Chapter 4: Considered Designs

The focus of this paper is on two-level experimental designs in which students are nested within units (such as schools or classrooms) that are randomly assigned to either a single treatment or control condition. Two-level designs are considered here to keep the presentation manageable and because they are the most common designs used in education research. The two-level considered designs are as follows: Design I, where schools are the unit of random assignment; and Design II, where classrooms are the unit of random assignment and school effects are treated as fixed (which occurs in the common case where schools are purposively selected for the study and school effects are treated as fixed strata, so that the impact results generalize to the study schools only).

Finally, this paper also considers Design III, where students are the unit of random assignment and purposively-selected site (school or district) effects are treated as fixed. This is a nonclustered, stratified RCT design that is a special case (collapsed version) of the two-level designs discussed above. These designs are discussed in more detail in Schochet (2008).