Collaborative Research: Modeling Sample Selection for Multi-Level Data Structures

Iowa State University, Ames IA

Investigators

Abstract

General Abstract An important part of political science research entails the analysis of events data in which the researcher accounts for when events occur, how often they occur, and tries to explain why they occur. Examples of interest to political scientists include topics like wars, mass protests, and coups, for instance. One of the problems associated with this type of research is that the event being observed is the product of a number of other factors or conditions. As a consequence, the researcher may not be able to identify the actual causes or the relative contributions of a set of causal factors for that outcome. Since the analysis of particular events is of such interest to political scientists, it is important to develop methods that allow researchers to draw the correct conclusions about how such events came to occur. The research team proposes to develop a statistical method to overcome the problems associated with events data analysis. The method has the potential to improve our understanding of how important types of events come to occur, making it of great interest to scholars across a range of social sciences. The research project will also provide a significant educational component as the team proposes to involve undergraduate students in the research experience. Technical Abstract The PI's note that much of the events data collected in the social sciences is the product of a structural data generation process that generally remains unanalyzed. As a result the observed events are correlated with a number of covariates. Statistical analyses of such data may lead to biased inferences if not appropriately addressed. These problems are of particular concern for small-n, or area-based research. The PI's propose to develop a theoretically-informed, statistical approach to how data is selected structurally. The project consists of three components. First, they develop an estimation procedure for repeated observations at higher levels of analysis with structural data. Second, they will produce work that bridges different periods of time-based aggregation across levels. Third, they extend the structural selection frame to move beyond two-level structures. Successfully developing the procedure will help to improve the collection and analysis of data, thereby helping to reduce the trade-off between thick descriptive approaches focusing on a few cases and thinly-informative operationalizations of variables in large-n studies. The work will have applications beyond political science, appealing to scholars in a number of social scientific disciplines. The project will provide important pedagogical benefits as the PI's will involve undergraduate students in the research.