Nonexperimental Designs

Nonexperimental designs are those research designs that lack manipulation of an independent variable and/or control of nuisance variables through random assignment into control or treatment groups. As a result, cause-and-effect relationships cannot be inferred from nonexperimental designs. This entry describes nonexperimental designs, discusses the reasons they are used, and gives recommendations for dealing with confounding variables.

Despite experimental designs typically being considered the gold standard within the context of science, researchers may opt for a nonexperimental design when manipulation of an independent variable and/or random assignment of study participants into groups is not possible, feasible, ethical, or of interest to the researcher. Nonexperimental designs do not allow for causal inferences to be made from relationships observed between variables. This should not, by itself, relegate [Page 1150]nonexperimental designs to being weaker, less rigorous, or incapable of making important contributions. They frequently serve a different purpose than experimental designs or are necessary given ethical, feasibility, or practicality constraints.

Common nonexperimental designs that use qualitative methodologies are case study, ethnography, and grounded theory. Common nonexperimental designs that use quantitative methodologies are comparative, survey research, and retrospective. It should be noted that there are so-called quasi-experimental designs that are not considered “true” experiments due to the lack of random assignment (with or without matching) but are still often considered strong designs within the scientific community. These designs include time series, regression discontinuity, and comparative design in which propensity score matching/analysis is incorporated.

Within the scientific framework, the importance of manipulation and random assignment to experimentation is clearly evident given that their absence is the distinguishing feature between designs classified as experimental and nonexperimental. As noted, these may be beyond the researcher’s ability, interest, or ethics. The remainder of this section discusses scenarios in which a nonexperimental design would be chosen.

First, a researcher may wish to make meaning and organize concepts from an extended series of observations to produce a theory, grounded in observation and experience, about how some social phenomenon might exist or occur. The design for such a study would likely involve purposeful sample selection and extended and extensive data collection via group or individual interviews, naturalistic observation, and/or review of artifacts (e.g., documents and archives) in order to develop a framework (i.e., theory) about the nature of a social phenomenon. The theory might later be subjected to additional investigation. This design involves neither manipulation of any variable(s) nor does it involve random assignment of participants into a control group or one or more treatment groups. As a result, cause-and-effect relationships cannot be inferred; however, potential explanations can be proposed and investigated further in future research.

Second, a researcher may wish to study the effect of being a victim of assault during college on degree completion. To carry out such a study, it would be necessary to collect, at minimum, degree completion data for students who had experienced assault. This approach would not allow the researcher to compare the degree completion rates of those students who had not been assaulted. An alternative approach would be to collect the degree completion rates from students who had experienced assault versus those who had not. An experimental design would require the researcher to randomly assign students to one of the assault conditions. Such a scenario is neither possible nor ethical; thus, a nonexperimental design would be necessary to examine the relationship between these variables.

...