Working Papers
Improving Estimates of Mean Welfare and Uncertainty in Developing Countries with David Newhouse. World Bank Policy Research Working Paper 10348. Revision requested, Review of Economics and Statistics.
[ Abstract | Draft ]
Reliable estimates of economic welfare for small areas are valuable inputs into the design and evaluation of development policies. This paper compares the accuracy of point estimates and confidence intervals for small area estimates of wealth and poverty derived from four different prediction methods: linear mixed models, Cubist regression, extreme gradient boosting, and boosted regression forests. The evaluation draws samples from unit-level household census data from four separate developing countries, combines them with publicly and globally available geospatial indicators to generate small area estimates, and evaluates these estimates against aggregates calculated using the full census. Predictions of wealth are evaluated in four countries and poverty in one. All three machine learning methods outperform the traditional linear mixed model, with extreme gradient boosting and boosted regression forests generally outperforming the other alternatives. The proposed residual bootstrap procedure reliably estimates confidence intervals for the machine learning estimators, with estimated coverage rates across simulations falling between 94 and 97 percent. These results demonstrate that predictions obtained using tree-based gradient boosting with a random effect block bootstrap generate more accurate point and uncertainty estimates than prevailing methods for generating small area welfare estimates.
Air pollution and agricultural productivity in a developing country. Revision requested, Journal of Development Economics.
[ Abstract | Draft | Most recent slides ]
I document negative externalities of air pollution in the Indian agricultural sector. Using variation in pollution induced by changes in wind across years, I show that higher levels of pollution lead to decreased agricultural productivity, with large changes in productivity being common. The negative effects of pollution are larger in areas growing more labor-intensive crops, indicating that the pollution works at least partly through direct effects on labor productivity. Finally, combining wind direction with the rollout of coal plants, results indicate that pollution from coal plants has a larger effect on agricultural productivity than other types of pollution. Given that the agricultural sector is a refuge for the poor in many developing countries, these results suggest that the negative externalities of pollution may hit the poorest particularly hard.
Gender and hiring in SMEs: Evidence from an RCT in Bangladesh with Emily Beam, Asad Islam, and Naveen Wickremeratne (funding: £225,000). Revision requested, Journal of Development Economics (pre-results review).
[ Summary coming soon. ]
Combining Survey and Geospatial Data Can Significantly Improve Gender-Disaggregated Estimates of Labor Market Outcomes (with David Newhouse, Michael Weber, and Partha Lahiri). World Bank Policy Research Working Paper 10077. Revision requested, Journal of the Royal Statistical Society: Series A.
[ Abstract | Draft ]
This article examines the extent to which combining survey data with publicly available geospatial indicators improves estimates of state and municipal labor force statistics in urban Mexico. Model-based estimates of labor force participation and unemployment are generated separately for men and women, using a population-weighted nested-error conditional random effect model following an arcsin transformation, specified at the level of the Área Geoestadística Básica (AGEB). Two types of hypothetical samples are used to estimate the model: a simple random sample of individuals within AGEBs selected using proportional to size sampling, and a full enumeration of all households within those same AGEBs. The resulting small area estimates are compared against results from the full census. Incorporating geospatial data improves the precision and accuracy of state-level estimates for all four indicators, despite the weak predictive power of the unemployment rate model. At the municipality level, small area estimates substantially improve on survey estimates of labor force participation. For unemployment rates, the results when using the simple random sample are mixed because of the large number of municipalities with no unemployed persons in the sample. Using the full enumeration sample greatly improves municipal predictions for all four indicators. These results are robust to the use of repeated simulations of alternative samples. Integrating survey data and publicly available geospatial indicators significantly improves the accuracy and precision of both state-level estimates and estimated municipal labor force participation rates at negligible cost, but accurately estimating low-probability events like unemployment with a linear model requires large samples within target areas.
Small Area Estimation of Monetary Poverty in Mexico using Satellite Imagery and Machine Learning (with David Newhouse, Anusha Pudugramam Ramakrishnan, Tom Swartz, and Partha Lahiri). Revision requested, Oxford Bulletin of Economics and Statistics.
[ Abstract | Draft ]
Estimates of poverty are an important input into policy formulation in developing countries. The accurate measurement of poverty rates is therefore a first-order problem for development policy. This paper shows that combining satellite imagery with household surveys can improve the precision and accuracy of estimated poverty rates in Mexican municipalities, a level at which the survey is not considered representative. It also shows that a household-level model outperforms other common small area estimation methods. However, poverty estimates in 2015 derived from geospatial data remain less accurate than 2010 estimates derived from household census data. These results indicate that the incorporation of household survey data and widely available satellite imagery can improve on existing poverty estimates in developing countries when census data are old or when patterns of poverty are changing rapidly, even for small subgroups.
Poverty at Higher Frequency (with Jonathan Morduch). Under review.
[ Abstract | Draft | Most recent slides ]
The poverty rate is an important focus of economic policy. We show, however, that in low- and middle-income countries, the poverty rate is often not what it seems. Poverty, as conventionally measured, is thought to be the proportion of households that are poor for the year, but we show that, under common data collection practices, the measure instead captures the average share of the year that households are poor. The resulting poverty rates are sensitive to the timing of household consumption, not just its total value. For policy, this means that, contrary to common assumptions, the *de facto* concept of national poverty in many countries is sensitive to households’ exposure to shocks and their ability to smooth consumption within the year. While created inadvertently, this *de facto* concept of poverty has appealing properties as a measure of well-being, and it raises new philosophical questions about the nature of deprivation. This transformation has happened without a change in the form of the poverty measures and without longitudinal data. Instead, the transformation follows from three common practices used when collecting household data: asking survey questions with short-term recall (often covering only the past week’s or month’s spending), stratifying on sub-periods (usually quarters), and surveying households only once during the year. We illustrate the implications with monthly panel data from rural India, showing that time-sensitivity in poverty measurement has quantitatively large impacts on measured poverty, improves predictions of health outcomes, and expands the scope of strategies to reduce global poverty.
Does the timing of the school year affect child labor decisions in developing countries? Under review.
[ Abstract | Draft ]
In developing countries, agricultural productivity shocks are an important predictor of the opportunity cost of time for children. This can lead to children dropping out of school to work during good rainfall years. However, this trade-off between current and future income is most salient only when the agricultural season and the school year overlap. In this paper, I show that this overlap is an important mediator of the effect of agricultural productivity shocks on both child labor and school enrollment. A long overlap between the harvest season and the school year leads to a lower elasticity of child labor with respect to agricultural productivity shocks relative to harvest season that does not overlap with the school year. The entirety of the effect is driven by self-employment, which is consistent with a story of children working on household farms.
Minimum wages and unemployment during economic shocks (with Matthew Sharp).
[ Abstract | Draft ]
This paper studies whether a minimum wage changes how labour markets respond to economic shocks. Using data from South Africa, we show that an agricultural minimum wage leads to higher mean wages with no significant impacts on mean employment. However, these positive aggregate outcomes hide important heterogeneity: the imposition of the minimum wage leads to substantial declines in employment -- especially overall hours -- in the sector in the wake of negative weather-related economic shocks, which typically exert downward pressure on wages. The increased variance of employment across years in the post-law period suggests caution in interpreting the overall welfare impacts of minimum wage laws.
Electricity, pollution, and structural transformation with Sunghun Lim.
[ Abstract | Preliminary draft | Most recent slides ]
We show how access to electricity drives structural transformation in India. Using village-level data from population and economic censuses, we document increases in manufacturing employment and decreases in agricultural employment following the opening of a coal-fired power plant near a village. We also show that these increases are driven by increases in employment in larger firms. Evidence suggests there are increases in both the availability and consistency of electricity. Importantly, we show that areas exposed to pollution from coal plants see decreases in access to electricity and decreases in population and literacy rates relative to less exposed areas, despite an increase in employment concentration in larger firms. These results suggest that access to electricity can be a driver of the structural transformation process, but that the resulting pollution can be an important mediator.
Works in progress
Structural Transformation in a Changing Climate with Rajeev Dehejia.
Improving Estimates of Human Capital with Lanu Kim, Peter Lanjouw, David Newhouse, and Michael Weber.
Climate change and agriculture across the world with Ariel Ortiz-Bobea and David Ubilava.
Roads and riots with Ashani Amarasinghe and David Ubilava.
