating jobs.”

This is why we need the discipline of deductive reasoning, economic theory, and also common sense.

That is from Kaushik Basu and Andrew Foster–respectively, chief economist at the World Bank and editor of the World Bank Economic Review (WBER)–in an article titled “Development Economics and Method,” which serves as an introduction of sorts to a special issue of the WBER summarizing this year’s Annual Bank Conference on Development Economics.

I have been saying for a few years now that the pendulum will swing back, that theory will make a comeback in development economics in order to help understand the mechanisms whereby the effects observed in randomized controlled trials occur. It looks like the pendulum is on its way back.

(January 16, 2016 Update: If you came here from the New York Times website, thank you for your visit, and please note that the findings discussed below have changed slightly due to our incorporating two more years of data since this was posted last summer. Our new findings will be presented and discussed in an updated version of the working paper discussed in this post, which I am hoping to release before the end of March 2016. In the meantime, some of those new findings are discussed in the New York Times article.)

When I arrived at the University of Minnesota in the fall of 2013, a few colleagues and I applied for a seed grant from the university’s Healthy Foods, Healthy Lives Institute by submitting a proposal to look at the impact of local and organic foods and food safety.

After working on it for almost two years, I am happy to finally be able to circulate my new paper titled “Farmers Markets and Food-Borne Illness,” coauthored with my colleague Rob King and my student Jenny Nguyen, in which we ask whether farmers markets are associated with food-borne illness in a systematic way. In order to answer that question, we use a US state-level panel data set for the years 2004, 2006, and 2008-2011 (i.e., the years for which we had a full data set).

Here is the abstract:

We study the relationship between farmers markets and food-borne illness in the United States. Using a state-level panel data set for the period 2004-2011, we find a positive relationship between the number of farmers markets per capita on the one hand and, on the other hand, the number of reported (i) outbreaks of food-borne illness, (ii) cases of food-borne illness, (iii) outbreaks of Campylobacter jejuni, and (iv) cases of Campylobacter jejuni. Our estimates indicate that a 1% increase in the number of farmers markets is associated with a 0.7% (3.9%) increase in the total number of reported outbreaks of food-borne illness (Campylobacter jejuni), and a 3.9% (2.1%) increase in the total number of reported cases of food-borne illness (Campylobacter jejuni) in the average state-year. Our estimates also suggest that a doubling of the number of farmers markets in the average state-year would be associated with an economic cost of over $900,000 in additional cases of food-borne illness. When controlling simultaneously for both the number of farmers markets and the number of farmers markets that accept SNAP benefits (i.e., food stamps), we find that they are respectively associated positively and negatively with reported food-borne illness outbreaks and cases. Our results are robust to different specifications and estimators, and falsification and placebo tests indicate that they are unlikely to be spurious.

This post is not about Gladwellian pabulum. Rather, it is about the econometric problem posed by outliers, whose presence of extreme-valued observations in a data set whose presence might cause problems of estimation and inference, and which a few colleagues have asked for a ‘Metrics Monday post on a few weeks ago.

Outliers cause estimation problems because they bias point estimates. They cause inference problems because they cause standard errors to be too large, thereby making it more likely that one will fail to reject a false null, i.e., a type II error. For example, if you collect data on a random sample of the population, the bulk of the people in your data might be between 18 and 80 years old, but you might also have someone in there who is 110 years old–that person is an outlier. Or the bulk of your sample might be making between $30,000 and $300,000 a year, but you might also have someone in there who makes $200,000,000 a year–that person is also an outlier.