Essays in Development Economics Open Access
Downloadable ContentDownload PDF
In this dissertation, I write three essays in Development Economics. In essay 1, I provide robust evidence on the effects of BRAC's Targeting the Ultra-Poor Program in Bangladesh (TUP). In addition to BRAC's own classification, I exploit type-1 errors in assignment (leaving eligible participants outside the program) to create an alternative treatment-comparison pair. This allows me to estimate the program effects on the target group not contaminated by mistarget-ing. The panel structure of the data enables me to eliminate the time-invariant unobserved heterogeneity at household level (i.e. innate entrepreneurial ability and risk preference) by using household fixed effects. Moreover, I carefully address selection on unobservables by implementing the heteroskedasticity based identification technique developed by Klein and Vella (2009) and two additional recently developed matching estimators. The results show that program participation had significant positive effects on food security, clothing, shoes, livestock and cash savings, but there is weak or no evidence of a significant impact on a number of household durables and assets, and on indicators of health and women's empow- erment. Furthermore, when one takes into account the differences in initial conditions, the effects of the TUP program on the alternative treatment-comparison group are much larger (as measured by the program effect normalized by the initial mean value of an outcome).In essay 2, I examine the design of social program targeting mechanisms and corresponding evaluation frameworks when poverty has multiple dimensions. I examine the sensitivity of the group identified as poor to the type and number of screens used, connecting multidimen- sional targeting to evaluation criteria. I apply this methodology to an assessment of Phase I of BRAC's Targeting the Ultra-Poor Program in Bangladesh, comparing characteristics of selected participants using alternative participation criteria, and compare program outcome measures associated with different targeting mechanisms. I relate this analysis to the mul- tidimensional poverty measurement technology developed by Alkire and Foster (2011), and show that the results represent an approach to conducting multidimensional poverty evaluation that parallels their framework. The approach offers an alternative way to examine the heterogeneity of the program impact across poverty levels. Findings of the application confirm that the BRAC TUP program has a significantly larger impact on health-related out- comes for the less poor households. On the other hand, I find that the poorest households (k = 5) have a larger impact than the less poor households (k = 1) on the net income increase variable (3481.95 Bangladeshi taka vs. 1759.97), on the probability of having a roof of good quality (0.24 vs. 0.13), on food availability (1.02 vs 0.67), on the probability of having meals twice a day (0.58 vs 0.37) and on the probability of owning shoes (0.27 vs. 0.15). Had the program concentrated on the poorest households, the average program impacts would have been larger in magnitude.In essay 3, I employ unconditional quantile-decomposition methods to conduct a careful accounting exercise analyzing the gender wage gap in the urban sector of twelve Latin American countries (Argentina, Bolivia, Brazil, Colombia, Costa Rica, Chile, Honduras, Mexico, Paraguay, Peru, Uruguay and Venezuela). Unconditional quantiles allow the researcher to compute marginal effects. By contrast, conditional quantiles only calculate wage effects for a subgroup with specific combination of years of education and experience. The data come from harmonized household surveys that contain information on employment status, wages and household's demographic characteristics. I control for the effect of individual characteristics (education and experience) and decompose the gender wage gap into an explained (due to differences in the endowment levels by gender) and an unexplained component (when the same endowment level is paid differently according to which gender the individual belongs) using a recent econometric technique developed by Firpo, Fortin and Lemieux (2009). This technique provides the user with asymptotic results instead of relying on simulations procedures as done by the method developed by Machado and Mata (2005) -based on conditional quantiles and resampling- which results in a loss of efficiency for a small number of calculations or in a computation burden with a large number. More importantly, by reweighting the data, a counterfactual distribution is generated and it is thus possible to calculate unconditional quantiles and therefore account for the contribution of each explanatory variable into the wage gap. I find that the wage gap is larger at the extremes of the distribution, which suggests the presence of sticky floors (defined as a gender wage gap favorable to males which is larger at the tenth percentile than at median levels) and glass ceilings (defined as a gender wage gap favorable to males which is larger at the ninetieth percentile than at median levels). The former are more frequent. Second, I find that the magnitude of the sticky floors is generally larger than that of the glass ceilings. Third, working women are more educated than working men all along the wage distribution, which partially hides the existent and `unexplained' pay difference. Thus, my estimates provide a lower-bound for the true gender wage gap. Fourth, the size of the gender wage gap at the bottom is highly correlated with measures of economic development, per capita GDP and income inequality (Gini). With only twelve observations, I am just suggesting some correlations. Nevertheless, I do find suggestive evidence that poorer countries and countries with higher levels of income inequality have higher unexplained gender wage gap differentials at the tenth percentile of the wage distribution. On the other hand, the unexplained portion of the wage gap at the top of the wage distribution (90th percentile) is larger in richer countries and in countries where income distribution is more even, although not statistically significant on the latter.