To them, "true" experiments are studies in which subjects are randomly allocated into "experimental" and "control" groups. The former receive a "treatment," such as an educational program, while the latter do not. Randomization allows the researcher to assume the similarity of groups at the beginning of the treatment, with known statistical chances of error, and to avoid a variety of "threats" to the validity of the conclusion that the treatment "caused" any found differences between the groups in post-treatment behavior or other outcome. In contrast, "quasi-experiments" (most other forms of research), which do not use randomization, require a variety of additional, often heroic, assumptions in order to make causal inferences (see Lieberson 1985 for an argument that these assumptions may be generally unacceptable).
A Model
Consider an experiment to test the proposition that increasing the sensitivity to Hispanic cultures of non-Hispanic second-grade teachers will improve the performance of Hispanic children in their classes, particularly students with poor English-language skills. Teachers are randomly assigned to attend one of three seminars on Hispanic cultures that vary in intensity (experimental treatments) or a seminar on Asian culture (control). Following the approach of Alwin and Tessler (1974), Figure 1 presents a model of this experiment.
This is a free page. This page contains 186 words. This
article contains 2,945 words (approx. 10 pages at 300
words per page).
Read the rest of this Article with our Experiments Access Pass.
