Models containing unobservable variables arise very often in economics, psychology, and other social sciences. They may arise because of measurement errors, or because behavioral responses are in part determined by unobservable characteristics of agents. To obtain meaningful statistical inferences for models containing unobservable variables requires either the expansion of existing data bases together with more precise measurements of variables of interest or the development of analytical frameworks to extract information from currently available data sources. The aim of this project is to continue developing analytical frameworks to enhance the reliability and validity of Federal statistical data. Attention is directed to cases where the use of latent variable models by researchers appears appropriate because of measurement errors, or because of the lack of measurable counterparts for the conceptual variables. The investigation centers on three issues (i) categorical or grouped data, (ii) the hedonic approach to demand analysis, and (iii) modelling unobserved heterogeneity in panel data analysis. The overall objective is to refine the methodology to establish empirically valid inferences. Professor Hsiao's research on modelling unobservable variables has been very fruitful. While his analytical approach, often requires the imposition of hypothetical assumptions, he has shown that it can yield useful information when properly conducted. Not only does the approach provide a means to take advantage of information that might otherwise be misused or totally ignored, it also can accomplish this desirable goal at relatively low cost.