The paper investigates the role of multi-level structures in poverty analysis based on household level data. We demonstrate how multi-level models can be applied to standard poverty analysis and highlight its usefulness in terms of assessing the extent community characteristics matter in determining poverty status and dynamics. We provide two applications. The first is an example of a growth model that control for characteristics measured at the initial time period, and considers directly to what extent the same characteristics contribute to explain changes in economic wellbeing over time. In the second application we model the determinants of escaping poverty. Both applications use longitudinal data from Vietnam recorded at two points in time during the nineties, a period where Vietnam experienced strong economic growth. We demonstrate that failing to control for multi-level data structures could give incorrect inference about the effect of covariates of interest. We also demonstrate how the multi-level models can be used for regional and community level policy analysis that otherwise is difficult to implement in more standard regression analysis.