ABSTRACT
This chapter’s focus is on quasi-experiments, which are distinguished from true experiments by some form of experimental control but lack randomization of participants to conditions. This feature often precludes quasi-experiments from producing results, allowing for confident causal interpretation, but not always. Quasi-experiments are useful in evaluation and applied research contexts, where randomization may be neither feasible nor ethical. A common threat to interpretability in QE designs is the regression artifact, which is covered in some detail in this chapter with respect to its relation to measurement error, which circles back to the content of Chapter 3. Use of matching designs and propensity score analysis to offset lack of randomization is discussed, and the advantages of the latter are presented. The remainder of the chapter is devoted to consideration of different quasi-experimental designs, including nonequivalent groups designs, and ways of improving their interpretability, case-control designs, time-series designs (simple, multiple, and comparison), and regression discontinuity designs.
In the preceding chapters we distinguished experimental from correlational research, in which the investigator’s role is largely that of an observer. In a real sense, all variables in correlational studies are dependent (or response) variables, and the researcher’s job is to assess their natural variation and discern the patterns and relations among them. In experiments, the researcher actively intervenes in the normal relational pattern by systematically controlling variation in the independent variable (or variables) to assess the causal impact of this variation on outcome variables. Controlled variation of the (presumed) causal variable and random assignment of subjects to research conditions are the necessary hallmarks of true experiments, which are the backbone of internally valid cause-effect analyses.
