10/21/2009 Lecture 13: Disease # Bleakly paper (not required reading). Studied malaria. Does Malaria hold back economic progress? Specifically, does childhood exposure to malaria could have long-term impact. Why could this be? - Childhood is best time to accrue human capital/education. - Hard to pay attention - Miss knowledge now makes it harder to learn later on (on the job or in future grades. - Physical stunting due to malaria might cause later hindrances to malaria - Children's parents being infected also affects them, as it makes them poorer How to study this? Start off by comparing kids who have malaria to those that don't, and see their difference in output. - Have to control to make sure that no other differences can be attributed to their change in outputs in the future: income, education, geographic region. Difference in differences - Take a difference over time for regions that have DDT introduced to reduce malaria. - Take a difference over geographic regions that have varying likelihood of malaria locations. Problems: - if some invention led the location to grow more over time, the difference in output after DDT introduction might not be attributed to DDT - maybe the DDT spraying program employed more people, which led to more output So consider eradicating malaria in 1920's US south - Take data of wages as adults in 1970s: Y - First difference: compare adults born before (e.g 1902) to those born after (e.g. born 1921): Y_1921 - Y_1902 - Second difference: do the first comparison again, but across adults from two regions, with high and low rates of eradication: (Y^H_1921 - Y^H_1902) - (Y^L_1921 - Y^L_1902) - This happened in US South in 1920's, and then in brazil/columbia/mexico in 1950s with DDT First difference: time variation - Imagine eradication occurred overnight in 1950's brazil - You have data on all brazillians of all ages in 1980 - Who among these people got lots of childhood exposure to malaria? Who got no exposure? Who got intermediate exposure? | exposure | to | malaria | 1 | /--------------- | / | / | / | / 0 |------------ ----------------------------------------------- 30 48 age in 1980 Second difference: spatial variation - study malaria exposure in different regions - charts on slide 12 showing x axis=initial malaria exposure, y axis=change in malaria exposure after treatment. None are perfect (45 degree line, but there is a positive slope---eradication kills malaria where it is more popular) It turns out if you do that difference in differences, the data for income, literacy, ... looks a lot like the chart above. If you look at Mexico, the middle slope for years of schooling (unlike literacy) is sometimes positive (USA and Columbia), sometimes negative (Mexico). That's potentially because there's a fixed ammt. of seats in schools which were already full, or perhaps because childhood wages rise with malaria eradication causes parents to pull kids out of school. # Acemoglu-Johnson on disease + development Different from Beleakley paper - Much more macro---compare entire countries to other countries - Outcomes (GDP) measured in real time---not a cross-cohort analysis based on effect of childhood exposure on outcomes observed later (e.g. adult wages) - Differen tquestion: effect of eradication of _fatal_ diseases (i.e. raise life expectancy). So check effect of mortality instead of morbidity. Why would we expect to see (or not) an effect of improved life expectancy on GDP growth? - Y = K + L (output = capital + labor). So if you increase L due to life expectancy, then Y/L goes down as L goes up, since capital won't necessarily go up. So people have to share same ammt. of capital, and output per capita goes down. - morbitity (sickness) might also go down, which would contribute to higher output. Turns out that there's positive correlation between life expectancy and GDP. How could we test causality? - Instrumental variable analysis: find things that increase life expectancy, and don't increase GDP except through life expectancy. - Hard to devise randomized evaluation of this over such a long time frame, so instead you have to look at history. This paper exploits the epidemiological transition around 1940s. - dramatic improvement in: international health interventions, public health measures, introduction o fnew chemicals and drugs - Diseases such as TB, malaria, pneumonia receded - Compare countries depending on their pre-treatment disease mix. Authors construct measure of predicted mortality M_{it} = sum_{disease in set of 15}((1-I_{dt})M_{di40} + I_{dt}M_{dft}) -> sum of differences in disease death before and after 1940's treatment. So we are interested in causal effect of LE (life expectancy) on Y (GDP per capita) - Since LE and Y are correlated, but no idea about causality or some external ommited variable affecting both. - Introduce intrument variable M' such that: - M' is correlated w/ LE [testable] - The only reason that M' is correlated with Y is because M' shifts LE, and LE shifts Y [not testable!!! have to think hard about whether you believe that!] - First stage: correlation between LE and M' - Reduced form: correlation between M' and Y - If you believe M' only affects Y through LE, then ratio of effect size of first stage and reduced form gives you the effect of LE on Y. So they found that M' is negatively correlated w/ GDP per capita, by way of LE. However, has no effect on GDP (doesn't lead to overall growth). Homework: what is the IV estimate of life expectancy on growth (look at first stage and reduced form in figures).