More than a Score: Metacognitive and Social-Affective Benefits of Cooperative Learning in STEM Classrooms

1.1 The state of STEM instruction: moving away from lectures alone

Quality instruction in science, technology, engineering, and mathematics (STEM) will continue to be a chief educational priority for the remainder of the twenty-first century [1]. Yet, students in STEM learning contexts often face unique barriers, like elevated levels of domain-specific anxieties (e.g., [2]) and reduced sense of belonging (e.g., [3]), which have been shown to predict lower performance, participation, and engagement in STEM fields [4, 5, 6], particularly for historically marginalized students [5, 7, 8]. Much work has aimed to support students’ STEM success by addressing these student-level psychosocial and motivational factors (e.g., values affirmations interventions, growth mindset training), with promising positive effects for test performance, grades, and degree completion (see [9] for a review).

Though well-intended, such interventions may be analogous to using a Band-Aid for a bullet hole. Recent research suggests that these light-touch, student-focused interventions are impactful and enduring only when coupled with concerted classroom-level changes [10]–namely, designing more inclusive STEM learning environments [11]. In this way, a hyperfocus on student-level factors may elide discussion of deeper, system-level challenges in STEM education (see [12]). Much research has documented common STEM teaching practices that have excluded or disengaged students, such as large student-to-faculty ratios, “weed out” cultures, and use of only high-stakes evaluative assessments [13].

Chief among these typical STEM teaching practices is the reliance on lecture alone. In his seminal book, Pedagogy of the Oppressed, Paulo Freire called for a move away from the “banking model of education”, where students’ roles were simply to passively receive and store deposits from the teacher [14]. Half a century later, unidirectional lecturing remains the modus operandi in university-level STEM instruction [15], despite the mounting evidence that interspersing student-student discussions into STEM lecture promotes deeper concept learning (e.g., [16]), attendance and engagement during class (e.g., [17]), and retention in the major [18]. These effects are particularly pronounced for students from racial/ethnic or socioeconomic groups historically underrepresented in STEM [11].

Thus, some have argued that instead of focusing on fixing students and their attitudes and perceptions of the STEM classroom, researchers should examine how to fix the classroom itself [13]. One increasingly well-studied classroom “fix” is the implementation of active learning techniques into lecture-based courses. Broadly, these contain a diverse range of pedagogical practices that encourage students to become more actively engaged during class time, either through peer discussion (e.g., think-pair-share) and/or response systems (e.g., clickers, minute papers). Large-scale meta-analytic reviews have shown considerable improvements in test performance, expertise-level concept discourse, and reduced failure rates in STEM classrooms that used active learning [11, 19]. Indeed, the effects are so large and consistent that some have argued the question is no longer whether active learning is better than lecture, but rather which types of active learning are most effective, and under which conditions [19, 20].

1.2 Cooperative learning

To this end, this chapter focuses on cooperative learning (CL), a form of interdependent active learning where students cooperatively work towards a shared learning goal. Specifically, CL is “an instructional use of small groups so that students work together to maximize their own and each other’s learning” ([21], p. 86). In this way, CL activities address the passive, unidirectional, and competitive nature of traditional STEM learning contexts and are distinct from other forms of active learning (e.g., collaborative learning, problem-based inquiry) that do not support students’ team building nor provide classroom level scaffolding of discussions [22]. CL can take many forms, including small group interviews, Jigsaw activities, and structured think-pair-share activities [22]. Meta-analytic reviews have suggested that the type of CL activity employed does not moderate students’ gains in understanding [23]. Rather, all CL activities share these five required elements that are key for achievement: positive interdependence (group succeeds only if each individual succeeds), promotive interactions, accountability at the individual and group level, fostering of teamwork, and group processing and reflection [24].

The remainder of the chapter is as follows: First, I review the benefits of CL for students’ performance and learning in STEM and consider the conditions under which CL is most effective–namely, how should groups be configured to maximize achievement? Towards that question, I then summarize evidence for important yet relatively underconsidered changes in students’ metacognitive processes and social-affective factors following CL. Lastly, I report on a recent experiment conducted by my team that empirically assesses CL group composition and its effect on performance, learning, and social-affective factors in introductory statistics classrooms. I conclude the chapter with future directions and implementation suggestions.

2.1 Evidence for performance and learning gains

In contrast to standard instructional approaches where students are passive during class and are individually accountable for their learning, CL paradigms encourage and scaffold students’ mutual learning efforts through collaborative activities [25]. Over half a century of empirical research supports the use of CL in the classroom for promoting students of all ages’ learning and performance (e.g., see [21, 23, 26] for reviews). For instance, Johnson and colleagues [27] synthesized evidence from 168 studies examining CL in undergraduate education. They found that, relative to competitive (mean weighted effect size = 0.49) and individualistic (0.53) learning paradigms, CL had a larger effect on undergraduates’ success across multiple achievement measures (e.g., higher-order thinking, persistence) and performance contexts (e.g., verbal tasks, mathematics tasks).

CL interventions have shown larger effects in STEM than non-STEM classrooms [23] and have demonstrated success in promoting achievement and persistence especially in undergraduate introductory STEM learning contexts [28]. Armstrong and colleagues [29] implemented CL in certain introductory biology courses and found improvements in performance on major exams and in-class activities, and on both conceptual and recall items, compared to business-as-usual control classes (see also [30]). Similar results were found for students’ final grades following optional CL activities in an introductory engineering course [31]. Further, meta-analytic reviews have consistently shown medium to large effects of CL on university students’ achievement in general and organic chemistry [32] and introductory statistics courses [33]. Studies have also documented increases in achievement-promotive behaviors among students in CL-designed classes, such as increased course attendance [29] and increases in self-reported time devoted towards studying [34], that are important for success.

At scale, CL interventions can fundamentally shape the STEM landscape of an institution. One university implemented a large-scale, interdisciplinary CL intervention in each of five of their introductory STEM courses [35]. Focusing on inquiry-based learning activities within small CL groups, the authors reported large gains in students’ average course grades, in the proportion of students’ receiving successful (A or B) grades, and in the retention rates of STEM majors [35]. Importantly, the use of CL eliminated the effect of high school SAT scores on STEM major retention at college, suggesting CL interventions might mitigate preexisting achievement gaps [35].

2.2 Theoretical perspectives

Though the effects on achievement are clear, the mechanisms underlying CL’s success are multifaceted. Slavin [36] outlined four major theoretical frameworks for understanding achievement gains following CL. Motivational perspectives posit that when students are rewarded based on the group’s performance, an interpersonal reward structure emerges that holds each individual accountable–thus increasing effort. Social cohesion perspectives emphasize that, through team building, cooperation can promote inclusion, care, and support for peers, which provides intrinsic rewards even without extrinsic incentives. At the same time, cognitive developmental and cognitive elaboration views focus on the impacts of peer interactions on thinking and reasoning processes occurring during CL. Compared to standard instruction, CL affords more opportunities for explaining one’s thought processes aloud (cognitive elaboration). This can prime students to experience cognitive conflict, reveal their misconceptions and/or challenge their preconceptions, and receive feedback from peers who are nearer in their abilities (cognitive developmental).

Still, the cognitive perspectives of CL do not divorce the learner’s thinking and reasoning from their social world, instead drawing heavily from Vygotskyan sociocultural theory and Piagetian views on cognitive development as inherently social processes [25]. Indeed, Slavin [36] and others [21, 37, 38] have argued that these frameworks are likely complementary, such that social cohesion and motivational aspects afford opportunities for deeper thinking and reasoning during group activities, which in turn boost positive interdependent feelings and increased motivation towards each group member’s learning in a recursive fashion.

2.3 Does group composition matter?

If how the CL group interacts with each other influences their success, then who should be in the group? One of the key unanswered questions in CL is whether group composition matters [39]–specifically, if groups should be heterogeneous (some students are more capable than others) or homogeneous (everyone’s at the same level) with regards to students’ abilities or prior knowledge in the material. There are many reasons this question remains, though I consider three primary observations drawn from studies that have evaluated CL in real STEM classrooms:

First, in many CL studies, heterogeneity is the default [40], thus this question of group composition is never considered. In practice, some (e.g., [41]) have formed groups at random due to feasibility constraints in large classes. Others investigated whether student-selected groups perform better than teacher-selected groups (e.g., [42]), but again did not empirically compare within-group heterogeneity. Still, others (e.g., [43]) have designed CL interventions on the assumption that a heterogeneous mix of abilities would be best, much to the dismay of CL researchers [44]. Even when describing how formal CL groups should be formed, Johnson et al. [21] recommend instructors use heterogeneous grouping sans explanation. The implicit assumption behind the endorsement of heterogeneity is often that higher ability students can scaffold discussions and learning opportunities, though rarely is this tested (see [45] for a discussion). Second, where heterogeneous and homogeneous cooperative groups have been compared in the classroom, teachers (knowingly or not) often differentially adjust their instruction depending on the group’s relative abilities [46, 47], thus potentially obscuring any treatment effects.

Third, and paradoxically, some CL studies have not used valid measures of learning, which may also obscure any differences by group ability (though see [16] for evidence from active learning research]. Many STEM CL studies operationalize achievement via assessments of performance (e.g., exam scores or end-of-course grades; [30, 31]). These performance measures do not necessarily imply long-term changes in learning. To the contrary, learning often only occurs after confusion, realization of misconceptions, or failure–meaning that long-term gains in learning may manifest only when initial performance first suffers [48, 49]. Applied to the question of CL group composition, this suggests the possibility that homogeneous groups without a more advanced peer may demonstrate low performance initially but engage in critical discussions with opportunities for cognitive conflict that are necessary to catalyze future learning potential.

In synthesizing the body of evidence on group composition and CL achievement, some researchers have found no effect of group composition [50] and many have reported conflicting evidence [51, 52, 53], some of which may be due to poor methodologies [54]. Most of this work has been done in K-12 settings, with substantially fewer experimental examinations of university-level CL group heterogeneity. In fact, one recent synthesis found only nine studies (most in STEM classrooms) that compared effects of group composition on undergraduates’ achievement following CL: two studies reported findings in favor of homogeneous grouping, four¹ were in favor of heterogeneous grouping, and four were inconclusive [55].

Beyond focusing on performance or achievement outcomes alone, which vary widely across classes and may not reflect learning, more insight about the impacts of group composition may be gleaned from understanding group processes, discourse, and students’ perceptions following groupwork. Particularly, insights from CL studies conducted inside the laboratory and from related research areas (e.g., peer tutoring, learning-by-invention) can elucidate the role of diverse group members’ abilities–including when heterogeneous versus homogeneous group composition matters most–on group interactions and perceptions during CL. These literatures argue that who is in the group influences how and what learners think about their reasoning and learning processes (metacognitive changes) and how learners feel about themselves and their group (social-affective changes).

3.1 Metacognition and lower ability students

In accordance with the cognitive elaborative and cognitive developmental explanatory perspectives of CL explained above, group discourse may be particularly facilitative for metacognitive reasoning when groups are homogeneously formed, even if students are all relatively lower ability. Though having a more knowledgeable student in a group might help expose students to diverse solution strategies and help students achieve the right answer (i.e., performance gain), it could also hinder opportunities for students to experience cognitive disequilibrium (i.e., conditions for learning). In one study of middle-school mathematics learning where group discussions were recorded, heterogeneous CL groups showed the smallest changes in achievement when group members only provided an answer without explanation or reflection [64]. Importantly, receiving quality explanations predicted gains in achievement through deeper, more constructive problem-solving attempts [64], evidencing a metacognitive pathway. However, analyses of undergraduate STEM students’ natural discourse in non-CL group settings have shown greater diversity of solution strategies in heterogeneous groups [65], but not necessarily greater metacognitive reflection (e.g., self-evaluations; [66]). When support for metacognitive practices was explicitly scaffolded into the group tasks and structure, students demonstrated greater transfer of learning and sustained gains throughout a course [61, 67].

Further evidence from researchers studying peer tutoring suggests that students in heterogeneous groups may default to the more capable peer to provide answers and resolve confusion, again limiting critical opportunities for cognitive engagement of all learners [37, 38]. Often, experiencing “productive failures” [48] can promote students’ long-term, conceptual understanding despite no immediate performance gains, nor differences in more surface level, procedural understanding. In explaining these effects, Kapur [48] argues for students’ increased opportunities to notice and identify prior knowledge gaps and critically evaluate theirs and others’ explanations. In fact, contrary to the idea that low ability students would be stuck without higher ability group members, much work has shown that reaching an impasse is a pre-requisite to learning gains in group settings ([68, 69], though see [66]).

3.2 Metacognition and higher ability students

Comparably less work has considered metacognitive gains for higher ability students working cooperatively in groups. Still, the evidence suggests these students also have something to gain from CL. Akin to the formation of gifted programs, it has been argued that homogeneous grouping of high ability students could propel their achievements beyond that which an instructor alone could provide [70] (though see [71]). Further, some have argued that the presence of a low ability student might hinder the potential of otherwise homogeneously paired high ability students [72]–as in, a group is only as strong as its weakest link. However, why these gains occur for homogeneously grouped high ability undergraduates has received little attention in CL research.

Here, again, peer tutoring and peer learning literature can provide additional insight into mechanism. A common belief is that high ability students can learn by teaching others in heterogeneous tutoring contexts, which has received empirical support [73]. However, the question lies not in that high ability students give answers, nor in the frequency with which they do, but rather in how they give answers [74]. Research suggests high ability students can experience learning gains insofar as answering questions affords opportunities to experience cognitive conflict, identify their preconceptions and misconceptions, restructure their schemas, and generate self-explanations [37, 38, 74]. This metacognitively rich process of higher ability students’ “reflective knowledge-building” contrasts with “knowledge-telling”, or simply providing the answer without evaluation of nor elaboration on their own understanding [74]. Even when homogeneously paired, high achieving students demonstrate gains in performance as they are able to exchange higher quality explanations with each other [72].

In sum, whether heterogeneous or homogeneous CL grouping is better, and whether they are better for high or low ability students’ achievement, is in part influenced by the extent to which students can engage in deep metacognitive reflection during discussion. Importantly, as mentioned earlier, these key processes are important conditions for deep learning [48] and may not necessarily evidence satisfactory immediate performance [49]. It is feasible to assume that effects of group composition and discourse practices on achievement may vary if measured at different timepoints following the initial CL activity (see also [67]). Duration engaging with group members may also affect achievement via social-affective changes–the longer one works with a group, the more comfortable and assured they may feel to contribute to and learn from group discussions.

4.1 Belongingness

Sense of belonging in college refers to “students’ perceived social support […], a feeling or sensation of connectedness and the experience of mattering or feeling cared about, accepted, respected, valued by, and important” ([75], p. 4). Feeling like one belongs to a group, in turn, reciprocally uplifts others’ felt belonging within the group [75]. In this way, CL is uniquely positioned to encourage feelings of belongingness, though rarely have CL researchers used that term specifically. With its focus on positive interdependence, promotive interactions, and fostering teamwork with classroom-level support [24], CL activities are designed to bring students together in open discussion towards a shared goal and often without added stakes. CL’s small group format may be perceived as less distressing and safer to ask questions and express confusion, especially CL activities that designate the roles of questioners and answerers [76, 77]. Evidence from K-12 classrooms show CL interventions predicted greater quality social interactions among peers [78], higher frequency of positive relationships with peers [26], and even lower rates of bullying [79]. In all cases, effect sizes were medium or higher and changes in positive interdependence among otherwise distant students was the presumed explanation.

These belonging-related benefits may be particularly helpful in the undergraduate STEM context, where low sense of belonging has remained a consistent negative predictor of STEM students’ performance and persistence [3, 80, 81]. CL activities differ substantially from the competitive and individualistic structures of typical STEM courses, predicting greater perceived social support and greater interpersonal attraction among undergraduates [21]. Increased belongingness has been associated with active learning interventions generally [82, 83] and in undergraduate STEM CL contexts specifically [84, 85, 86]. Here, undergraduates have reported increased belongingness in algebra, physics, biology, chemistry, geology, and engineering courses incorporating CL activities for the duration of an academic term [84, 85, 86].

CL may predict greater sense of belonging, but little literature has directly examined the question of CL group composition on students’ sense of belonging or related perceptions. Some studies argue that heterogeneous CL groups may psychologically alienate gifted students from the larger class. This is either because they feel exploited having to adopt a teacher role towards their peers [87, 88] or they feel frustrated by their low ability peers who may slow their learning and lower their grade [89]. The body of evidence on low ability students’ and/or undergraduates’ sense of belonging in CL remains even more limited. What has received comparably more empirical attention is students’ self-beliefs following CL–namely, their self-concept and self-efficacy.

4.2 Self-concept and self-efficacy

When students participate in CL activities, they tend to feel more capable in the academic domain: CL studies have often reported changes in students’ self-beliefs, including increases in confidence, self-esteem, and positive attitudes towards the course content [21, 28, 83, 90]. Other changes following CL include decreases in undergraduates’ anxiety towards the subject, distress, and fear of negative evaluations [76, 91, 92], particularly when instructors designed CL groups to support students’ interactions [76]. Of all social-affective changes following CL, most evidence reports increases in students’ self-efficacy and self-concept. These positive self-beliefs are related but conceptually distinct [93]. Each is defined in turn, followed by a summary of the relevant empirical evidence in CL.

Self-efficacy refers to beliefs about one’s future capacities in a domain (“what I can do”; [93]). Feeling more self-efficacious predicts improved performance, greater task-directed effort, and deeper metacognitive reflection [94, 95]. Analyses of undergraduates’ self-efficacy following CL are varied. Some studies report considerable increases in students’ self-efficacy [82, 90]. Conversely, Griffin and colleagues have found either no effects or very small, inconsistent effects of CL on undergraduates’ self-efficacy [96], even when students described CL favorably [97]. These, however, were not conducted in STEM courses.

Whereas self-efficacy is a descriptive, prospective belief about one’s competence, self-concept refers to one’s retrospective beliefs about their capacities [93]. Self-concept also both describes and evaluates one’s capacities, and in this way is influenced by one’s frame of reference [93]. In other words, self-concept judgments change with whom an individual compares themselves, whereas self-efficacy does not [93]. In this way, self-concept may be particularly relevant to the question of group composition in CL.

An application of frame of reference is the big-fish-little-pond-effect, which posits that students evaluate their academic self-concept with reference to others in their class [98]. Though this big-fish-little-pond-effect is premised on classroom-level ability grouping, it suggests students of the same capacity may have different self-concepts depending on who is in their CL group. Meta-analytic reviews of peer tutoring research and CL show consistent gains in K-12 students’ self-concept [73, 78, 99, 100] and in undergraduates’ mathematics self-concept in introductory statistics courses [91, 92]. Little, if any, research has examined whether these changes differ based on students’ relative abilities or the heterogeneity of groups (though see [70] for K-12 ability grouping at the classroom level). As described below, my team recently considered this question in a study of CL in introductory statistics courses.

5.1 Unique challenges in introductory statistics instruction

Why introductory statistics? Most of the literature on undergraduate STEM CL has been conducted in chemistry classrooms [40]. Though in many ways similar to other STEM classrooms, there are unique challenges in introductory statistics instruction that make it a prime candidate for CL. First, introductory statistics may enroll a considerable number of STEM-averse students: Introductory statistics is a common requirement for social sciences and education programs, which some students have reported selecting precisely to avoid math and statistics courses [103, 104]. Relatedly, students’ statistics anxiety has been shown to be a strong negative predictor of their performance [105] and of their achievement-related behaviors (e.g., procrastination, [106]). These anxious feelings may be exacerbated by the considerable diversity of prior knowledge typical for introductory statistics courses [107].

Second, learning statistics is different than what students might expect. Inferential statistics are premised on hypothetical yet fundamental sampling distributions that are difficult for students to grasp [108]. Additionally, the field’s highly technical use of everyday language (e.g., “significant”) or double meanings for the same word (e.g., “independent”) may confuse students further [109]. Other challenges include an instructional focus on mathematical and procedural knowledge over conceptual understanding [107] and, relatedly, a dearth of valid assessments of students’ statistics reasoning and transfer of knowledge [110]. As such, even students who pass introductory statistics courses do not show proficient conceptual understanding [111, 112].

5.2 Design and results of our CL intervention

A brief review of the CL study design and results are reported here. For more details, please see the publication [101]. Participants in this study were graduate and undergraduate students enrolled in five introductory statistics courses at the same institution. Students were identified as having relatively high (n = 54) or low (n = 88) prior knowledge about statistics at the beginning of the term. Within each classroom, students with low prior knowledge were assigned at random to CL groups of 3–4 students that were either heterogeneous (at least one high prior knowledge student; n = 45) or homogeneous (all low prior knowledge students; n = 43). Students completed three in-class exercises with their CL groups throughout the term. These exercises assessed conceptual understanding of key statistics concepts and were designed to spark discussion and deep reflection on key topics for which students often express confusion, make mistakes, or hold misconceptions. Each exercise had three phases: a pre-test that was completed by students independently, a post-test in which students answered the same items collaboratively with their CL groups, and an individual assessment of learning that required students to independently answer new items for assessing their transfer of learning from the group discussions (see [16] for a similar three-phase design).

In comparing the homogeneous groups relative to the heterogenous groups, we operationalized students’ gains in collective performance as the change from pre-test to post-test. We operationalized individual learning as the change from pre-test to individual test. We rescaled items to allow comparison and used multilevel linear modeling to account for the nested nature of our data (using low prior knowledge students in heterogeneous groups as the reference group). Our analyses revealed that low prior knowledge students showed similar performance gains in heterogeneous and homogeneous groups. Low prior knowledge students had greater performance gains than high prior knowledge students in heterogeneous groups. Between-group differences in individual learning were not detectable in this study.

At the end of the term, we assessed students’ perceptions of their CL experience through a series of survey questions. The responses loaded onto two factors: the first was context-specific perceived concept of self and peers (e.g., “during group discussion, I find myself able to answer questions without the help of others”), and the second was perceived value of the CL activities (e.g., “I have found that work in small groups consumes too much time”). Regardless of prior knowledge and group composition, all students perceived the CL activities to be equally valuable. We did not find major differences between homogeneous and heterogeneous groups in low prior knowledge students’ beliefs about peers/self-concept; yet we found that higher ability students generally held more favorable beliefs.

A qualitative analysis of responses to an open-ended question probing students’ perceived efficacy of CL further elucidated the role of quality peer interactions within various groups. CL was deemed effective when students were able to collaboratively clarify key concepts, compare their explanations to reveal misconceptions, and teach each other. Students also reported greater sense of comfort discussing in the small group space versus whole-class discussion. However, some students reported that CL was less effective when the items or the discussion made students feel more confused, or when their group members did not know enough to participate. Importantly, these themes emerged at similar frequencies across group compositions and across student prior knowledge levels.

The findings from this study suggest that, for low prior knowledge students, interactions among other low prior knowledge students were as valuable as interactions including a higher prior knowledge peer. The results shed light on the question of group composition: Heterogeneously and homogeneously grouped students’ qualitative responses explicitly remarked on the metacognitive affordances of the CL activities at equal frequency (self-explanation, misconception revelation). Still, an explicit assessment of students’ discourse is an important next step to better understand learning potential [63]. Moreover, this study measured self-concept at one timepoint–though large differences within heterogeneous groups emerged, we cannot be sure these were a result of the CL activities without a baseline assessment. The items comprising this self-concept measure more so captured students’ beliefs about their and their teammates’ capacities within the CL settings and may not represent academic domain self-concept as typically measured in CL interventions (e.g., [91]). Next, more future directions for the field are provided.

7.1 Recommendations for classroom implementation

Though future research is needed to further elucidate the conditions of success for CL, what is clear is that including any active, student-student discussion is better than lecture alone [11, 19, 20]. Below, I provide practical recommendations for STEM instructors to facilitate student learning and engagement by embedding student-student discussion into lecture. These recommendations are informed by the CL literature reviewed above, our team’s findings [101], and practices I use in my own classrooms. The first set of recommendations may suit instructors in classes of any size–including large, lecture hall-style STEM courses. The second set of recommendations characterize more formal CL designs and may better suit classes with smaller student-to-faculty ratios.

First, all STEM instructors should incorporate some form of student-student discussion into their lecture, even on top of active learning activities. For instance, one common STEM active learning tool is the use of quick polls to quickly gauge understanding during lecture. This can be made into a collaborative activity by reserving time for student discussion following the poll. It is best to have students discuss prior to revealing the correct answer. In this way, students must verbalize their explanations and rationales for selecting certain answers (something that is not guaranteed during individual polling alone) which provides opportunities to reveal misconceptions in their logic and gaps in their understanding, thus priming them for subsequent learning. Instructors can re-share the poll following discussion to see if consensus was achieved (performance change) and, critically, should provide a new question to assess whether students can transfer their understanding gained from discussion to a new problem context (learning; [16, 101]). These kinds of activities can be done in small, ad hoc groups (i.e., cooperative base groups; see [21]), meaning instructors need not worry about assigning individuals to groups.

Instructors teaching smaller classes, or those with more teaching support staff, may consider implementing formal CL groups into instruction. To this end, instructors must take care to foster teamwork and mutual interdependence among CL group members [24]. It is best if students are assigned to groups as close to the start of the course as possible, in order to establish community and trust prior to any evaluative or summative CL activities. Additionally, instructors should build in opportunities for group collaboration and discussion throughout the course–even outside of the CL-relevant activities. In my classes, I have posed brief catalyzing questions before lecture (e.g., “What was confusing from today’s assigned readings?”) and ice breaker-type questions (e.g., “How is midterm week going?”) to reinforce the practice of sharing challenges or difficulties within one’s group and increase camaraderie and trust for subsequent CL-relevant activities. Lastly, having students and instructors reflect on and evaluate the efficacy of the group’s efforts can help identify any challenges early and facilitate promotive interactions. In my classes, I have done this through peer and self-evaluations at the midpoint and end of the term. This is particularly helpful as the instructor can anonymously share the favorable and critical feedback with the entire class–thus modeling how groups may work most efficiently and how individuals can contribute towards their group’s success.

Finally, I make no broad recommendation for heterogeneous versus homogeneous CL grouping–what matters more than who is in the group is the quality of social interactions and the potential for metacognitive reflection during CL group discussion. Still, instructors concerned about heterogeneity in students’ prior knowledge can consider CL activities that explicitly and evenly divide the labor of cooperative tasks. For instance, in Jigsaw activities [123], each student independently takes on one “piece” of a task, thus becoming an expert on that topic. Then, the group convenes, each student shares what they have learned, and the group fits all the pieces together to complete the task. Because students are responsible for teaching something new to others, gaps in relative ability or prior knowledge may not be as evident to the group, nor matter as much for their success. Additionally, instructors concerned about heterogeneity may consider changing their reward structures to promote non-hierarchical equitable exchanges: incentive structures that explicitly embed positive interdependence during groupwork (e.g., a student’s grade is determined by the average of all group members’ scores) produce greater cooperative efforts and, consequently, performance across all group composition types [25, 78].

7.2 Closing thoughts

Taken together, the evidence is clear: on average, students who participate in cooperative learning engage in deeper metacognitive reflection; have more positive feelings about themselves, their learning environment, and those in it; and achieve better performance and learning in undergraduate STEM classrooms. Still, with more institutional support for instructors, and with more nuanced empirical investigation into group setup and composition, cooperative learning is well suited to be the necessary “fix” [13] that STEM classrooms need to support students’ performance and learning, eliminate preexisting achievement gaps, and retain more students in the field. This is a promising area in which a joint, concerted effort between educational researchers and instructors can fundamentally alter traditional STEM instruction and student trajectories.