This paper analyzes how the explained and unexplained part of the Oaxaca-Blinder decomposition has changed due to a shift in attitudes towards immigrants in Germany. We use the migration crisis in 2015 as an exogenous source of variation in attitudes towards immigrants. As many studies analyze labor market discrimination, there are very few studies examining discrimination in the German labor market by conducting a natural experiment. In contrast to previous empirics, this paper sets out to explain and decompose the differences between immigrants and natives using a rather unique method. This paper is based on data from the European Social Survey (ESS) conducted in 2014 and 2016 and is analyzed using measures of public attitudes, the linear probability model, and the Oaxaca-Blinder decomposition. The results indicate that there are rather great differences in employment probabilities, human capital and family constellations between immigrants and natives. Measuring public attitudes, it is evident that there has been a general decline in positive attitudes towards immigrants post the migration crisis in 2015. Analyzing the Oaxaca-Blinder decomposition, it is evident that the unexplained (discriminatory) part of the decomposition has decreased which is referred to as a decrease in discrimination in the labor market. Even though that a general decline in positive attitudes towards immigrants is present, an increase in the unexplained part of the decomposition is not observed. As no correlation between public attitudes and the unexplained part of the decomposition is apparent, it is most likely that employers base their employment decision on the extent of information provided. This indicating that discrimination in the labor market most likely is due to statistical discrimination and not due to the preferences of the employer.

Cette thèse est composée de quatre essais sur l'économétrie des mesures d'inégalité et de pauvreté. Elle fournit un traitement statistique fondé sur l'analyse de modèles probabilistes de mélange fini de distributions et de modèle de régression quantile, le tout dans une approche Bayésienne.Le deuxième chapitre s'intéresse à la modélisation d'une distribution de revenus par un mélange fini de lois log-normales dont les paramètres sont estimés par la méthode d'échantillonnage de Gibbs. Ce chapitre propose une méthode d'inférence statistique pour certains indices d'inégalité par une Rao-Blackwellisation de l'échantillonnage de Gibbs. Le troisième chapitre propose une estimation Bayésienne de la récente régression quantile non-conditionnelle basée sur la fonction d'influence recentrée (regression RIF) dans laquelle la densité est estimée par un mélange de lois normales. De cette approche, on déduit une inférence Bayesienne pour la méthode de décomposition d'Oaxaca-Blinder. La méthode proposée est utilisée pour analyser la dispersion des salaires aux Etats-Unis entre 1992-2009.Le quatrième chapitre propose une inférence Bayésienne d'un mélange de deux lois de Pareto simples pour modéliser la partie supérieure d'une distribution de salaires. Cette approche est utilisée pour analyser la répartition des hauts salaires aux Etats-Unis afin de tester les deux modèles (Tournoi et Superstar). Le cinquième chapitre de la thèse est consacré à l'analyse des rendements privés de l'éducation sur le revenu des ménages et des inégalités entre les populations urbaines et rurales. Il considère le cas du Sénégal et utilise les dépenses totales de consommation comme indicateur du revenu. / This dissertation consists of four essays on the econometrics of inequality and poverty measurement. It provides a statistical analysis based on probabilistic models, finite mixture distributions and quantile regression models, all using aBayesian approach.Chapter 2 models income distribution using a mixture of lognormal densities. Using the analytical expression of inequality indices, it shows how a Rao-Blackwellised Gibbs sampler can lead to accurate inference on income inequality measurements even in small samples.Chapter 3 develops Bayesian inference for the unconditional quantile regression model based on the Re-centered Influence Function (RIF). It models the considered distribution by a mixture of lognormal densities and then provides conditional posterior densities for the quantile regression parameters. This approach is perceived to provide better estimates in the extreme quantiles in the presence of heavy tails as well as valid small sample confidence intervalsfor the Oaxaca-Blinder decomposition.Chapter 4 provides Bayesian inference for a mixture of two Pareto distributions which is then used to approximate the upper tail of a wage distribution. This mixture model is applied to the data from the CPS ORG to analyze the recent structure of top wages in the U.S. from 1992 through 2009. Findings are largely in accordance with the explanations combining the model of superstars and the model of tournaments in hierarchical organization structures. Chapter 5 makes use of the RIF-regression to measure both changes in the return to education across quantiles and rural urban inequality decomposition in consumption expenditure in Senegal.