Disease and Development: A Reply to Bloom, Canning, and Fink

Beginning in the 1940s, a wave of health innovations and more effective international public health measures led to a rapid and large improvement in health; for example, in some relatively poor countries, life expectancy at birth quickly rose from around 40 years to over 60 years. In Acemoglu and Johnson ð2006, 2007Þ, we constructed an instrument for these changes in life expectancy: “predicted mortality,” which is calculated from initial mortality by disease and the timing of global disease interventions. Across a wide range of specifications, our work suggests no positive effects—over 40or 60-year horizons—of life expectancy on GDP per capita ðor GDP per working-age populationÞ. Bloom, Canning, and Fink ð2014, in this issueÞ argue that the level of life expectancy in 1940 affected subsequent growth rates and should be included in our long-difference specifications; that is, the level of life expectancy in 1940 should be included on the right-hand side when 1940–80 or 1940–2000 changes in GDP per capita are the dependent variables. In a linear regression framework, their specification introduces a great deal of multicollinearity, and the standard errors become very large.