The effect of team formation on team performance
A comparison between self-formed teams, partly self-formed teams and randomly formed teams in the short term
Although a lot of previous studies investigated the optimal composition of teams, only a few of them investigated the effect of team formation on team performance. This paper investigates the effect of team formation on team performance in the short term by comparing 65 teams of students. Those participating teams are self-formed, partly self-formed or randomly formed. It is found in previous studies that self-formed teams are more homogeneous than randomly formed teams and that homogeneous teams outperform heterogeneous teams in the short term. This subsequently leads to the expectation that self-formed teams will outperform randomly formed teams in the short term. Although the results of this study provide no significant results that support the predictions that self-formed teams are more homogeneous and that homogeneous teams outperform heterogeneous teams, the results do support the expectation that self-formed teams outperform randomly formed teams. Those remarkable findings indicate the need for future research to further investigate the effect of team formation on team performance.
Name: S.C.M. Korrel
Student number: 10636242
Thesis supervisor: mw. dr. S. Dominguez Martinez BSc- programme: Economics & Business
Specialization: Finance & Organization Submission date: 24-01-2016
Hierbij verklaar ik, Simone Korrel, dat ik deze scriptie zelf geschreven heb en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud ervan.
Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die welke in de tekst en in de referenties worden genoemd.
De Faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de begeleiding tot het inleveren van de scriptie, niet voor de inhoud.
1
1. INTRODUCTION ... 2
2. THEORETICAL FRAMEWORK ... 4
2.1. HOMOPHILY ... 4
2.2. DIVERSITY ... 4
2.3. PERFORMANCE OVER TIME ... 7
2.4. PERFORMANCE OF SELF-FORMED TEAMS AND RANDOMLY FORMED TEAMS ... 9
3. METHOD ... 11 3.1. RESEARCH SETTING ... 11 3.2. DATA COLLECTION ... 122 3.3. RESEARCH SAMPLE ... 12 3.4. DATA VARIABLES ... 13 4. RESULTS ... 16 4.1. HYPOTHESIS 1 ... 16 4.2. HYPOTHESIS 2 AND 3 ... 17 4.3. ADDITIONAL ANALYSES ... 20 5. DISCUSSION ... 24 5.1. FINDINGS ... 24
5.2. CONTRIBUTIONS AND IMPLICATIONS FOR THE LITERATURE ... 25
5.3. PRACTICAL IMPLICATIONS ... 25
5.4 LIMITATIONS ... 25
5.5. DIRECTIONS FOR FUTURE RESEARCH ... 26
6. CONCLUSION ... 27 REFERENCES ... 28 APPENDIX A ... 32 APPENDIX B ... 33 APPENDIX C ... 34 APPENDIX D ... 34 APPENDIX E ... 35
2
1. INTRODUCTION
Organizations increasingly recognize the value of working in teams to accomplish goals (Chapman, Meuter, Toy & Wright, 2006). Because of this, the use of teams in the organizational context has become very common (Chapman et al., 2006). Besides inside the organisational context, also outside the organizational context people work in teams in a lot of situations. For example when people play sports like football or hockey or at school where students have to work in teams and have to deliver team assignments. In response to the increasing focus of organizations on teamwork, business schools increased the frequency in which students are working together teams (Chapman & Van Auken, 2001).
A lot of previous studies investigated the ‘best’ composition of teams (e.g. Ancona & Caldwell, 1992; Bacon, Stewart & Anderson, 2001; Hansen, Owan & Pan, 2006; Williams & O’Reilly, 1998). This is very relevant since working efficiently and effectively in a team has become essential in the workplace (Chapman et al., 2006; Hoogendoorn, Oosterbeek & Van Praag, 2013). Most of the studies which investigated team composition focused on a specific kind of diversity, for example gender (Hoogendoorn et al., 2013; Pelled, Eisenhardt & Xin, 1999), age (Tsui, Egan, & O’Reilly, 1992), race (Pelled et al., 1999; Tsui et al., 1992) or personality (Harrison, Price & Bell, 1998; Hoffman & Maier, 1961) and investigated how much of this kind of diversity teams should have to perform optimal (e.g. Ancona & Caldwell, 1992; Bowers, Pharmer & Salas, 2000; Cox, Lobel & McLeod, 1991; Hoogendoorn et al., 2013; Rulke & Galaskiewicz, 2000; Williams & O’Reilly, 1998). Up till now, not a lot of studies investigated the most optimal type of team formation. This is remarkable since team formation has a substantial influence on team performance (Bacon et al., 2001; Chapman et al., 2006; Mahenthiran & Rouse, 2000). Team formation implies the method of assigning people to teams. A team could for example be formed by team members on their own by giving them full self-control or a team could be formed by a third party, for example a leader.
The study by Mahenthiran and Rouse (2000) showed that partly self-formed teams had a higher group performance and satisfaction in comparison to completely randomly formed teams. Moreover, Chapman et al. (2006) showed that totally self-formed teams perceived more positive group dynamics, better outcomes and had a more positive attitude towards the group experience in comparison to completely randomly formed teams. In addition to these studies, this paper will investigate the difference in the influence of self-formed, partly self-formed and randomly formed teams on team performance. The research question that will be addressed in this study is: ‘Do self-formed teams perform better than randomly self-formed teams in the short run?’ Whereas Mahenthiran and Rouse (2000) and Chapman et al. (2006) only considered two different types of team formation, this paper will consider three types. Furthermore, whereas the study of Chapman et al. (2006)
3 focused on performance perceived by students, this study will focus on performance assessed by a third party and while Mahenthiran and Rouse (2000) considered different grades for each individual team member, this study will focus on one group grade for all team members.
The findings of this study will be useful for all people who are responsible for or involved by working in teams, like leaders, managers, teachers or students. By showing that people work better or worse when they compose teams by themselves, leaders can choose to let them self-select or not. This subsequently influences the performances of the team.
This paper starts with a literature review which will provide theoretical background for the most relevant variables and for expected relations in this study. Thereafter, in the method section it will be explained how the research question will be investigated by explaining the research setting, the research sample, the data collection method and the data variables used in this empirical study. This method section will be followed by the results section. In the results section the empirical study will be performed and the results will be analysed. The results will be critically overthought and discussed in the discussion. Finally, this paper will finish with a conclusion.
4
2. THEORETICAL FRAMEWORK
This chapter will provide theoretical background of the key variables used in this paper and will provide theoretical arguments for the expected relations. The key variables used in this paper are homophily, diversity and team performance over time.
2.1. HOMOPHILY
There are a lot of studies which investigated the effect of homophily (e.g. McPherson, Smith-Lovin & Cook, 2001; Reagans, 2005). The homophily principle can be defined by the sentence ‘similarity breeds connection’ (McPherson et al., 2001, p.415). This principle states that there is more contact between similar people than between dissimilar people (Huston & Levinger, 1978; McPherson et al., 2001) since people mostly associate themselves with people who are most similar to them (Feld, 1982; McPherson et al., 2001; Reagans, 2005). Homophily can be based on demographic characteristics, like age, sex and ethnicity (e.g.Loomis, 1946; Richardson, 1940), as well as on psychological characteristics, like intelligence and attitudes (e.g. Loomis, 1946; Richardson, 1940). Aristotle (1934, p. 1371) stated: ‘people love those who are like themselves’ and Plato (1968, p. 837) stated: ‘similarity begets friendship’. Friends are chosen on the basis of similarities, as McPherson et al. (2001, p. 417) stated: ‘birds of a feather flock together’. An example is found in the study of Bacon, Stewart & Stewart-Belle (1998) which shows that members of self-formed teams are more likely to have the same level of ability. Several empirical studies show that children at school are more likely to play and become friends with children who have similarities with them (e.g. Bott, 1928; Hubbard, 1929; Wellman, 1926). This homophily principle is also applicable on students who have to work together. If students are able to self-select team members, they are most likely to choose friends (Bacon, Stewart & Silver, 1999; Chapman et al., 2006). On the basis of the previously mentioned literature, this paper predicts that self-formed teams will have more similarities among team members than randomly formed teams. This leads to the first hypothesis:
Hypothesis 1: Self-formed teams are more homogenous than randomly formed teams.
2.2. DIVERSITY
As it seems that people tend to choose people most similar to them, it is likely that self-formed teams are less diverse than randomly formed teams. Therefore, in this paper the impact of (non-) diversity in teams will be considered. A lot of previous studies investigated the effect from levels and types of diversity within teams on team performance (e.g. Ancona & Caldwell, 1992; Bowers et al., 2000; Chatman, Polzer, Barsade & Neale, 1998; Cox et al., 1991; Hansen et al., 2006; Harrison et al.,
5 1998; Jehn, Northcraft & Neale, 1999; Rulke & Galaskiewicz, 2000). In this paper, the definition of diversity formed by Van Knippenberg, De Dreu and Homan (2004, p. 1008) is used: ‘differences between individuals on any attribute that may lead to the perception that another person is different from self’. This definition is very broad and complete; it includes surface-level characteristics, like age and gender, as well as deep-level characteristics, like political and sexual preference (Jehn et al., 1999; Van Knippenberg et al., 2004).
Up till now, the results of studies which investigated the effect of diversity on team performance are really inconsistent (Tsui et al., 1992; Williams & O’Reilly, 1998). Williams and O’Reilly (1998) looked at more than 80 previous studies about diversity and concluded that all types of diversity can have positive as well as negative effects on team performance. A few examples will be considered. First, functional diversity may lead to more innovation and better performance (Ancona & Caldwell, 1992; Bantel & Jackson, 1989; Jehn et al., 1999; Pelled et al., 1999; Williams & O’Reilly, 1998) through increased external communication (Ancona & Caldwell, 1992), different perspectives (Bantel & Jackson, 1989; Williams & O’Reilly, 1998) and increased task conflict (Jehn et al., 1999; Pelled et al., 1999; Tjosvold, 1998). However, functional diversity may also lead to less cohesion in teams (Ancona & Caldwell, 1992) and more relational conflicts (Jehn et al., 1999; Pelled et al., 1999). Second, demographic diversity may lead to more available resources (Bantel & Jackson, 1989; Wanous & Youtz, 1986), which can result in higher performance. For example, Wanous and Youtz (1986) show that the high availability of resources, a lot of different perspectives, leads to more generated alternative ideas and a final solution of high quality. On the other side, demographic diversity may lead to less communication (Triandis, 1960), less team integration and commitment (Harrison et al., 1998; Tsui et al., 1992) and more relational conflicts (Ancona & Caldwell, 1992; Jehn et al., 1999), which has a negative effect on team outcomes (Jehn, 1995). Furthermore, it is found that demographic diversity has a negative relation with the focus on team objectives (Chatman et al., 1998) and cooperative norms (Chatman & Flynn, 2001). Less cooperative group norms implies that people place a higher degree of importance to personal interests relative to shared pursuits (Wagner, 1995). Decreased focus on team objectives and less cooperative norms may lead to decreased performance and decreased effectiveness (Chatman & Flynn, 2001; Wagner, 1995).
One of the most popular demographic characteristics in diversity literature is gender (Bowers et al., 2000; Clement & Schiereck, 1973; Hoffman & Maier, 1961; Hoogendoorn et al., 2013; Tsui et al., 1992). For example Clement and Schiereck (1973) found that teams consisting of both male and female underperformed teams consisting solely of males or consisting solely of females. Clement and Schiereck (1973) give sex-clustering, which implies the forming of sex coalitions in mixed groups, as possible explanation. Sex-clustering may lead to less communication between the two like-sex
6 coalitions (Clement and Schiereck, 1973). Contrary, there are also studies which found that the influence of sex diversity on performance is positive (e.g. Hoffman & Maier, 1961). Hoffman and Maier (1961) found that teams which are heterogeneous with respect to gender are superior relative to homogeneous groups in problem solving. This superiority in problem solving leads to solutions of higher quality (Hoffman and Maier, 1961). Hoogendoorn et al. (2013) found that teams with an equal gender mix outperform teams which are male-dominated. However, they did not succeed in finding an explanation for this result. Unfortunately, they could not draw a conclusion about teams with equal gender mix relative to female-dominated teams since the sample of the female-dominated teams was too small.
In response to all those contradictory findings of positive as well as negative consequences of diversity, Van Knippenberg et al. (2004) and Williams and O’Reilly (1998) assumed that the processes underlying diversity determine if the effect of diversity on performance will be positive or negative. The main underlying processes include social categorization and similarity/attraction and information/decision making (Van Knippenberg et al., 2004; Williams & O’Reilly, 1998).
The social categorization and similarity/attraction theories are mostly based on surface-level characteristics (Chatman & Flynn, 2001; Jehn et al., 1999; Tsui et al., 1992) and predict that homogenous teams will outperform heterogeneous teams (Jehn et al., 1999; Watson, Kumar & Michaelsen, 1993). The theories state that people categorize themselves on the basis of similarities and prefer people from their own group above people from other groups (Brewer, 1979; Tajfel, 1982). With teams consisting of more similar people, there will be more cooperation (Chatman & Flynn, 2001; Cox et al., 1991), cohesion (Ancona & Caldwell, 1992; Tsui et al., 1992), communication (Reagans, 2005; Triandis, 1960; Zenger & Lawrence, 1989), satisfaction (Tsui et al., 1992) and less relational conflict (Jehn et al., 1999; Pelled et al, 1999) within the teams. According to Williams and O’Reilly (1998) the negative effects of social categorization probably diminish over time since over time there will be more attention paid to deep-level characteristics instead of surface-level characteristics. Negative effects of social categorization can be further diminished by creating shared goals and a collectivistic culture (Williams & O’Reilly, 1998).
The information/decision making theory, on the other hand, is mostly based on deep-level characteristics (Tsui et al., 1992; Van Knippenberg et al., 2004) and predicts that heterogeneous teams will outperform homogenous teams (Bantel & Jackson, 1989; Cox et al., 1991; Jehn et al., 1999). With more diverse people in a team, there will be more different perspectives resulting in more alternative generated ideas. This may result in higher quality solutions (Ancona & Caldwell, 1992; Bantel & Jackson, 1989; Hoffman, Harburg & Maier, 1962; Wanous & Youtz, 1986). Hoffman et al. (1962) found that for achieving high quality solutions, tolerance of each other’s point of view
7 within a team is important. This implies that a heterogeneous team can be more beneficial than a homogeneous team due to more alternative perspectives if the condition of tolerance of other viewpoints is met (Hofmann et al., 1962). Furthermore, heterogeneity within teams may be even more beneficial relative to homogeneity when working with complex task problems rather than simple problems (Bowers et al., 2000).
2.3. PERFORMANCE OVER TIME
Team performance changes over time (Chatman & Flynn, 2001; Watson, Michaelsen & Sharp, 1991; Watson et al., 1993). Several studies give possible explanations for this change, like better use of knowledge of team members (Watson et al., 1991), more cooperation within teams (Chatman & Flynn, 2001) and a modification in the collectivistic culture of teams (Chatman et al., 1998). Several researchers found that under certain conditions heterogeneous teams will improve more than homogenous teams (Chatman & Flynn, 2001; Watson et al., 1993).
The improved performance over time of diverse teams relative to homogeneous teams becomes very clear in an empirical study by Watson et al. (1993). Watson et al. (1993) assigned students to teams, making 19 culturally heterogeneous teams and 17 culturally homogeneous teams. Teams were defined culturally diverse when the members were different in both ethnicity and nationality. The teams consisted of four or five members. Performance was measured according to four tasks; range of perspectives, problem identification, alternatives generated and quality of solutions. Watson et al. (1993) found that in the first weeks of the experiment, the homogenous teams outperformed the heterogeneous teams on all measured tasks. During the observed weeks, both teams improved performance. However, the heterogeneous teams improved more than the homogenous teams. At the end of the period, after 17 weeks, the heterogeneous and homogenous teams had the same overall performance. However, the heterogeneous teams outperformed the homogenous teams on two tasks; range of perspectives and alternatives generated. Watson et al. (1991) argued that the knowledge of members can be used better over time caused by increased experience.
Another relevant factor for team performance which could also be subject to modification over time is the level of cooperative norms within a team (Chatman & Flynn, 2001). According to Chatman and Flynn (2001), heterogeneity has a negative relationship with cooperative group norms. Diversity will lead to more individual norms and less group norms (Chatman & Flynn, 2001). This is not beneficial since cooperative norms have a positive effect on performance (Chatman et al., 1998). However, this negative influence of heterogeneity on cooperative norms can diminish over time. Chatman and Flynn (2001) did an empirical study with MBA students who had to work on a
8 consulting project. This project counted for 42.5 percent of the final grade for the organizational behaviour course they participated in. The project took 15 weeks and the students had to work in teams existing of five people. Students formed those teams by themselves. Chatman and Flynn (2001) paid attention to diversity in sex, race and citizenship of students. They considered the effect of those demographic diversity characteristics on cooperative norms within teams. Individual perception of cooperative norms was measured by reports of students in which they express how much some statements were applicable to their team. Those statements were about the level of harmony, collaboration, cooperation, no self-interest and sharing within teams. Cooperative norms at team level were measured by averaging the individual cooperative norms perceptions within the teams. The level of cooperative norms was measured during the third week and at the end of the project, in the fifteenth week. Chatman and Flynn (2001) found that diversity had a negative effect on cooperative norms, but this negative effect of diversity on cooperative norms decreased significantly between the two measurements. People pay less attention to surface-level characteristics and more attention to deep-level characteristics as time passes by (Chatman & Flynn, 2001; Harrison et al., 1998; Jehn et al., 1999; McKnight, Cummings & Chervany, 1998). The negative effects of heterogeneity due to social categorization based on surface-level characteristics will diminish (Chatman & Flynn, 2001; Harrison et al., 1998). The level of communication within teams will increase over time and this leads to more cooperation within teams. This leads subsequently to improved performance (Chatman & Flynn, 2001).
Furthermore, Chatman et al. (1998) state that demographic diversity will be more beneficial in organizations with a collective culture. There is a negative relationship between demographic heterogeneity and interaction among team members (Chatman et al., 1998). Since there is a positive relationship between collectivistic cultures and interaction, it is useful to create a collectivistic culture within teams (Chatman et al., 1998). Chatman et al. (1998) did an empirical study with MBA students and noticed that within teams with a collectivistic culture, students interact more with each other and have less conflicts with each other, which results in better performance. Creating a collectivistic culture is especially useful for heterogeneous teams since these teams, in general, originally have less interaction (Chatman et al., 1998). A collectivistic culture can be created by emphasizing to individuals that they are part of a team (Chatman et al., 1998). This can be done by making values of the team or organization salient and ensuring that members share some goals (Chatman et al., 1998). Sharing the same goals and fate within a team is an important basis to create a collectivistic culture (Chatman et al., 1998; Dovidio, Gaertner & Validzic., 1998; Jehn et al., 1999). A low value-diversity within teams creates a high added value for teams (Chatman et al., 1998; Dovidio et al., 1998; Jehn et al., 1999). For example Jehn et al. (1999) found in their study that value-diversity
9 increases relational conflict, which results in decreased team functioning. An example to converge individual’s goals is to make wage dependent on organizational outcomes, in this way all team members will strive to a high organizational outcome and team performance will improve (Chatman et al., 1998).
To conclude, it is hard to define the direct relationship between diversity characteristics and team performance. However, processes underling diversity explain the relationship a bit more. Previous literature suggests that newly formed teams experience a higher negative than positive effect from diversity (Ancona & Caldwell, 1992; Bacon et al., 2001; Chatman et al., 1998; Dovidio et al., 1998; Williams & O’Reilly, 1998), largely due to the social categorization problem (Williams & O’Reilly, 1998). Williams and O’Reilly (1998) conclude in their overview of more than 80 studies on diversity that diversity on itself is most likely to have a negative effect on group functioning. This overall effect of diversity can become positive if the negative effects of social categorization diminish, for example due to the creation of a collectivistic culture (Ancona & Caldwell, 1992; Chatman et al., 1998; Dovidio et al., 1998; Williams & O’Reilly, 1998). According to this, the following hypothesis is formed:
Hypothesis 2: Homogeneous teams outperform heterogeneous teams in the short term.
2.4. PERFORMANCE OF SELF-FORMED TEAMS AND RANDOMLY FORMED TEAMS
In this final part of the theoretical framework some comparable studies will be considered. The first study is the empirical study of Mahenthiran and Rouse (2000) which investigates team performance of completely randomly assigned teams in comparison to team performance of teams which were formed by randomly assigning self-chosen pairs. The teams consisted of four or six people. The researchers analysed the differences in grades of the teams and the differences in satisfaction of the team members. They concluded that teams consisting of randomly assigned pairs performed better according to the obtained grades and the team members were more satisfied than in the teams where all students were completely randomly assigned (Mahenthiran & Rouse, 2000).
The second study is the experimental study by Chapman et al. (2006). In line with our study, Chapman et al. (2006) considered the differences between self-formed teams and randomly formed teams. By means of a survey about group projects among marketing students, Chapman et al. (2006) investigated the effect of the type of assignment, self-formed or random, on group dynamics, group outcomes and students’ attitude towards the group experience. Out of 15 group dynamics measures, eight measures show significant positive effects and only two measures show significant negative effects of self-formed teams relative to randomly formed teams. Out of the nine measures of
10 students’ attitudes only four measures were significant, which all show positive effects of self-formed teams relative to randomly formed teams on attitudes towards the group experience. The overall attitude towards the team is by members of self-formed teams significantly higher (p < 0.10) than by randomly formed teams. Out of the nine group outcome measures five measures were significant and these all show positive effects of self-formed teams on group outcome relative to randomly formed teams. The overall conclusion of Chapman et al. (2006) is that randomly formed teams are inferior to self-formed teams, since randomly formed teams perform less than self-formed teams on group outcomes, attitudes and dynamics.
According to the homophily principle, this paper predicts that self-formed teams are more homogenous than randomly formed teams (Feld, 1982; McPherson et al., 2001; Reagans, 2005). Since it seems that homogeneous teams perform better than heterogeneous teams when they are recently formed (Ancona & Caldwell, 1992; Bacon et al., 2001; Chatman et al., 1998; Dovidio et al., 1998; Williams & O’Reilly, 1998), this paper predicts that self-formed teams outperform randomly formed teams in the short term. This prediction is also in line with described comparable studies and leads to the following hypothesis:
11
3. METHOD
3.1. RESEARCH SETTING
To test the formulated hypotheses, this paper uses data from a third year course at the University of Amsterdam from the years 2013/2014 and 2014/2015. The course is compulsory for all third year Business and Economics students who follow the specialization Finance & Organization and the course is elective for exchange students and for students of the specializations Economics & Finance and Economics. Students have to follow this course for eight weeks. The first three weeks and week five till week seven contain one lecture and one tutorial each week. All lectures and tutorials take three hours. During lectures the coordinator explains the relevant theory and during tutorials students practice this theory with exercises. The fourth and final week of the course do not contain any lecture or tutorial and during the final week of the course students have to make an individual written exam.
The grade of the course consists for 70 percent of the individual written exam. Students can earn 0.5 bonus point on their exam grade by submitting a homework assignment in groups of three to five students. The students have one week to complete the homework assignment, week five. The other 30 percent of the final course grade is determined by a group assignment. For the group assignment students have to form teams of five people in the first week of the course. Since this is a course for third year students, this paper assumes that a major part of the students knows each other. Students are allowed to form teams by themselves. However, the students who do not form a team by themselves will be assigned to a team by the coordinator of the course. This team is either a self-formed team of less than five members or a team totally formed of people who did not assign themselves in a team. Hence, there are three categories of teams: self-formed, partly self-formed and randomly formed. In those teams, the students have to work on an assignment which has to be submitted in week four. This means that students have only the first three weeks to work together on the assignment. The group assignment is a case designed by consultants from an established consultancy company. In the case study students have to design and write a business strategy, a remuneration philosophy and a remuneration design for the CEOs of a publicly listed firm after some changes occur in the company. The set-up is the same in both years, the only difference is the company considered. The cases are graded by the coordinator of the course, based on the quality of the proposed solution, originality, quality of the argument and choice of the literature.
12
3.2. DATA COLLECTION
The data used in this paper is collected by the University of Amsterdam. I obtained the data from the coordinator of the course. The data was fully anonymized before the coordinator of the course handed the data over to me. I used the obtained data only in the interest of this paper. The data contain individual- as well as team- level data. First of all, for each participant the gender is known. Furthermore, the individual- level data consist of information about the individual result of the students; if someone earned the bonus point or not, if he/she participated in the final exam and/or final resit and in case of participation the grade is known. With the individual-level data it was possible to determine some team-level data, namely the male-female ratio and the team average of the obtained exam grades. Most important, the dataset shows how the teams are formed; by themselves, by the coordinator of the course or partly formed by themselves and partly formed by the coordinator. In the case that teams are partly formed by themselves, it is known how many of the students within the team chose each other and how many students are assigned to the team through the coordinator of the course.
3.3. RESEARCH SAMPLE
The data originally included 322 students, divided over 67 teams; 35 teams from year 2013/2014 and 32 teams from year 2014/2015. From those teams, 59 teams consisted of five students, six teams consisted of four students, one team consisted of two students and one team consisted of only one student. The teams which contained only one or two students are not included in the study. The reason to exclude those teams is because some social phenomena can only take place with a minimum of three individuals (Moreland, 2010). Moreland (2010) states that two individuals cannot be considered as a group. Because of this restriction, the number of teams in the total sample dropped from 67 to 65 and the number of students dropped from 322 to 319.
A relevant issue with the dataset is that 29 of the 319 students do not have an individual grade since they did not participate in the final exam or in the resit exam. Unfortunately, it is not known why these students did not participate in any of the exams. It is possible that they did not participate in the group assignment either, but that the team did not exclude them from the case grade the rest of the team earned. Since this is only speculative and not stated, those students cannot be totally excluded from the sample. If the speculation is true that the people who did not make any of the exams did not put in effort in the group case either, the composition of all team members could be different from the composition of team members that actually put in effort. Those compositions could differ in for example percentage male within a team.
13 In this paper the two extremes are considered. In the beginning the total sample will be investigated. Thereafter, a selected sample which excludes the 29 students that did not make any of the exams will be investigated. If the difference in compositions between the total sample and the selected sample seems large, all regressions done with the total sample will be done again but this time with the selected sample. This will be discussed in more detail in a separate chapter in the result section.
3.4. DATA VARIABLES
The variables used in this paper are analysed at team-level. Each variable will shortly be discussed.
DEPENDENT VARIABLE
Group grade. The group grade is used as performance measure. This grade is the same for all members of the team. This grade is given by the coordinator of the course and is based on quality, originality and choice of literature.
INDEPENDENT VARIABLES
Formation. There are three types of formation; completely self-formed, partly self-formed and completely randomly formed. There are two dummy variables included to cover the total self-formed teams and the partly self-self-formed teams. The teams randomly self-formed through the coordinator are the reference group, subsequently the dummy variable for the randomly formed teams is excluded from the regressions.
Diversity in Gender. The variable ‘diversity in gender’ will be included in the regression to investigate the effect of diversity in gender on team performance. Blau’s diversity index (1977) is used to measure the level of diversity in gender in teams. Blau’s diversity index is a widely accepted method to capture qualitative distinctions (Biemann & Kearney, 2010). Blau’s index measures diversity by dividing the number of people in a certain category by the total number of categories. This is shown by the following formula (Biemann & Kearney, 2010):
The p in this formula indicates the proportion of category i in the group (Biemann & Kearney, 2010). For example, if we look at gender diversity and a team consists of two boys and two girls, Blau’s diversity index will be 1- (0.52+0.52) = 0.5. The range of this index starts at a value of zero and the
maximum value is determined by the function of the number of categories ((k-1)/(k)) (Biemann & Kearney, 2010). In this paper there are only two categories; male and female. This implies that in this
14 paper the possible range for Blau’s diversity index is from 0 to 0.5 ((2-1)/(2)). Whereas a value of 0 indicates a perfect homogeneous group and a value of 0.5 indicates a perfect heterogeneous group. In this paper the actual range is equal to the possible range. There are some teams which are totally homogeneous (24 teams), some which are as heterogeneous as possible (31 teams) and some in between those extremes (10 teams). Teams consisting of five members cannot have a diversity index of 0.5, because there cannot be an exact equal fraction of male and female in a team with an odd number of members. For teams consisting of five members the maximum attainable index of diversity is 0.48, which is the case in 30 out of the 31 teams which are as heterogeneous as possible. Since this study includes 219 males and only 100 females, on average there are more males than females in teams and it is likely that there are more teams solely consisting of males than solely consisting of females. However, we do not have to account for this in comparing Blau’s diversity indexes between teams, since all teams draw members from the same sample.
Diversity in Ability. The variable ‘diversity in ability will be included in the regression to investigate the effect of diversity in ability on team performance. Just as in the study of Hansen et al., (2006), the standard deviation of the individual grades relative to the average individual team grade is used as measure for the level of diversity with respect to ability. This paper predicts that diversity has a negative effect on team performance in the short term (Ancona & Caldwell, 1992; Bacon et al., 2001; Chatman et al., 1998; Dovidio et al., 1998), therefore the coefficients of both gender diversity and diversity in ability are expected to be negative.
CONTROL VARIABLES
There are several control variables included in the regression.
Average individual grade for the course. The individual grade achieved for the course is used as control variable to capture the ability of the team members. On team level the ability is measured as the average individual exam grade within the team, which is a good measure according to Gigone and Hastie (1993). For students who only participated in the final exam, this grade is taken, for students who only participated in the resit exam, this grade is taken and for students who participated in both the final exam and the resit exam, the highest of those grades is taken. The exam grades are just like the case grades determined by the coordinator of the course. It is expected that a high average individual grade for the course will affect the case grade in a positive way (Gigone & Hastie, 1993). Teams which consist of high ability individuals are expected to perform better in the group assignment (Gigone & Hastie, 1993).
Percentage Male. This variable is included to account for different effects of males and females within teams on team performance. The percentage of males within teams is used as measure. It is predicted that this coefficient is not significantly different from 0 since nothing
15 indicates that males or females perform different in the group assignment.
Percentage Male2. This control variable is added to investigate if the relation between percentage male within a team and case grade is linear or not. It is expected that the relationship is non-linear (e.g. Clement & Schiereck, 1973; Hoffman & Maier, 1961). Therefore, the coefficient is expected to differ from zero. If this variable is not significant, it will be excluded from the regression and only ‘Percentage Male’ will be included in the regression.
Year. Because the data is originated from two different years, 2013-2014 and 2014-2015, in all regressions the control variable ‘year’ is added. Year is a dummy, with a value of zero if the data is from year 2013-2014 and a value of one if the data is from year 2014-2015. This variable is expected to be zero, since there are almost no differences between the years. The assignment is comparable as well as the grading.
16
4. RESULTS
4.1. HYPOTHESIS 1
The first hypothesis predicts that self-formed teams are more homogeneous than randomly formed teams. In this paper, the level of diversity is based on gender diversity and ability within the team. To measure the level of diversity in gender Blau’s diversity index is used and to measure level of diversity in ability the standard deviation of individual grades relative to the average individual grade within a team is used. Table 1 gives an overview of all variables.
Table 1: Descriptive statistics of the sample (N=65)
Self-formed Partly self- formed Randomly formed Total Number of teams 42 15 8 65 Number of students 205 74 40 319
Average number of students per team 4.88 4.93 5 4.91
Average percentage male within teams 68.57 % 64.00 % 77.50 % 68.62 %
Average individual grade in teams 5.64 5.24 4.89 5.45
Average value of Blau’s diversity index per team 0.270 0.299 0.300 0.280 Average value of std. dev. of indiv. grade per team 1.117 1.299 1.347 1.187 Percentage of members who did at least one exam 94.29% 86.67 % 82.50 % 91.08 %
Average Case grade 8.01 7.47 7.06 7.77
The average Blau’s diversity index is 0.300 for randomly formed teams, 0.299 for partly self-formed teams and 0.270 for self-formed teams, this is also shown in table 2. Those values indicate that the level of gender diversity is the highest in randomly formed teams, thereafter in partly self-formed teams and the lowest in self-formed teams. This is in line with hypothesis one which predicts that self-formed teams are more homogeneous than randomly formed teams. However, performing the Bonferroni, Scheffe and Sidak tests shows that none of the differences in level of diversity between self-formed, partly self-formed and randomly formed teams is significant (see appendix A). Based on these results, hypothesis one with respect to gender is rejected.
The average standard deviation of individual grades relative to the average individual grade within a team is 1.347 for randomly formed teams, 1.299 for partly self-formed teams and 1.117 for self-formed teams, this is also shown in table 2. Those values indicate that diversity with respect to ability is the highest in randomly formed teams, thereafter in partly self-formed teams and the lowest in self-formed teams, which is in line with hypothesis one. However, when performing the Bonferroni, Scheffe and Sidak tests on the differences in standard deviations again none of the differences between types of teams is significant (see appendix A). Therefore, hypothesis one with respect to ability is rejected as well. By analysing diversity in gender and ability none of the diversity levels is significantly lower in self-formed teams relative to partly self-formed teams or randomly
17 formed teams and the diversity levels in partly self-formed teams are not significantly lower than in randomly assigned teams. Although the results are in the good direction, with the highest levels of diversity for randomly formed teams, lower levels of diversity for partly self-formed teams and the lowest levels of diversity for self-formed teams, they are significant. Since the results are non-significant with respect to gender as well as with respect to ability, hypothesis one, which states that self-formed teams are more homogeneous than randomly formed teams, is rejected.
Table 2: Statistics of diversity within teams with respect to gender and ability
Blau’s diversity index on gender Diversity in ability
Type of team N Mean diversity index
std. dev. Mean std. dev. individual grade std. dev. Randomly formed 8 0.300 0.199 1.347 0.419 Partly self-formed 15 0.299 0.225 1.299 0.511 Self-formed 42 0.270 0.230 1.117 0.485 Total 65 0.280 0.222 1.187 0.486
4.2. HYPOTHESIS 2 AND 3
To investigate hypothesis two and three, a number of multiple regressions will be done with group grade as dependent variable. Since hypothesis one is rejected, which implies that self-formed teams are not more homogeneous than randomly formed teams, the level of diversity and the type of formation will not be correlated to such an extent that including them both in one regression will lead to multicollinearity. Therefore, the diversity measures and type of formation can be included as independent variables simultaneously. It is important to mention that since some relevant variables are not available, like personality profiles or level of cooperation within teams, there will be some omitted variables in this study. Even though the regressions are not completely reliable due to these omitted variables, the regressions are still useful to consider relations between the variables. This will be explained in more detail in the discussion section of this paper.
The first model only includes the type of formation as independent variable and the dummy variable ‘year’ as control variable. This model shows that both self-formed teams (β = 0.96, p < 0.01) as well as partly self-formed teams (β = 0.43, p < 0.05) have a positive relation with case grade relative to randomly formed teams. An additional test shows that the relation of self-formed teams with case grade relative to partly self-formed teams is also significant positive (F = 12.36, p < 0.01). These results support hypothesis three which states that self-formed teams outperform randomly formed teams in the short term.
18 In model 2 the control variables ‘percentage male’, ‘percentage male2’ and ‘average
individual grade’ are added. Self-formed teams (β = 0.90, p < 0.01) and partly self-formed teams (β = 0.39, p < 0.10) have again a significant positive relation with case grade relative to randomly formed teams, as well as self-formed teams have again a significant positive relation on case grade relative to partly self-formed teams (F = 9.28, p < 0.01). These results once more support hypothesis three. From the added control variables only ‘percentage male2’ is marginally significant (β = -1.58, p <
0.10), the other variables are not significant (βpercentage male = 1.63, p > 0.10; βaverage individual grade = 0.11, p > 0.10). The regression is also done without ‘percentage male2’, but this model
explains less (F = 6.60 < F = 6.71) and excluding ‘percentage male2’ from the regression makes
‘percentage male’ still not significant (see appendix B). Therefore ‘percentage male2’ is included in
model 2. The marginally significant negative ‘percentage male2’ indicates that the function of the
variable ‘percentage male’ is a parabola that opens downward. This finding suggests that having male in a team improves group performance up to some point and from that point on the effect on group performance becomes negative. This result can be due to the issue that the control variable ‘percentage male’, besides saying something about the effect of male/female on group case, says something about the effect of diversity in gender on group performance. This is the case since a variable of 0.5 indicates that the team is as heterogeneous in gender as possible, 0 indicates that the team is homogeneous in gender since the team only consists of females and 1 indicates that the team is homogeneous in gender since the team only consists of males. According to this issue, the results suggest that diversity is beneficial, since having only female or male in a team is inferior to a mixed-gender team. This suggestion about diversity is not in line with what was predicted by hypothesis two, namely that homogeneous teams outperform heterogeneous teams in the short term.
In model 3 the diversity measures of gender and ability are added as control variables. The diversity measure of ability, average standard deviation of individual grades to average individual grades within teams, is not strongly correlated with the measure of ability, average individual grade (see appendix C). Therefore, both can be included in the same model. However, since there is a strong correlation between the diversity measure of gender, the control variable ‘percentage male’ and the control variable ‘percentage male2’, the control variables are excluded in model 3. In model 3
self-formed teams (β = 0.94, p < 0.01) and partly self-formed teams (β = 0.38, p < 0.05) have again a significant positive relation with case grade relative to randomly formed teams and self-formed teams have again a significant positive relation with case grade relative to partly self-formed teams (F = 11.42, p < 0.01), which is in line with hypothesis three. Average individual grade has a significant positive coefficient (β = 0.15, p < 0.10), this means that a higher average individual grade has a
19 positive relation with case grade, this is in line with what was predicted. However, both measures of diversity have a significant positive coefficient as well (β gender diversity = 0.72, p < 0.05; β ability diversity = 0.35, p < 0.05), meaning that a higher diversity leads to a higher case grade. This is not in line with hypothesis two, which predicts that diversity would not be beneficial in the short term. By comparing the results of model 2 and model 3, it seems that diversity in gender and ability is more important than the presence of explicitly male or female in a team and the individual ability in a team, since only the diversity measures are significant at a significance level of five percent. Furthermore, the variable ‘year’ is non-significant, which is the case in all the models, see table 2, and is in line with what is predicted.
With respect to hypothesis two, in models 2 and 3 the coefficients of diversity are significant positive, meaning that diversity has a positive relation with team grade. Therefore, hypothesis 2, which predicts that diversity would not be beneficial in the short term, is rejected. With respect to hypothesis three, in all regressions the relation of self-formed teams with group grade (in all models p < 0.01), as well as the relation of partly self-formed teams with group grade (p < 0.05 and p < 0.01 in model 2) relative to randomly formed teams is significant, see table 2. Furthermore in all regressions the relation of self-formed teams with group grade relative to partly self-formed teams is significant positive (in all models p < 0.01). In all three models the coefficients of the dummies ‘self-formed’ and partly ‘self-‘self-formed’ are almost similar even though other control variables are added in the regressions. These results support hypothesis three. Therefore, hypothesis 3, which predicts that self-formed teams outperform randomly formed teams in the short term, is supported.
20 Table 3: regression analysis of the relationship between team formation and level of diversity on group performance (group grade)
Independent variables Model 1 Model 2 Model 3
Self-formed (1 = yes, 0 = no) 0.957*** (0.209) 0.896 *** (0.204) 0.942*** (0.200) Partly self-formed (1 = yes, 0 = no) 0.434** (0.216) 0.386* (0.200) 0.383** (0.188) Percentage male 1.635 (1.099) Percentage male2 -1.583* (0.865)
Average individual grade 0.114
(0.098)
0.153* (0.091)
Blau’s diversity index on gender 0.723**
(0.344) Average std. dev. on individual
grade within team
0.349** (0.143) Year (1 = 2014, 0 = 2013) -0.133 (0.148) -0.056 (0.139) -0.057 (0.131) Constant 7.113 (0.184) 6.248 (0.647) 5.648 (0.462) N 65 65 65 F 8.59 6.71 9.53 R2 0.270 0.361 0.417 Adjusted R2 0.235 0.295 0.357
The unstandardized coefficients are given, between parentheses are the robust standard errors. ***P < 0.01 **P < 0.05 *P < 0.10
4.3. ADDITIONAL ANALYSES
Since the level of self-selection within partly self-formed teams is diverse it seems useful to consider the levels of self-selection in more detail instead of just labelling them self-formed, partly self-formed or randomly formed. All randomly formed teams have the same self-selection level of 0.0 and all self-formed teams have the same self-selection level of 1.0. However, the level of self-selection within partly self-formed teams is diverse and ranges between 0.4 and 0.8, meaning that at least 40 percent and at maximum 80 percent of the team members chose each other within the partly self-formed teams. To say something about the effect of the level of self-selection on group grades an additional regression will be done. Model 3 will be used as basis since this model includes both type of formation and level of diversity. The only difference with model 3 is that the dummies for the type of formation are replaced by a continue variable, namely ‘Level of self-selection’.
𝐶𝑎𝑠𝑒 𝑔𝑟𝑎𝑑𝑒 = ∝ + 𝛽1 ∗ 𝐿𝑒𝑣𝑒𝑙 𝑆𝑒𝑙𝑓 𝑆𝑒𝑙𝑒𝑐𝑡𝑖𝑜𝑛 + 𝛽2 ∗ 𝐴𝑣𝑔. 𝐼𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝐺𝑟𝑎𝑑𝑒 + 𝛽3 ∗ 𝐷𝑖𝑣𝑒𝑟𝑠𝑖𝑡𝑦 𝐼𝑛 𝐺𝑒𝑛𝑑𝑒𝑟 + 𝛽4 ∗ 𝐷𝑖𝑣𝑒𝑟𝑠𝑖𝑡𝑦 𝐼𝑛 𝐴𝑏𝑖𝑙𝑖𝑡𝑦 + 𝛽5 ∗ 𝑌𝑒𝑎𝑟 + 𝜀
21 Within this regression linearity is assumed, even though we do not know yet if the relationship is linear. The regression is done in this way, since there are too few observations of all levels of self-selection available to do it otherwise. The performed regression shows that he level of self-self-selection has a significant positive relation with case grades (β = 0.98, p < 0.01) (see appendix D). However, doing the regression with only the partly self-formed teams gives a non-significant positive coefficient for the variable ‘Level of self-selection’ (β = 1.48, p > 0.10) (see appendix D). This result shows that the level of self-selection within partly randomly formed teams does not have a significant positive relation with case grades. However, this non-significant result might be due to the small sample of partly self-formed teams. The non-significant relation between level of self-selection and case grades in the regression which only includes the partly self-formed teams shows that the results of the regression which includes all teams is mainly driven by the extreme types of teams, namely completely self-formed and completely randomly formed. Even though the results cannot be used to draw a conclusion about the relation between the level of self-selection and group grades, the results are again supporting hypothesis three, since the results again show the positive relation of self-formed teams with group grades relative to randomly formed teams.
As stated in the ‘research sample’ section, 29 students did not participate in any of the exams. The team compositions of the total sample, which are used up till now, will be compared to the team compositions of a selected sample which excludes the 29 students who did not participate in any of the exams (see appendix E). In the selected sample the number of students decreases from 319 to 290 while the number of teams stays the same compared to the total sample. Therefore, the average number of students per team decreases, now 34 teams consisting of five students, 27 teams consisting of four students and four teams consisting of three students. Resulting in a drop of the average number of students per team from 4.91 to 4.46 students. The total average percentage male within teams falls slightly from 68.62% to 67.95%.
Since average individual team grades are calculated in both samples on the basis of available grades, for both samples exactly the same grades are used. Therefore, there is no difference in the total average individual grade in teams. There is only a slight change in average case grade between self-formed (from 5.64 to 5.62) and partly self-formed (from 5.24 to 5.23) teams. This small change is due to the categorization of two teams. Two teams are categorized as a partly self-formed team when looking at the total sample and are categorized as a self-formed team when looking at the selected sample. Those teams consist of five students, from which four select themselves and one was randomly assigned to the team by the course coordinator. The ones who were randomly assigned did not participate in any of the exams. For this reason, the teams are categorized as completely self-formed in the selected sample. Those two teams had a 7 and 7.5 as case grades. This
22 leads to a slight decrease in the average case grade of the self-formed teams and a slight increase in the average case grade of the partly self-formed teams when looking at the selected sample relative to the total sample. The total average case grade stays the same, since exactly the same case grades are considered. The ability diversity measure among the total and the selected sample stays the same as well, since this measure is also in both samples based on exactly the same grades. The gender diversity, on the other hand, will change. This is because the percentage of male changes. However, this change will be very small since the percentage of males changes with less than one percent. Therefore it is not needed to calculate the gender diversity again for specific the selected sample.
To conclude, the differences between the total and the selected sample are really small with respect to average percentage male and gender diversity within teams and for average individual grade and ability diversity within teams the change is even zero. The only large differences are the change in group size and the relative amount of available data. For the selected sample all individual grades are known while this is not the case in the total sample. To account for the decrease in team members and the increase in the percentage known grades, the control variable ‘percentage grades known’ can be added. When the speculation is right, which is assumed in this part of the paper, the ‘percentage known grades’ equals the percentage of team members who put in effort in the group assignment. This implies that the variable for teams of which all students participated in at least one exam has a value of one and if nobody participated in any exam the value will be zero and thereby the decrease in team size will be one hundred percent. Since we assume that effort has a positive relation with performance, we predict the coefficient of ‘percentage known grades’ to be positive. The control variable ‘percentage grades known’ is added in model 2 and 3 of table 3. The results show that the percentage grades known has no significant effect on group grade, see table 4. Those results differ from what was expected. A possible explanation can be that the speculation was not true, which means that the people who did not participate in any of the exams did participate in the group assignment. Another explanation could be that the sample is too small to get a significant relation. By comparing the results with the results in table 3 the coefficients of self-formed and partly self-formed teams are comparable. Even though another control variable is added, the results again support hypothesis 3.
23 Table 4: regression analysis of the relationship between team assignment and level of diversity on group performance (group grade), with additional control variable ‘percentage grades known’
Independent variables Model 1 Model 2
Self-formed (1 = yes, 0 = no) 0.856*** (0.205) 0.911*** (0.204) Partly self-formed (1 = yes, 0 = no) 0.374* (0.196) 0.371* (0.186) Percentage male 1.703 (1.137) Percentage male2 -1.607* (0.884)
Average individual grade 0.105
(0.099)
0.145 (0.093)
Blau’s diversity index on gender 0.711**
(0.343)
Average std. dev. On individual grade within team 0.345**
(0.144)
Percentage grades known 0.430
(0.501) 0.294 (0.518) Year (1 = 2014, 0 = 2013) -0.044 (0.141) -0.050 (0.134) Constant 5.893 (0.738) 5.453 (0.587) N 65 65 F 5.91 8.14 R2 0.365 0.419 Adjusted R2 0.287 0.348
The unstandardized coefficients are given, between parentheses are the robust standard errors. ***P < 0.01 **P < 0.05 *P < 0.10
24
5. DISCUSSION
5.1. FINDINGS
The aim of this paper is to answer the question whether self-formed teams perform better than randomly formed teams in the short term. The results support the prediction that self-formed teams outperform randomly formed teams in the short term. Both regressions, the regressions in which self-formed teams were included as dummies and the regressions were the level of self-selection is included as explanatory variable, show that self-formed teams have a significant positive relation with group grade relative to randomly formed teams. However, it is remarkable that the hypothesis which predicts that self-formed teams outperform randomly formed teams in the short term is accepted, while the hypotheses underlying this hypothesis are rejected.
The first hypothesis predicted that self-formed teams are more homogeneous than randomly formed teams. The results of this study show that self-formed teams had the highest average level of diversity with respect to gender and ability, followed by partly self-formed teams and randomly formed teams had the lowest level. However, those differences appeared to be non-significant. Therefore, this hypothesis is rejected. The second hypothesis predicted that homogeneous teams will outperform heterogeneous teams in the short term. This hypothesis is rejected as well. Significant positive coefficients of diversity in gender and ability show that diversity has a positive effect on team performance in the short term. This is in line with some previous studies, for example Hoogendoorn et al. (2013) and Hoffman and Maier (1961). The lack of support for hypothesis one and two could be explained by the way in which this study analysed diversity. This study focused on gender and individual grade to measure the diversity level within teams. Other relevant diversity factors, like age and ethnicity, might influence the level of diversity within a team as well and subsequently influence the relationship between diversity and team performance.
Besides showing a significant positive relation between self-formed teams and team performance relative to randomly formed teams, the results of this study show more positive relations. They also show a more positive relation between partly-self formed teams and team performance than between randomly formed teams and team performance and a more positive relation between self-formed teams and team performance than between partly self-formed teams and team performance. The investigation to the relation between the level of self-selection within partly self-formed teams and team performance gives a non-significant positive relation. A reason for this non-significant result could be the small sample; only 15 teams are partly self-formed.
25
5.2. CONTRIBUTIONS AND IMPLICATIONS FOR THE LITERATURE
Up till now, not a lot of studies investigated the most optimal type of team formation (Mahenthiran & Rouse, 2000; Chapman et al., 2006). In line with Mahenthiran and Rouse (2000) and Chapman et al. (2006), this study predicted that the outcomes of teams improve if people have some control over their team composition. However, there are some differences between those studies and this study. Whereas Mahenthiran and Rouse (2000) considered partly self-formed and randomly formed teams and Chapman et al. (2006) considered self-formed and randomly formed teams, this study considered all three types of team formation. In the study of Mahenthiran and Rouse (2000) the grade given by a teacher was considered as measurement for performance, just as in this study. However, an important difference with their study is that performance of the teams in our study is one constant grade which is applicable for the whole team, whereas Mahenthiran and Rouse (2000) measured performance by a grade which varies among team members. An important difference with the study of Chapman et al. (2006) is that our study focuses on an outcome measured by a third party (group grade imposed by the teacher), whereas Chapman et al. (2006) looked at how students experience the team dynamics and outcomes within self-formed teams compared to randomly formed teams. The results of our study contribute to the assumption that outcomes of teams improve if students have some control over their team compositions, by validating that self-formed teams have a positive relation with team outcomes measured by a third party relative to randomly formed teams and partly self-formed teams. However, further research is necessary to investigate the reason(s) underlying the positive relation between self-formed teams and team performance.
5.3. PRACTICAL IMPLICATION
S
The findings of this study are highly relevant for all parties involved by working in teams. Especially universities and other educational institutions can benefit from the findings in this study. Teachers can decide to give students more control on the formation of teams to improve their performance. Whether previous studies already showed that students have a more positive attitude towards working in teams and value their outcomes higher with members they self-select, this paper shows that this self-selection also has a positive influence on outcomes measured by a third party.
5.4 LIMITATIONS
Although the results of this study are valuable, this study has several limitations. The first limitation is about the external validity of this paper. The results found in this study might also be relevant for
26 teams in companies. However, additional research has to prove if the results of this paper are also applicable for those teams. For example Chatman and Flynn (2001) investigated the effect of diversity within teams first with MBA students and thereafter with officers from business units in financial services. By means of this, Chatman and Flynn (2001) could determine if their results were generalizable to work settings in companies. This might also be very relevant for this study and might increase the generalizability of the findings.
The second limitation is about the internal validity. To test the first hypothesis about diversity within teams, the only factors which are used to measure diversity are gender and ability. However, there are many other factors which are relevant to test this hypothesis. For example personality profiles of members, ethnicity, age or intelligence. Adding diversity measures about other relevant characteristics might lead to other results. Hence, there is omitted variable bias.
Furthermore, it could be that there are differences between the type of students in self-formed teams and the type of students in randomly self-formed teams, which causes (partly) the effect on team performance. For example, it could be that the more social students are in the self-formed teams or that students who are in the randomly formed teams did not form a team by themselves since they are not much present at the university and do not know fellow students. Another explanation could be that they had difficulty in selecting a team since they do the course for the second time or they are exchange students. This means that the sample of the teams is not a totally random sample since the teams consist of certain types of students. This might make the results of this paper biased.
The last limitation is about the construct validity of this paper. In total there are 65 teams in the sample. 42 teams are self-formed, 15 teams are partly self-formed and only 8 teams are randomly formed. The results of this study would have been more reliable if more partly self-formed and randomly formed teams participated in this study.
5.5. DIRECTIONS FOR FUTURE RESEARC
H
When taking the implications and limitations of this study into account there are several opportunities for future research. First, future research could investigate the same research question but should enlarge the amount of participating teams. By means of this, the construct validity of the findings will improve. Second, to overcome the problem of unknown differences between the type of people in self-formed teams and the type of people in randomly formed teams, it is useful for future research to make the research setting somewhat different. In this study there is no random assignment of student to teams which can lead to selection, which means that students in
