Group work in undergraduate STEM classes

Group work is a key component of several different evidence-based pedagogical strategies, including course-based undergraduate research experiences [2], peer instruction [3], many active learning strategies, such as think-pair-share [4], and constructivist approaches in which students construct their understanding through discussion and problem-solving with groups [5]. Incorporating group work into classes can yield a variety of benefits for students. Multiple meta-analyses of group work, including one specific to STEM undergraduate courses, have concluded that group work tends to boost a number of desirable outcomes, including achievement, productivity, attitudes towards learning, and psychological health in students [6,7]. In STEM courses like biology, group work is particularly common and amenable in laboratory-based classes [8]. Lab courses that employ inquiry-based activities can simulate how scientists work and collaborate in real life in addition to training students in science process skills. This is best demonstrated and practiced when undergraduates work in groups in the lab.

Despite its numerous benefits, group work can be challenging to implement effectively and may also cause undesirable outcomes. Students dislike group work when they have group mates who show up to class late, are unprepared, or are not engaged in the group work [9]. Social loafing is a common group work challenge, in which some students take on the majority of the work while others contribute little to the group project [10]. Due to these and other challenges, many students have negative perceptions of group-based projects [11]. In an attempt to provide practical guidance to instructors, researchers have investigated how the way groups are formed and group compositions impact the effectiveness of group work and student outcomes.

Group formation and composition

There is a great variance in the results of research on the best way to form groups or the ideal group composition. Generally, groups may be formed by students self-assorting or by instructor assignment. Self-assortment tends to result in more homogenous groups, since people generally tend to be more comfortable around and attract to others like themselves [12,13, but see 14]. There are multiple axes of heterogeneity that the literature unpacks: cognitive heterogeneity and demographic heterogeneity. Below we explore both.

When student choice factors into group selection, students could experience higher satisfaction in completing tasks, mainly due to increased comfort among group members leading to increased cohesiveness of the group. Theobald and colleagues showed that when students had a friend in their group, they were more comfortable, and this increased students’ content mastery by 27.5% [15]. This lends support to the idea that positive interdependence is one of the factors that impacted students’ experiences with group activities. Chapman and colleagues found that for first-year business students, self-selected groups perceived themselves to have better communication, more enthusiasm for group work, and a positive outlook towards their course outcomes [16].

However, the self-selection of students into groups can lead to a phenomenon of “cronyism.” Chapman and colleagues describe this as a situation where groups are based on prior relationships and familiarity and can cause cultural divide among students between groups [16]. Self-selection of groups is also shown to affect student achievement and lead to poorer attitudes towards learning [17,18]. Donovan and colleagues also observed that when asked to self-select, certain students reported feelings of anxiety when they could not find a group and ultimately approached the instructor for group assignment [14]. Thus, allowing students to self-assort into groups could pose challenges to creating inclusive and equitable learning environments for marginalized students if their peers do not seek them out for inclusion in their groups.

An alternative that potentially relieves stress on students to find a group they fit in is for instructors choose to assign groups. However, instructors then must decide how to create these assignments. The ideal composition of groups remains elusive as literature contains mixed results (see Table 1 in [14]). Creating diverse groups is an appealing approach because heterogeneity in perspectives and ways of thinking can boost overall group performance in problem solving [19,20]. Donovan and colleagues examined student performance in heterogeneous groups in a large-enrollment nonmajors biology course and found that group work improved learning outcomes of students who scored low on a biology competence assessment and did not impair the performance students who scored intermediate and high [14]. Students of all ability levels had more positive views of group work in heterogenous groups compared to students in more homogenous groups [14].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Student demographics.

https://doi.org/10.1371/journal.pone.0323799.t001

In addition to cognitive heterogeneity, research on group composition has considered the role of demographic heterogeneity. A survey of over 1,000 students from 6 different institutions found that cultural diversity promoted behavioral and cognitive engagement [21]. Similarly, a large-scale study of over 11,000 undergraduates at 20 institutions found that courses with higher representation of first-generation and racially marginalized students had higher average grades for the entire class [22]. Samudra and colleagues suggest that student groups with diverse representation tend to interact with each other more and thus transfer information and knowledge among themselves, leading to opportunities of peer learning [9]. This effect has also been found on a smaller scale by comparing groups within a STEM class; Donovan and colleagues found that demographically diverse groups have had higher learning outcomes than self-selected, more homogenous groups [14]. Rienties and colleagues showed a similar outcome in their study with students in an event management course and observed that although students in groups with demographically diverse members initially faced cultural differences, they were able to overcome them using coping strategies, leading to a stronger team identity [23]. Working in heterogenous groups can yield career-lasting impacts for professional scientists; Campbell and colleagues showed that journal articles in ecological science published by groups with gender diversity (at least one man and woman on the team) were 34% more likely to be cited than those published by gender-uniform groups [24]. This trend was repeated in journal articles published in the medical sciences [25] and in innovation process teams in management organization [26]. All these studies emphasize not only the real-world application of teamwork, but also the value of heterogeneity among team members.

Despite the numerous documented benefits, working in both cognitively and demographically heterogenous groups also poses challenges. When groups are heterogeneous in members’ identities, there is potential for individuals belonging to underrepresented and marginalized groups to experience feelings of marginalization and be subjected to bias and discrimination by their group mates [27]. Studies find that students with minoritized identities or those facing stereotype threats are more likely to be reluctant and passive while participating in a group, especially if they are outnumbered by members with majority identities [28,29]. For example, Adams and colleagues found that women tended to decrease their participation in a group that contained more men [28]. Another study with students in an introductory physics lab course indicated that task division in heterogenous groups became gendered and the women experienced stereotype threat and isolation, leading to inequity in opportunities for learning [30,31].

Academic heterogeneity can also produce challenges, such as low-ability students relying on high-ability students to complete the task. For example, some studies show that students tend to improve reasoning ability in groups with homogeneity in ability as they needed to interact and find solutions to tasks [32]. These groups likely could not rely on “social loafing” that could have been possible in a group with ability heterogeneity.

There is relatively little information available on heterogeneous group dynamics and outcomes in contexts where members of typically underrepresented groups are more highly represented (e.g., minority-serving institutions). One possibility is that the risk of negative effects from marginalized students being isolated as the only individual with that identity is reduced when the population of students overall is more diverse.

The present study

In the present study, we test whether students in groups that are strategically assigned to be maximally heterogeneous experience better outcomes than students in groups that are randomly assigned. Many prior studies examining group composition compare student self-selected groups to randomly-formed groups or examine existing variation among groups without an intervention (e.g., [12–14,16]). However, the existence of tools such as the Comprehensive Assessment of Team Member Effectiveness (CATME) SMARTER Teamwork [32–34] tool enables instructors to strategically form groups, an alternative to both random assignment and self-assortment. CATME is a system of online tools designed to help instructors manage student teams effectively. One of the available tools, which we use for the current study, is a team-maker, which sorts students into teams based on their responses to a survey and criteria that instructors select. In this study, we seek to explore whether strategically creating heterogeneous groups might result in positive benefits for students over randomly-assigned groups. Our study furthers the conversation about optimal group composition in two key ways.

First, we consider both types of heterogeneity – cognitive and identity – in combination as “maximally heterogeneous.” Based on prior work indicating benefits of both cognitive heterogeneity and identity heterogeneity, we hypothesized that combining both types of heterogeneity would yield positive attitudes towards learning, superior group performance, and reduced amounts of conflict during group work.

Second, we examine the effects of heterogeneity in a minority-serving institution where a group that is typically underrepresented is well-represented. Specifically, this study was conducted at a Hispanic-Serving Institution, which had 29% enrollment of Hispanic students and more broadly speaking 39% enrollment of students with underrepresented racial/ethnic backgrounds during the year the study took place. We hypothesize that the risk of marginalized students having negative social experiences from isolation will be mitigated due to their relatively high representation and therefore students in heterogeneous groups will experience improved outcomes.

Study setting and participants

We conducted this study at a single large, public, Very High Research Activity (R1) Hispanic-serving institution in the United States. The introductory biology course sequence for majors integrates both lecture and lab in the same course and spans two semesters. The course in the fall covers cellular biology, genetics, and energetics and the course in the spring covers evolution, ecology, and physiology (only one of the courses is available each semester). There are no pre-requisites for either course in the sequence, and students could take the courses in either order, though one is only offered in each semester (more details below). Two of the authors are the instructors of record for this course sequence: ASN is the lab coordinator who supervises the graduate Teaching Assistants (TAs) who are the lab instructors but does not directly instruct students and LBL is the overall course instructor of record who teaches the lecture sections but is not directly involved in lab instruction.

During each semester, groups worked on all lab activities together. Mid-semester, groups planned and executed an independent investigation project, i.e., an experiment based on previous lab activities that was developed and designed by groups themselves with guidance from TAs. Students were not provided explicit guidance on group work, but before working on the lab report, they completed a group contract in which they outlined expectations of contributions from each group member and designated an internal deadline for revisions prior to submission. At the end of the semester, groups submitted a collaborative lab report that the team submitted as a single report and all group members receive the same grade. However, students completed a peer review in which they indicated whether all group mates contributed equally and fairly. If group members reported that one group member did not contribute sufficiently to the report, that individual was or could have been assigned a lower grade per the TA’s discretion. In both semesters, the lab report was worth 8% of the lab grade (in turn, the lab grade was worth 30% of the overall course grade).

Many students take both courses in sequence, but students may take just one if they only require one for a major in a different department (e.g., agriculture, kinesiology, bioengineering) or transfer credit to replace one of the courses. Thus, the population of students in the two courses typically contains a large amount of overlap but is not exactly the same population. In our sample, 528 participants took both courses, 268 participants enrolled in the fall only (not the spring), and 100 participants enrolled in the spring only (not the fall). Due to the large but incomplete overlap, we implemented the quasi-experimental procedures in both semesters but treated them separately and considered the spring a replication attempt. In the fall, there were 798 students enrolled in the course across 36 lab sections (~24 students in each section) taught by 18 graduate Teaching Assistants (TAs; each TA teaches two lab sections). In the spring, there were 628 students enrolled in the course across 28 lab sections taught by 14 TAs. The demographic diversity of the participants is described in Table 1. Note that the number of participants is lower than the total enrollment in the course due to some students choosing to not participate in the research.

Group assignments

Each TA taught two lab sections; in one of their sections, they created maximally heterogeneous groups and in the other section created randomly-assigned groups. All groups were composed of 4 members at the start of the semester and groups were retained throughout the semester. If a group member dropped the course, the group remained intact with 3 members. In this academic year, no group had more than one member drop the course.

Maximally heterogeneous groups were created using CATME. To create groups using CATME, students completed a short survey at the start of each semester, providing information about their identities (i.e., race, gender, generation in college), prior preparation (GPA), weekly schedule, and future goals (complete survey text available in S1 File). To create maximally heterogeneous groups, CATME settings were assigned to create groups dissimilar in race, gender, generation in college, and GPA. Schedules were assigned to be similar to facilitate groups finding common time to work together on lab group projects. Random groups were assigned either by TAs before the first lab session or during the first lab session (e.g., by randomly handing out papers with group numbers on them).

We examined the structures of groups in sections where groups were created by CATME and randomly assigned to check that the intervention changed the pattern of group structures in the sections (Table 2). For each demographic characteristic, we characterized groups as being homogenous (e.g., being all white students or all women), having a student with a marginalized identity isolated in a group (e.g., 3 white students and a Black student or 3 men and a woman), or being mixed (e.g., 2 white students and 2 Hispanic students or 2 men and 2 women). We observed clear differences in the group makeup across the different strategies for creating groups. In both semesters, the sections where groups were created to be maximally heterogeneous had fewer homogenous groups and more groups that were mixed and with isolated URM students or isolated women. This provided confirmation that using CATME created groups that were noticeably different in composition from when groups were assigned at random.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Group structures.

https://doi.org/10.1371/journal.pone.0323799.t002

Data collection

Data analysis

Our analysis strategy treats our two semesters as a replication test of the same experiment. We decided to proceed with analyzing data from the two semesters separately for a variety of reasons, including that some students were in both classes, the course content was different (so the samples were not easily combinable), and our hypothesis testing procedure included interactions and we thought our sample size would not support three-way interactions with semester. Therefore, we analyze data from the two semesters separately and qualitatively compare the robustness of findings across both semesters.

Preliminary exploratory analyses

First, we conducted basic preliminary data exploration. We visually inspected the data by graphing the lab report grades, overall lab grades, and frequency of students reporting conflict disaggregated by group formation method for both courses (Fig 1). A simple visual inspection suggested that there were not any obvious differences across the two modes of group formation in students’ lab report grades or overall lab grades (Fig 1A). Conflicts were very infrequently reported, ranging from 4–9% of students being in groups that reported conflict (Fig 1B).

Hypothesis testing model selection results

We tested a series of models to identify which variables best explained the variance in each outcome in each semester. The final, best fitting models for both outcomes (lab report grade and final lab grade) are presented in Table 3. For all of the models, the best random effect structure included the group and TA variables, but not the section variable. For all but one model, the group formation fixed effect was absent from the final, best-fitting model. This means that whether groups were formed randomly or strategically to be maximally heterogeneous had no impact on the outcomes.

Parameter estimates are shown with standard error in parentheses. For indicator variables, the reference group is indicated in parentheses.

Model results for lab report grade

For lab report grade in biology 1 (fall semester course), the null model was the most parsimonious, best fitting model, which indicates that we were unable to significantly detect meaningful effects of the hypothesized fixed effects (gender, race/ethnicity, generation in college, nor group formation; Table 3). However, in biology 2 (the spring semester course), the model with an interaction between group formation method (heterogeneous vs. random) and URM fit the best. The retention of the interaction term in the best-fitting model revealed that the lab report grades for URM students (m = 33.22) were lower than non-URM students (m = 33.82) in the randomly-assigned groups only. In the heterogenous groups, URM students’ lab report grades were improved (m = 33.75) whereas non-URM students’ lab report grades were unchanged (m = 33.74), which removed performance disparity. It is important to note the caveat that this finding was only in biology 2 and that we could not detect any effect of race/ethnicity in biology 1, thus this result was not robust and could be spurious.

Model results for overall lab grade

In both courses, the group formation variable (random vs. heterogeneous) was not included in the final, best-fitting model, meaning that lab group formation method did not impact students’ overall lab grades in either course.

In both courses, the final model included all 3 demographics as fixed effects (but not group formation or any interaction effects), indicating that gender, racial/ethnic identity, and generation in college all impacted students’ overall lab grades. These results repeated trends known from prior literature that students with minoritized social identities experience worse academic outcomes. Final lab grades were lower for students identifying as women than men in both biology 1 (m_women = 83.05; m_men = 85.22) and in biology 2 (m_women = 80.87; m_men = 82.96). Similarly, URM students scored 3.17 points less than non-URM counterparts in biology 1 and 2.09 points less in biology 2. Finally, students who were the first in their family to attend college (first generation) scored 1.78 points less than their counterparts whose family members have attended college (continuing generation) in biology 1 and 3.66 points less in biology 2.