The individual performance of people, whether work-related or academic, is affected by other members of their peer groups. This effect is attributable to factors such as peer pressure and mutual learning among those people. In many cases, firms and other organizations take such “peer effects” into consideration, and try to improve the work environment with the aim of improving worker productivity by taking measures such as introducing the team production approach (e.g., Ichniowski et al., 1997; Hamilton et al., 2003). On the other hand, there is contention over the ideal approach to forming peer groups. Past studies have suggested that peer interactions lead to better overall work performance in groups with a greater spread in abilities and skills (“heterogeneous group”) rather than the one with similar abilities and skills (“homogeneous group”). That is because a heterogeneous group is expected to generate a “selection effect:” i.e., high-ability members tend to form subgroups with other like-minded, high-ability members, leading to a further improvement of their performances. However, due to such selection, low-ability members suffer as they are unable to enjoy the benefits of peer learning from high-ability members. It is therefore difficult to judge the effects of ability heterogeneity normatively in comparison with homogeneity.
In this column, I will introduce readers to a paper I coauthored with Professor John Ashworth from Durham University in the United Kingdom (Kamei and Ashworth, 2023) which shows that forcibly pairing more able and less able individuals and providing them with opportunities for peer learning within pairs is useful for improving the entire pair’s productivity. This study is based on a randomized field experiment in a university classroom. The experiment results showed not only that low-achieving students improved academic performance through working with their matched high-achieving students, but also that the high-achieving students were not hurt by being matched with the low-achieving ones. In a previous column , based on the data obtained from a different economic experiment, I explained that moral hazard behavior, such as shirking, is prevalent among workers assigned with “undesired tasks” (externally imposed tasks that they do not enjoy) due to a negative psychological reaction (Kamei and Markussen, forthcoming). What is interesting in the present experiment is that no such negative effect was observed in the peer-learning setting even when undesired persons were assigned as a partner.
Background and Experimental Design
The Kamei-Ashworth study is based on a field experiment conducted in the “Introduction to Economics” module at Durham University taught by this author in the 2019-2020 academic year (Kamei was Associate Professor at Durham University at that time). This module is compulsory for the first-year undergraduate students in accounting and finance in the university. The subject of the module is comprised of two parts—introductory microeconomics (the first term) and introductory macroeconomics (the second term). Each student receives a grade based on the result of a single year-end written examination called the “summative assessment” (a perfect score = 100 points), which is conducted at the end of the second term.
In the Introduction to Economics module, all students were required to attempt and submit a “formative assessment” in each term (two pieces in total). This was equivalent to a mid-term exam. Specifically, students must answer a problem set (whose format is the same as the summative assessment) and submit their answers officially to the university. The tutors (appointed by the university) marked the work online, and provided individual feedback to each student. Because of the characterization of the formative assessment as practice for the summative assessment, the marks given for the submitted work were not reflected in the final assessment grades for students (although the scores were recorded in the students’ information sheets in the university).
For the purpose of implementing a randomized control trial, a “peer review assessment” was introduced in the module as another written assignment in each term. Under the peer review assessment, students were required to attempt a problem set whose format was the same as the formative assessment and submitted their work. However, unlike in the case of the formative assessment, the submitted pieces were assessed not by the teacher but by other students in the module. In the peer review assessment, students were divided into pairs, with each member of the pair critically assessing the other’s work and providing feedback and a mark using a prearranged format sheet (called “proforma”). They had to hold a meeting in their pair to discuss the details of the problem set and to explain the feedback and their mark for each other. The discussion continued until a consensus was reached on their feedback and marks given (it was possible to revise the score as necessary). Under the rules set for this assignment, the student had to go through the whole process without seeing an answer sheet of the problems set: a solution was distributed only after the peer review assessment activities ended. Each pair was encouraged to study and find answers by themselves if both the students in the pair did not solve the problem set. In other words, cooperation on finding the correct answers within the pairs was essential.
An intervention was made for how pairs were formed in the peer review activities. In the control condition, students were arranged in descending order of the marks recorded in the formative assessment (interim class performances) in the first term, and pairs were formed by matching students whose scores were similar to each other (“sorting” approach). In the treatment condition, pairs were formed in a completely random fashion, i.e., through computer random number generation (“random matching” approach). Students were not allowed to choose their own partner, nor was the exchange of assigned partners between pairs permitted. The average intra-pair absolute difference in interim academic performance in terms of the mark of the formative assessment was very small, 3.26 points, in the control condition (sorting), but was much larger, 25.16 points, in the treatment condition (random matching). In other words, as intended, pairs in the treatment condition had a greater variation in interim achievement levels between pair mates, relative to those in the control condition. We analyzed how the use of these two different pair-forming approaches affected the students’ performances of the summative assessment.
To maximize the external and internal validity of the project, the students remained uninformed of the presence of the intervention experiment and the use of the two different pair-forming approaches. They were merely briefed on the effectiveness and educational objectives of peer learning while being left to assume that the peer review assessment activities were nothing more than part of the module. This experiment was conducted upon receiving approval from the institutional review board (IRB) at Durham University.
Peer Learning Effect Is Stronger in Pairs with a Greater Spread in Interim Performances
As already explained, the assessment grade for students in the Introduction to Economics module was determined solely on the basis of their own marks in the year-end summative assessment. Panel A in Figure 1 shows the average mark in the summative assessment categorized by their respective matching condition (“random matching” or “sorting”). It reveals that students in the treatment condition (random matching) achieved significantly better marks than those in the control condition (sorting). In the class, a small number of students failed to submit their work under the peer review assessment, although attempting the problem set and submitting their answers was the premise for peer learning activity. Even if students did not submit the assessments, they were still encouraged to meet with their partners and discuss the problem set within the pairs as an academic commitment. However, because having an effective within-pair discussion might be difficult without having their partners’ work, it is useful to study a possible treatment effect also by limiting data to those pairs both of whose members submitted the assessments (91.2% of the total number of pairs). Panel B shows a similar pattern for the restricted dataset to Panal A, and the performance difference is significant at two-sided p = 0.015. Our conclusion is therefore that pairing students with different prior achievement levels for the purpose of peer learning activities is more effective in improving academic performance than pairing students whose achievement levels were similar.
Low-Achieving Members in Pairs are the Ones that Mainly Benefited from Peer Learning
Why did pairs of students with a large interim performance difference obtain significant benefits from peer learning? To answer this question, we split members of the pairs formed under random matching into two sets—one comprising members whose interim performance level gauged by the formative assessment was higher than their pair partner (“high achiever group”) and the other comprising members whose interim performance level was lower (“low achiever group”), and then analyzed how much each set of students improved their academic performance as a result of peer learning activities. The analysis indicated that the students in the low achiever group in the treatment condition (random matching) improved performance by more than five points on average than those in the control (sorting approach) – see Row I.ii of Table 3.A in Kamei and Ashworth (2023). On the other hand, despite the fact that the students in the high achiever group were forcibly paired with the students in the low achiever group for peer learning, the performance of the high achievers was not undermined as a result.
The strong positive effect of peer learning in pairs of students with a large interim performance difference can be explained by two types of effects. One type is social effects through social comparison, such as guilt, shame and pride, among the members of the pairs. Each student not only had their work viewed, critically assessed and marked by their partner, but they also assessed and marked the partner's work. When students recognize their own performance level to be insufficient, they incur negative social effects; and when they recognize their own performance to be exemplary, they can enjoy positive effects (e.g., Bowles and Gintis, 2015). The psychological effects of performance information could become motivation for students to enhance self-study efforts. The second type of effect is the so-called mutual learning effect. This effect is generated when high-ability (-achieving) members have strong non-selfish preferences, such as social preferences. In this case, high-ability members are motivated to educate their matched, low-ability partners to improve the partners’ performance, even at the expense of significant effort. This effect is higher when the ability difference within the pair is larger.
Recent research in the fields of labor economics and economic education has discussed the merits of dividing, respectively, workers and students, into peer groups by ability (achievement) level. Those merits are attributable to the fact that persons who are in the position of teachers can better tailor their instruction levels and methods that are appropriate to the capabilities and skills possessed by members in each peer group (Duflo et al., 2011). On the other hand, the results of our experiment suggest that forming peer groups with a greater spread in abilities without dividing workers or students by their ability level may be more beneficial if peer learning is an important element in achieving the goals within corporate organizations or in school. This indicates that when designing the peer group structure, it is necessary to consider whether the approach of teaching whole groups through top-down education or the approach through peer learning brings greater benefits, and also that it may be useful to create multiple layers of peer groups in order to obtain benefits from both approaches.
In order to maintain the productivity level in Japan, it is essential to raise labor productivity. It has recently become usual for employers, whether they be firms or government organizations, to take measures to promote labor mobility, such as hiring mid-career workers, while education, training, and skills development activities other than on-the-job training (OJT) have started to gain popularity. In this situation, worker heterogeneity in terms of abilities, skills and other personal characteristics in each workplace has increased far beyond previous levels. Presumably, it will become more and more important for employers to meticulously design approaches to peer learning and human resource management within their organizations in order to benefit from the spillover effects brought to their workforces by trends such as labor mobility and reskilling.