Longitudinal surveys have revolutionized empirical research and our understanding of the dynamic processes that affect the economic prosperity, health and well-being of the population. This dissertation explores and provides evidence, through three empirical applications, on the costs and benefits of designing, implementing and using data from a new, innovative longitudinal survey, the Mexican Family Life Survey (MxFLS). The survey, which is representative of the Mexican population living in Mexico in 2002, is designed to follow movers within Mexico and also those who move to the United States. This design lies at the center of the contributions of my research to the scientific literature. Attrition is the Achilles heel of longitudinal surveys. The first essay of the dissertation focuses on the cost of attrition for scientific knowledge. Following the same individual through time allows a researcher to trace the evolution of a respondent's behaviors and outcomes in a dynamic framework; however, if attrition is selected on unobserved characteristics, the advantage of using panel data could be severely hindered. Exploring different methods to adjust for attrition, this essay provides evidence of limitations of standard post-survey adjustments strategies that are the standard in the literature. These approaches, exploit only baseline characteristics of the respondents and, conditional on those characteristics, treat attriters as missing at random. I provide evidence that this assumption is substantively important and rejected in the MxFLS in spite of the fact that attrition in that survey is low relative to other nationally-representative surveys conducted in the United States and abroad.