The Nuts and Bolts of Quasi-experiments

McBride Ch. 12

Quasi-Experiments
A type of research design where a comparison is made, as in an experiment, but no random assignment of participants to groups occurs.
Quasi-independent/subject variable: variable that allows comparison of groups of participants without manipulation
(i.e., no random assignment).
Many applied studies that examine realistic behaviors make use of quasi-experiments because random assignment would be difficult or impossible or unethical.
For example, research done in educational settings often involves a quasi-experiment, because the researcher cannot affect the classroom students are in or the teacher they have. Thus, different classes of students may be compared on a behavior of interest, but the researcher will not gain strong causal evidence because of the lack of random assignment. 
Pretest-Posttest Designs
A behavior is measured twice: once before a treatment or condition is implemented (the pretest) and once after it has been implemented (the posttest).
This design is a quasi-experiment because there is no random assignment of participants to the treatment. In this case, all the participants get the treatment. The researcher compares the scores from the pretest and the posttest, looking for a change based on the treatment or condition occurring between the two measurements.
Researchers can attempt to deal with some of the alternative explanations in pretest-posttest designs by including a control group.
If participants are randomly assigned to the control group and the treatment group, the study becomes an experiment. However, in many situations random assignment to groups may not be possible. Without random assignment, the study is still a quasi-experiment, where it is difficult to rule out alternative explanations of the results.
This design is a pretest-posttest design with nonequivalent groups, because group differences that might account for the results are not controlled by random assignment to groups.
Sources of Bias
History effects--events that occur during the course of a study to all or individual participants that can result in bias.
Example: A study was reported in the media during the time between measurements of smoking cessation that concluded that smoking has extremely harmful effects.
Testing effects--occur when participants are tested multiple times and each subsequent test is affected by the previous tests.
Example: participants may get better on the tests over time with practice. Alternatively, they may become fatigued or bored with the test after taking it once.
Regression toward the mean--can occur when participants score higher or lower than their personal average—the next time they are tested, they are more likely to score near their personal average, making scores unreliable.
Example: A high score achieved at posttest may be an extreme score in some cases, and with additional testing, these students may score closer to their original mean (which is typically lower than the norm).

Maturation--natural changes that occur to the participants during the course of a study that can result in bias.

Example: this can include actual maturation, as in developmental studies where individuals are observed at different ages, or other types of changes that occur to individuals over a period of time.
Attrition--occurs when participants choose not to complete a study.
Example: in the depression treatment study, suppose that some of the participants in the study drop out during the 12-week period and do not return for the posttest.
It is possible that these participants who have chosen not to complete the study did so because they have worse depression symptoms than the participants that remained in the study.

 

Questions to think about:

a) What aspects of a study allow tests of causal relationships?

b) Describe three main sources of bias that can affect the results of a quasi-experiment.