This dissertation develops new methods for estimating treatment effects using panel data, with a focus on flexible and robust identification under treatment effect heterogeneity across units and over time. Chapter 1 provides an overview of the subsequent chapters. Chapter 2 introduces a simple time-series transformation—termed the Rolling Method—for Difference-in-Differences estimation. This method converts panel data into a sequence of cross-sectional datasets through unit-specific outcome transformations, enabling consistent estimation of group-time-specific treatment effects even in the presence of treatment heterogeneity. Chapter 3 extends the Rolling Method to small-sample settings, particularly when the number of treated or control units is limited, and demonstrates improved finite-sample inference properties. Chapter 4 further generalizes the framework to accommodate dynamic treatment paths, allowing for treatment reversals and subgroup-specific moderating effects. This extension is applied in an empirical case study examining the effects of the entry and exit of chain pharmacies in rural areas.
Read
