Within-Subject Experimental Designs

Chapter 10 — Within-Subject Experimental Designs In a between-subject design, different units are assigned to treatment and control. In a within-subject design, the same unit is observed in both—a powerful but demanding approach. This chapter develops the potential outcomes framework for within-subject settings, introducing three new identification assumptions: balanced panel, temporal stability, and causal transience. It covers threats to each assumption, solutions including Latin squares, washout periods, and stealth designs, and shows why within-subject designs offer greater statistical power and access to the full distribution of individual treatment effects.

1. Within-subject designs are a key departure from between-subject designs in that the same unit is observed in treatment and control sequentially.
2. Within-subject designs require stronger identification assumptions than between-subject designs but have more statistical power and generate more information.
3. Balanced panel, temporal stability, and causal transience assumptions are three new identification challenges when using within-subject designs.
4. Within-subject designs can usefully be categorized based on whether these new assumptions are plausible.