As a warning, I’m going to talk about methods again for part of this post because I can’t seem to help myself, but I will actually make a recommendation by the end, so there is a payoff if you bear with me (or if you skip to the last couple paragraphs).

Within the first few weeks of any semester that I am teaching, the odds are good that I have repeated the phrase, “correlation does not imply causation” to my students enough times that it has become obnoxious. There is a good reason I do this, though. It’s because in the wider world, people continue to make causal inferences based on correlational studies, and I’m trying to inoculate my students against fallacious reasoning. I mention this here because when we are trying to identify features of group projects and collaborative assignments that make them effective, we are inherently trying to make a causal claim, and a lot of the research that exists on this topic is not well-suited for this.

In my last post, I mentioned two studies that showed contradictory results for how method of group assignment (self- vs instructor-selected) relates to the subjective experience that students have in groups. Both studies used a retrospective survey, in which they asked people to describe their best and worst experiences working in a group for a school project. One found that self-selected teams made up 59% of the best experiences and 41% of the worst experiences (Bacon et al., 1999). The other found that self-selected teams made up 22% of the best experiences and 40% of the worst experiences (Fiechtner & Davis, 1984). Finally, despite both of these studies being correlational, both made explicit (contradicting) recommendations to instructors that assume a causal interpretation of their results. (Oddly, though the studies used very similar methodologies and were both published in the Journal of Management Education, the 1999 study did not include any reference to the 1984, so the contradictory nature of the results was not addressed.) One went so far as to say, “At this point we are confident we can identify a set of tactics that…when employed in combination will virtually ensure that learning groups will be counterproductive,” followed by a list titled “What Not To Do” (Fiechtner & Davis, 1984, pg. 70).

While we can rule out the possibility that the students’ subjective experience of the group at the end of the project influenced whether the group was self-selected or instructor assigned, it’s still possible that some other variable influenced both method of group assignment and subjective experience in a way that has little to do with group composition. For example, in the Fiechtner and Davis (1984) study, it could be that instructors who were more hands-off overall, were more likely to let students self-select groups and were also more likely to engage in other behaviors that contributed to a worse experience (such as providing limited or vague instruction and less opportunity for feedback).

Even if we assume that the causal explanation put forward by the authors of the two articles is accurate, we are still left with contradictory results. There are a few different reasons I can imagine for why two studies that were so similar would produce opposite results and I want to mention two here that have potential implications for researchers and instructors; one is methodological and the other is conceptual.

The methodological concern is that both studies may have had problems with their samples. In the Fiechtner and Davis article, participants were asked which courses the best and worst group experiences were in, and a little over 30% of the best group assignment experiences came from an Organizational Behavior class. Could it be that people who teach about organizational behavior might have some insight into how to set up effective groups? It seems possible. If that is the case, the results could in part be an artifact of one very good teacher. Similarly, the study by Bacon and colleagues had a restricted sample. The researchers used an in-class survey administered in a set of first and second-year MBA courses, so preexisting cohorts existed within the sample (by course section and by year). This could lead to a similar effect, where a common group experience shared by multiple participants disproportionately impacts the results. Unfortunately, in neither study are the data and analyses presented in a way that would allow us to test these possible alternative explanations

A potential conceptual explanation for the varied findings may be the presence of a moderator. That is, the relationships in the studies may be contingent upon some other variable. In this case, that other variable may have to do with the class composition prior to group formation. In particular, it is possible that group self-selection is preferred by students if they are in a class with several friends, but not in a class where they are unacquainted with their peers. This point is highlighted by comments made by participants in the Fiechtner and Davis study and in a more recent study that used a similar reflective survey approach to learn about students’ experiences with group projects:

“We got to choose our groups and I was the only one not in a sorority. I felt left-out all semester”

(Fiechtner and Davis 1984, p. 61).

“If there was someone I knew, or a good friend, the advantage would be to choose that person. But if you don’t know anyone in your class, which is me for most classes, then it’s going to be harder.” 

(Rusticus and Justus 2019, p. 450)

“It’s a shot in the dark. I knew no one in the class.” 

(Rusticus and Justus 2019, p. 450)

The point is that self-selected groups can be harmonious and pleasant experiences in some circumstances, but they also have the potential to create scenarios reminiscent of being picked last for dodgeball. Not only is the selection process unpleasant in those situations, but once groups are formed, there is the potential for cliques of preexisting friends to establish within groups, and newer students could be made to feel isolated.

Bacon and colleagues (1999) attempted to explore this possibility in their data by analyzing the experience of first-year and second-year students separately. They found that self-selection was the preferred method of group assignment by students in their second year of the MBA program, but this preference was dramatically reduced when they examined data from students who were only in their first quarter of the program (and therefore were unlikely to know many other people when the groups were created).

So, we have our first qualification that instructors should consider when determining how to assign groups. What relationships exist in the class? In classes where students are unlikely to know each other, self-selection processes may be stressful, and they may not result in greater student satisfaction. A challenge for instructors is that it can be difficult to know what the existing relationships are like within a class, and even in upper-level classes where students may be more familiar with each other, problems may arise.

A compromise might be a strategy that blends features of self-selection and instructor selection. For example, Mahenthiran and Rouse (2000) tested a method of group assignment that involved having students identify one other person that they wanted to be in a group with. These pairs of students were then combined by the instructor (randomly) to form groups of four or six. They found that the groups comprised of student-selected pairs performed better on a class project than groups that were created entirely through random assignment. Satisfaction with the group experience was also marginally higher among students in the paired groups. Further examination of the project grades found an interesting potential moderator in student GPA. That is the difference in project grade was driven by the improved performance of students who had lower GPAs.

Based on the one study they conducted, there is no way to know exactly why low-GPA students benefited so much from this hybrid of self-selection and random assignment. An unappealing but plausible explanation is that these students did not learn more, but they were carried by the friend in the group, whereas the students in the randomly assigned groups were less willing to take on the work of a lower performing student. However, an appealing, and equally plausible interpretation might be that having even a single pre-existing relationship with someone in the group gave the lower-performing student someone to turn to within the group to help them learn. In a way, this explanation represents an ideal scenario in collaborative learning – students learning from each other.

Instructors and researchers alike might benefit from continuing to explore hybrid grouping strategies like the one utilized by Mahenthiran and Rouse, which capitalize on the possible benefits of students being in groups with people they know and like, while reducing the stress of having to come up with a full group, and increasing the likelihood of some degree of heterogeneity within the groups.


  • Bacon, Donald R., Kim A. Stewart, and William S. Silver. 1999. “Lessons from the Best and Worst Student Team Experiences: How a Teacher can Make the Difference.” Journal of Management Education 23(5): 467–488.
  • Fiechtner, Susan Brown, and Elaine Actis Davis. 1984. “Why Some Groups Fail: A Survey of Students’ Experiences with Learning Groups.” Journal of Management Education (4): 58-73.
  • Mahenthiran, Sakthi, and Pamela J. Rouse. 2000. “The Impact of Group Selection on Student Performance and Satisfaction.” International Journal of Educational Management 14(6): 255–264.
  • Rusticus, Shayna A., and Brandon J. Justus. 2019. “Comparing Student- and Teacher-Formed Teams on Group Dynamics, Satisfaction, and Performance.” Small Group Research 50(4): 443-457.

David Buck, associate professor of psychology, is the 2020-2022 Center for Engaged Learning Scholar. Dr. Buck’s CEL Scholar project focuses on collaborative projects and assignments as a high-impact practice.

How to cite this post:

Buck, David. (2020, September 15). A Complicated and Qualified Recommendation for Creating Groups [Blog Post]. Retrieved from