This dissertation comprises three essays on the micro-foundations voters' support for welfare policies. In the first paper, I propose a hdpGLM to deal with context-dependent latent heterogeneity in the effect of observed covariates. The model is motivated by the problem of investigating how people’s socioeconomic characteristics affect their support for redistributive policies. The problem is that (1) many variables that condition the effect of observed socioeconomic characteristics of the individuals are latent or remain unmeasured, and (2) those latent conditioning features vary from country to country. The omitted variables can produce latent heterogeneity in the effect of observed covariates not only across countries but within each country as well. The proposed model allows us to investigate that context-dependent latent heterogeneity. It is general enough to be used with any hierarchical data in which the latent heterogeneity in the effect of lower-level characteristics is a function of upper-level features of the contexts, such as schools, hospitals, countries, or other institutional settings. The paper also shows how the model can be used to investigate the occurrence of Simpson’s paradox in the context of generalized linear models. In the second paper, I apply the hdpGLM approach to investigate the latent heterogeneity and polarization in voters' redistributive policy preferences. Previous studies only investigated polarization among observed socioeconomic groups (rich versus poor, white versus non-white), overlooking or ignoring within-group latent polarizations and cross-groups latent coalitions in policy preferences. The major contribution of the paper is a polarization model that uses the hdpGLM approach, accounts for latent heterogeneity in the determinants of redistributive preferences, and demonstrates how that heterogeneity in the effect of observed covariates can lead to different latent structures of polarization in different political contexts. In the third and final paper, I dive into the theoretical question of the association between socioeconomic positions and welfare preferences, which is the topic that motivated the first two methodological papers. The theoretical question I address is: What emphexplains the negative association between socioeconomic positions and welfare policy preferences? That is, what is the mechanism that links socioeconomic positions to welfare policy preferences? The predominant explanation for that association is that people evaluate welfare policies from the point of view of their material self-interest. Looking at their own pocket, low-income groups see welfare policies as benefits for themselves, and high-income groups see that as cost and constraint to their consumption power. Based on political sociology and socially situated cognition literature, this paper questions that mechanism and argues that socioeconomic positions affect redistributive preferences in part because it affects perceptions about the socioeconomic environment and people’s cognitive patterns of outcome attribution. Lower classes are more affected by the constraints imposed by external conditions, so they tend to attribute outcomes to exogenous forces rather than to individuals' agency, and they develop a more pessimistic perception about country’s economy and unemployment. Those perceptions about the socioeconomic environment, affected by socioeconomic positions, affect redistribution preferences. The empirical analysis is based on a series of structural equations estimated using cross-national data from the European Social.