Preexisting Differences in Iron Consumption Are Distributed Quasi- Randomly

reduced- form estimates of the program impact as coming solely from niacin would imply an unreasonable individual effect. Consequently, I focus on iron defi ciency, but at the very least the results can be interpreted as the total effect from reductions in defi ciencies of all three micronutrients.

B. Preexisting Differences in Iron Consumption Are Distributed Quasi- Randomly

The identifi cation strategy relies on the assumption that preexisting geographic dif- ferences in iron consumption across SEAs are not correlated with SEA heterogene- ity in omitted characteristics that induce changes in the outcome. I conduct a direct, albeit partial, test of the identifying assumption by individually regressing SEA aver- age iron consumption on several preprogram SEA characteristics. The assumption is supported if these characteristics do not predict iron consumption, and the exercise suggests specifi c controls if characteristics do have predictive power. 20 Point estimates provided in Table 3 suggest that iron consumption is uncorrelated with several eco- nomic, labor market, agricultural, and demographic characteristics. For example, the unemployment rate, New Deal spending, World War II mobilization rates, retail sales and manufacturing output are essentially unrelated to the preexisting differences in iron consumption. Exceptions include per capita war spending, the fraction of the population native born, net migration, and growth in median home values over the 1930s. The results, therefore, suggest adding specifi c controls to the regressions in the following empirical sections. The empirical strategy provides plausibly exogenous variation in health improvements during the 1940s when specifi cations include the additional controls. Table 4 explores the proximate cause of geographic differences in iron consumption and variation in the types and quantities of food consumed. Areas with an iron intake below the median of the sample consume less than areas above the median in every food category. However, some of this has to do with geographic differences in tastes for food. In Panel B, diets are broken out by regions of the United States. The South obtains a relatively high proportion of its iron intake from plant- based sources, which are more diffi cult for the body to absorb. The West and Northeast consume a relatively high amount of meat, which has a high absorption rate.

C. Outcome Data