We propose a method for aggregating evidence from observational studies, which may be subject to internal selection bias, and randomized controlled trials (RCTs) which may be subject to site selection bias. We show that it is possible to nonparametrically debias both types of studies using an Instrumental Variables (IV) strategy, uncovering the true distribution of treatment effects for complier studies. As we often have a small number of studies, imperfect instruments, and observe study results with error, parametric hierarchical Bayesian models work well in practice. Our specific implementation uses the presence of a facilitating organisation such as the Jameel Poverty Action Lab (JPAL) or Innovations for Poverty Action (IPA) as a novel Differences in Differences “Plausibly Exogenous” IV. This model point-identifies the internal selection bias for studies switching to the experimental design due to JPAL-IPA entry which is the negative local average treatment effect (LATE) in our setting. Applying this strategy to Conditional Cash Transfers (CCTs) and Microcredit shows substantial internal selection bias in observational studies of CCTs and much less for Microcredit; neither application shows evidence of RCT site selection bias, though credible intervals are wide with minimal updating.
We derive a formal, decision-based method for comparing the performance of counterfactual treatment regime predictions using the results of experiments that give relevant information on the distribution of treated outcomes. Our approach allows us to quantify and assess the statistical significance of differential performance for optimal treatment regimes estimated from structural models, extrapolated treatment effects, expert opinion, and other methods. We apply our method to evaluate optimal treatment regimes for conditional cash transfer programs across countries where predictions are generated using data from experimental evaluations in other countries and pre-program data in the country of interest.
Developing country cities are characterized by informal housing–slums–but as incomes grow their governments will pursue a host of urban renewal policies that promote the construction of modern, formal sector housing. This paper examines spatial spillovers from urban renewal using a unique policy experiment in Mumbai that led 15% of central city land occupied by the city’s defunct textile mills to be redeveloped during the 2000s. We digitize a host of new spatially disaggregated datasets on population, employment and house prices, and provide the first application of a deep convolutional neural network to measure changing slum cover from daytime satellite imagery. We find reduced form evidence of sizable spatial spillovers that impact surrounding locations by (i) increasing formal sector house prices and reducing slum cover, (ii) reducing informal employment density with no increase from the formal sector and (iii) increasing the share of high-skill residents and reducing population density. We disentangle the source of these spillovers by developing a quantitative urban model with formal and informal land and labor markets, and use it to quantify the equity-efficiency trade-off associated with slums and urban renewal policies.
How informative are treatment effects estimated in one region or time period for another region or time? In this paper, I derive bounds on the average treatment effect in a context of interest using experimental evidence from another context. The bounds use information embedded in the experimental outcome distributions and observational data from the context of interest to address treatment effect heterogeneity due to unobservables whose distribution can differ across contexts. Empirically, I use results from an experiment on returns to cash transfers given to microentrepreneurs in Leon, Mexico to predict average returns among microentrepreneurs in other Mexican cities. I show that the benchmark extrapolation method from the literature using treatment effect heterogeneity due to observed covariates yields implausibly precise predictions for other cities considering the very small experimental sample. Using data from a pair of remedial education experiments carried out in India, I show the bounds are able to recover average treatment effects in one location using results from the other while the benchmark method cannot.
Indian Labor Regulations and the Cost of Corruption: Evidence from the Firm Size Distribution
(with Amrit Amirapu)
Review of Economics and Statistics, March 2020.
Final draft, VoxDev article, Code and data
In this paper, we estimate the costs associated with a suite of labor regulations in India whose components have gone largely unstudied in developing countries. We take advantage of the fact that these regulations only apply to firms above a size threshold. Using distortions in the firm size distribution at the threshold together with a structural model of firm size choice, we estimate that the regulations increase firms’ unit labor costs by 35%. We document a robust positive association between regulatory costs and exposure to corruption, which may explain why regulations appear to be so costly in developing countries.
Selected work in progress
How Do Environmental Firm Location Policies Affect Workers, Firms, and Environmental Quality?
(with Namrata Kala)
(with Keisuke Hirano)