Call Us: US - +1 845 478 5244 | UK - +44 20 7193 7850 | AUS - +61 2 8005 4826

Stochastic dominance, poverty and the treatment effect curve

There is no meaningful range of sensible ‘poverty lines’ expressed in terms of changes in MUAC Z-scores. Therefore, we test for stochastic dominance over the entire domain rather than the typical righttruncated domain used in consumption or income poverty analysis. Our stochastic dominance based method for program evaluation has one potential disadvantage compared to regression-based difference-in-difference estimators. In the standard regression program evaluation approach we can include other covariates as right hand side variables. In practice, this doesn’t matter if we are primarily interested in whether the program has had an effect or not. Furthermore, our stochastic dominance method can be used to evaluate program impact net of other covariates; it just cannot do it simultaneously with estimating the program impact. To account for covariates we first run a regression of the outcome variable on the desired covariates before using the residuals, which represent the variation in the outcome variable net of observables, in the stochastic dominance estimation.