We wrote a short summary for @voxdev.bsky.social here 👉 voxdev.org/topic/energy...

The full paper is now out in the Journal of Development Economics 👉 www.sciencedirect.com/science/arti...

This kind of evidence matters. If we care about protecting the Amazon while supporting development, we need better tools to measure the real impacts of infrastructure -- and design smarter policies.

The big takeaway? It’s not just about what happens near the road/railroad/port itself. The deforestation footprint is spatially complex, spreading in ways that standard impact assessments often miss.

In our new paper, just published in the Journal of Development Economics -- with the brilliant Juliano Assunção and @arthurbraganca7.bsky.social -- we look at how roads, railroads, and waterways affect deforestation in the Amazon.

We highlight that our proposed methodology can be used in many empirical contexts. To employ our partial identification strategy, we only require an exogenous mismeasured treatment variable and two exogenous proxies.

However, our method reaches very different conclusions when compared to the findings of Nunn and Qian (2011) and Acharya et al. (2016). In particular, we find that the impact of agricultural productivity may be smaller than previously reported.

To illustrate our methodology, we reevaluate the results of three empirical studies: Nunn and Qian (2011), Bustos et al. (2016) and Acharya et al. (2016).

We find that the results by Bustos et al. (2016) are robust to measurement error.

We derive bounds around the true linear effect of crop productivity on the outcome of interest. These bounds exhaust all the information contained in the first two moments of the data distribution and have intuitive closed-form solutions.

We propose to partially identify the effect of crop productivity in a linear model with two proxies that are subject to nonclassical measurement error.

We impose that the true unobservable productivity is exogenous and that its effect on the outcome is non-negative.

We document the possibility of measurement error in the GAEZ variables. Their versions differ with respect to calibrated parameters and sources of data. These differences cause large changes in productivity for many crops and regions, suggesting the existence of mismeasurement.

Measurement error in the FAO-GAEZ data is rarely recognized in the empirical literature and is not properly addressed by current statistical methods. We propose a novel method that partially identifies the effect of crop productivity while accounting for measurement error.

The impact of crop productivity on economic outcomes is important in many fields. The most common measure of crop productivity is the FAO-GAEZ measure. However, this measure is based on a prediction model and is subject to measurement error.