# Drought and Death in Northeast Brazil

As an outside observer of Brazil, it’s hard to be optimistic. Who can pick just one reason? The country is struggling to emerge from a historically severe economic crisis. The president is historically unpopular, credibly accused of corruption, and lacks democratic legitimacy. The political class has been utterly discredited by the ongoing Lava Jato scandal. A front runner for the upcoming presidential election is a Duterte-like authoritarian demogogue. One could go on.

It’s in this depressing context that I made the following plot in the course of some work on the poltiics of drought relief in the most drought prone areas1 of Northeastern Brazil:

The plot infant mortality rate by month is calculated from government administrative records.2. As the plot makes clear, infant mortality more than halved over the period. To state the obvious, this is an amazing transformation for one of Brazil’s poorest regions.3

The starkness of this change led me to revisit Rocha and Soares (2015), one of my favorite recent papers in Brazilian development economics. They use rich microdata and finely-grained panel data to demonstrate a tight link between drought and birth outcomes in the Northeast semi-arid zone. In other words, drought systematically led to infant deaths and other negative health outcomes in this region. It’s a great model of how to use administrative panel data to make causal inferences.

Given the huge decline in infant mortality in the region, I wondered if the relationship they documented using data from 1996 to 2010 still exists. Using the same data sources on infant mortality, but a different4—though quite similar—source of data on rain, I replicated their analysis using a very similar method.5. The estimates of the effect of rain on infant mortality in various time periods are presented in the following plot:

When averaging over all years, we replicate what Rocha and Soares (2015) found: infant health outcomes respond to fluctuations in the weather. But note the fairly extreme heterogeneity by time period: large effects in 1996-2001 and negligible effects in the later periods.

The pace of change is evident when I disaggregte the data further. In the plot below, I estimate the drought effect every 2 years: The plot makes clear that there was a real transformation in the human consequences of weather in the late nineties and first half of the 2000s. And this change isn’t because of better weather: one of the worst droughts in the region’s history just ended this year. I don’t precisely know what caused this change, though expanded social programs and economic growth are likely reasons.

So yes, Brazilian democracy is in a bad state. But data like this help me temper the despair that the daily headlines bring.

## References

Harris, IPDJ, Philip D Jones, TJ Osborn, and David H Lister. 2014. “Updated High-Resolution Grids of Monthly Climatic Observations–the Cru Ts3. 10 Dataset.” International Journal of Climatology 34 (3). Wiley Online Library: 623–42.

Rocha, Rudi, and Rodrigo R Soares. 2015. “Water Scarcity and Birth Outcomes in the Brazilian Semiarid.” Journal of Development Economics 112. Elsevier: 72–91.

1. Technically, the “Semi-Arid” zone, which is a government designation for areas most prone to severe drought.

2. Total births are calculated using the microdata from the Brazilian National System of Information on Birth Records (Datasus/SINASC). Infant deaths are calculated from microdata provided by the Brazilian National System of Mortality Records (Datasus/SIM).

3. To get a sense of the history of the drought problem and its consequences for human development, I recommend Albert Hirschman’s chapter on the region in Journeys towards Progress. It’s of course out of date, but Hirschman’s prose makes it worth it.

4. Rocha and Soares (2015) use precipitation data generated by Kenji Matsuura and Cort J. Willmott. I use very similar data described in Harris et al. (2014).

5. Specifically, I estimate the following model: $H_{iyt}=\alpha+\beta R_{iyt} + \psi_{it} + \lambda_y + \rho T_{iyt} + \epsilon_{iyt}$ where $H_{iyt}$ is the infant mortality rate, $R_{iyt}$ is the rainfall variable, $\psi_{it}$ is a fixed effect for the municipality $i$ and calendar month $t$, $\lambda_y$ is a year fixed-effect, $T_{iyt}$ is a linear municipality-specific time trend, and $\epsilon$ is the error term (clustered on municipality). Rocha and Soares (2015) also control for municipality-specific average temperature.