Comparaisons multidimensionnelles de bien-être et de pauvreté ...

AbstractThe main objective of this thesis is to purpose a suitable statistical method for comparingwelfare when we deal with multivariate distributions. After a critical literature review onstatistical inference based composite hypotheses, the intersection-union (IU) type of formulationhas been retained for making unambiguous robust comparisons in the sense of strictstochastic dominance. Davidson and Duclos (2006) suggest in this way, a method using theempirical likelihood ratio for testing the stochastic dominance in the univariate distributionscontext. This method is extended by this thesis to the multivariate distribution, which is inphase with the recent literature on poverty and welfare analysis that advocates the use ofseveral dimensions to measure the welfare.The first chapter consists in analyzing the performances of the suggested method in bidimensionalcontext. This method, based on the maximization of an empirical likelihoodfunction, tests for the null hypothesis of non dominance against the alternative of dominance.The test statistic is pivotal, which enable to perform Monte Carlo simulations for analyzingthe size and the power of asymptotic and bootstrap tests.Once the test performances judged satisfactory, illustrations are made to study the stochasticdominance relations in poverty between some African countries. Two dimensions areconsidered for drawing the bivariate distributions : The nutritional status and an asset indexderived from factor analysis methods using DHS (Demography and Health Surveys) data.The third chapter consists in considering the case where one of the two dimensions is adiscrete variable. We then test for the sequential stochastic dominance relations in welfare