The use of panel studies, in which the same people are interviewed at least twice in two different time periods, provides researchers with the ability to explore a multitude of issues. Researchers can explore change over time, better establish temporal order of events, and have the option to use a wide array of models that are only possible with longitudinal survey data (Johnson, 1988; Toon, 2000). With more waves of data, however, comes the potential for more problems in data collection and analysis. Researchers may feel pressured to choose between question consistency and updating surveys for contemporary language or issues (Olsen, 2005).
For my dissertation, I explore how the detection and correction of attrition may be performed after data is collected. Although the prevention of attrition in the collection phase may be ideal, the proliferation of large and widely available datasets from the government and academia means that a considerable amount of research is performed by those without any ability to impact data collection. Given the rise of computing power and the increasing knowledge of multiple imputation and complex statistical models, such as fixed and random effects, today’s researcher may have an increased ability to perform necessary adjustments for the problems that plague panel studies. The central research question for this dissertation is: Given the presence of attrition, and the implications of attrition for biased estimates, what procedures will aid researchers in estimating valid findings?