Team-based learning is a type of collaborative learning that is increasingly prevalent throughout all disciplines in higher education (Espey 24; Kim et al. 225–226). It is a common practice for faculty members to include team-based learning into an undergraduate curriculum, and for some, into a postgraduate curriculum. Inclusion of teamwork and team-based assignments can be commonly found in engineering, business, and social sciences programs across the world. Specifically, some faculty members will include team-based assignments in first-year introductory courses as well as final-year capstone projects. Inclusion of teamwork is considered to be beneficial to students in terms of learning to be good team members and this is why Gardner and Korth mentioned that “To remain innovative and competitive, businesses are looking for employees who can work and learn effectively in teams” (28). Besides, previous studies have shown that by learning in teams, students’ academic achievement and self-efficacy may increase.

Nonetheless, team-based learning is not a universally positive experience for all students, as some of the obstacles in teamwork include communication difficulties, uneven work allocation, free-riders and unfair grading experiences. (Wilson et al. 794; Pfaff and Huddleston 38; Medaille and Usinger 240–42). As teams are often made up of students who come from different backgrounds, it is normal for them to worry about potential obstacles in team-based learning. For example, all ten participants interviewed as part of Medaille and Usinger’s study noted that they had negative experiences with team-based collaborative learning due to the presence of free riders in their groups (Medaille and Usinger 246). In another study, non-high-achieving students were found to have difficulties in expressing their ideas, as high-achieving students in the groups had prepared for the projects on their own and were more likely to persuade the other students to follow their ideas rather than negotiating to resolve any conflicts. (Lee et al. 423; Lee and Lim 222)

In order to combat the perceived inequalities and negative experiences expressed by students regarding team-based projects, faculty members have introduced and developed techniques to assist in the teamwork process. (Chin et al. 3). Both the use of computer-supported collaborative learning environments and the use of peer mentors to assist teams throughout the semester have shown beneficial outcomes for students on teams. (Chin et al. 4; Ruël et al. 17–18). Teamwork assessment and support tools such as CATME encourage students to rate their teammates and themselves, while instructors can easily retrieve large amounts of data gathered by the tools. (Beigpourian et al. 11; Chin et al. 5; Maneeratana and Sripakagorn 5). With the help of these tools, instructors are able to look at one of the main aspects of a student’s negative experience of students in team-based learning, such as communication difficulties, and try to help the team out if there are disruptions in team dynamics (Beigpourian et al. 11)

Communication difficulties in teams can be due to students being reluctant to share their thoughts or just being shy and introverted. These may be the reasons why students are quiet in teams, but other reasons for being silent are often the result of personal, social, academic, cultural and contextual constraints (Medaille and Usinger). While previous studies have explored quiet students’ behaviors (Jin) and how they perceive themselves in collaborative learning (Medaille and Usinger), this study was designed to understand the following research questions (RQ):

RQ 2: To investigate whether there is a relationship between self-rated previous team experiences (number and positive/negative valence) for the three variables mentioned. It is crucial for instructors to understand these two questions so that they can divide the students into groups that fit the students’ personality and traits based on the students’ responses to a survey administered at the beginning of term, rather than randomly grouping students. In order to test the research questions mentioned above, I propose the following hypothesis to be investigated in this study:

It is common to have students who tend to listen more and speak less in a group. These students are considered quiet and will often express agreement with the thoughts of others regardless of whether they actually agree with the ideas (Medaille and Usinger 242;

While previous studies find these relationships (Medaille and Usinger 254), it is important to reproduce this finding quantitatively. Thus, by calculating the correlation between variables to find the R-squared value, we are able to be more confident in saying that there is a relationship between “Extraversion” & “BT_BelongingConcern” and “Extraversion” & “SpeakUp” among the students who responded to the survey.

I will be using Kendall’s tau-b (τb) statistic to calculate the correlation between the variables mentioned above. τb is chosen over other statistics to calculate the correlation as τb is the better in calculating the correlation between two ordinal variables (Khamis 159) as the questions in the survey were in the form of seven-point scales with identified end points.

For the two hypotheses above, I believe that when a student has many teamwork experiences or has had prior good experience with teamwork, they are able to recognize the benefit obtained through collaborative learning in groups such as increased individual achievement and persistence when facing adversity (Pfaff and Huddleston 38). Many prior teamwork experiences might have given students the confidence and more understanding of what to expect from a class that contains team projects while prior positive experiences might make the students want to work as a team again. Thus, both of the 2a and 2b hypotheses will use the same null and alternative hypotheses but will be tested using different smaller filtered data sets (as will be explained in the

After testing the three hypotheses mentioned above, I will continue this paper by calculating which predictors have a lower loss in predicting the variable “Extraversion” (rate of a student speaking up in groups) and will also perform cluster analysis on the students to facilitate team formation in the future.

The data for this study was collected from 2088 students enrolling in Engineering, Business, Informatics, and Architecture courses at the University of Michigan using a team assessment tool. The students answered a Beginning of Term survey before they were put into groups by their respective instructors. Therefore, the survey used in this research will collect the students’ personality and traits before the semester started and before they were divided up into teams. Although the students are from different courses, a similarity between these students is that the courses are conducted in a team-based collaborative learning format. In each course, students are required to complete project(s) assigned by the instructors in their respective teams. Projects vary by courses, but most of the projects require students to brainstorm ideas, solve challenges, and present their findings or products.

The survey forms contained different types of questions that vary from courses to courses. Among the questions included in the initial survey, 13 of them are the same across the 17 different courses. The final cleaned data is stored in a single file containing only the responses to the 13 questions. Note that while the team assessment tool included more questions and other assignments that students had to complete each week, only five variables from the responses are studied for the purposes of this project. In the original data file, six participants that contain NA values in one or more variables were removed in order to prevent error from occurring, leaving us with 2082 responses. Since the total number of samples collected was 2088, the 6 samples removed will not affect the computation in any important way. Some basic analysis on the five variables can be seen in

Most students report many team experiences.

Most students report fairly positive (4) or very positive (5) experiences in their previous teamwork.

The SpeakUp variable appears bimodally distributed, with more people toward the “It’s easy for me to speak up about my ideas or preferences even if it disrupts my group” end of the scale than the “I’d rather hold back ideas or preferences if my group stays happy” end.

The Extraversion variable appears bimodally distributed, with more people toward the “I often speak up in groups” end of the scale than the “I tend to listen more than speak” end.

The BT_BelongingConcern variable appears to be right-skewed, with more people toward the “I expect to fit right into the course” end of the scale than the “I expect to feel pretty out of place in the course”.

The top-right portion of the density plot is denser, with more students toward the “It’s easy for me to speak up about my ideas or preferences even if it disrupts my group” and “I often speak up in groups” end.

The top-left portion of the density plot is denser, with more students toward the “I expect to fit right into the course” and “I often speak up in groups” end.

For this study, I used the “Beginning of Team” (BoT) survey which comprised a total of N = 2082 students (6 students’ responses were removed as mentioned above). The students responded to the following survey questions at the beginning of the Fall 2020 semester. Ordinal data was converted into numerical data so that computation can be carried out easily. I analyzed all the following variables in the range of 1–7 (or 5 in some cases) without performing any modification such as mid-ranking as I believed the students had clearly expressed their opinion using the Likert scales. The variables that were studied in this paper and its details are as:

This paper aims to study students’ self-rated “Extraversion” (tendency of students being quiet or speaking up in a group) and to understand what might affect students’ perception that they will be quiet in a team. Thus, “SpeakUp” and “BT_BelongingConcern” are chosen as predictors since they have the highest correlation coefficient with “Extraversion” as observed in

To answer the research questions, I first performed data visualization to see if there are any interesting trends among the data. As the data are discrete, most of the points ended up overlapping one another, thus a density plot in the form of a heat map was chosen so that the trends can be observed clearly as in

After looking at the mean score of “SpeakUp”, “BT_BelongingConcern”, and “Extraversion”, I was curious to find the population mean for these variables. In order to approximate the means, I used the Bootstrap method. The 95% confidence interval was calculated using R’s in-built boot.ci() function. As the sample size was large and the three variables were approximately normally distributed, I used the normal confidence interval of the function. The normal confidence interval can be expressed as the following:

After obtaining the basic information about the three variables to be tested, bootstrap is once again used to test the third hypothesis by bootstrapping 10000 times (B=10000). The data was separated out into four smaller datasets, which were data of students with many teamwork experiences (6 or more times) and students with less or no teamwork experience (less than 6 times); and data of students with past negative experiences (score of 3 or less) and students with past positive teamwork experience (score of 4 or more). For hypothesis 2a, the difference in means for the “SpeakUp”, “BT_BelongingConcern”, and “Extraversion” variables were calculated through:

For hypothesis 2b, the difference in means for the “SpeakUp”, “BT_Belongingness”, and “Extraversion” variables are calculated through:

Moving on, assuming that for some reasons, the instructors were unable to access the students’ Extraversion scores and had to predict students’ rate of speaking in groups based on other variables collected in the survey form, then it is essential to figure out what variable is the best predictor in determining a student’s Extraversion score. In order to achieve this, Leave-One-Out Cross-Validation can be used. In this study, a cost of one would be paid if quiet students are classified as talkative and vice-versa. The cross validation implementation uses 10000 replications to determine which of the two variables has the lowest average loss.

In addition, hierarchical clustering is used to cluster the students so that when new points (future students’ responses) are obtained, instructors will have a better understanding of which clusters the students belong to, easing lecturers in the process of assigning the students into teams.

Before I start testing the hypotheses, performing prediction or clustering the data, I first computed the correlation between each variable and the “Extraversion” variable to determine the best two predictors to be used. Thus, according to

Correlation coefficient between variables and “Extraversion”

| cor.val | | Predictors |
---|---|

1 | Extraversion |

0.3485 | SpeakUp |

0.2217 | BT_Belongingness |

0.168 | ManyTeamEXP |

0.1056 | BT_PastPositive |

0.0747 | Procrastination |

0.0654 | Group_Preference |

0.0481 | PositiveExp |

0.0447 | Control |

0.0337 | BT_PastDiverse |

0.0033 | BT_PastWorkDifferent |

After finding the variables that have the highest correlations with “Extraversion”, the sample mean, bootstrapped mean confidence interval, bias and mean squared error (MSE) of “Extraversion”, “SpeakUp”, and “BT_Belongingness” are calculated as shown in

Mean, confidence interval, bias, and MSE of the “Extraversion”, “SpeakUp”, and “BT_BelongingConcern” for overall sample

Variable | Mean | Bootstrap C.I. (95%) | Bias | MSE |
---|---|---|---|---|

Extraversion | 4.5269 | (4.463, 4.591) | –0.00013 | 0.001073 |

SpeakUp | 4.4424 | (4.386, 4.499) | –0.0001845 | 0.0008213 |

BT_BelongingConcern | 3.2051 | (3.147, 3.263) | 0.0001997 | 0.00087 |

From _{b} = 0.32 with a confidence interval of (0.3160, 0.3810). Therefore, since 0 is not included in the confidence interval, I will reject the null hypotheses, _{0} in favor of the alternate hypotheses. This result actually agrees with the conclusions obtained in previous research where quiet students will often express agreement while talkative students are seen to be quite willing to disagree with other group members.

From _{b} = –0.2217 with a confidence interval of (–0.2560, –0.1875). Therefore, since 0 is not included in the confidence interval, I will reject the null hypotheses,

From

Difference in mean between students with many teamwork experiences and less teamwork experience and confidence interval

Variable | ∆ | Bootstrapped C.I. (95%) |
---|---|---|

Extraversion | 1.989247 | (0.976, 6.086) |

SpeakUp | 0.8129032 | (–1.2603, 3.8474) |

BT_BelongingConcern | –1.651613 | (–5.334, –0.147) |

Thus, I will reject the null hypothesis in favor of the alternate hypothesis that there is indeed a difference in the mean score for “Extraversion” and “BT_BelongingConcern” among the two types of students. However, there is no clear evidence for me to reject the null hypothesis to conclude that there is a difference in mean score for the “SpeakUp” variable among the two types of students.

Difference in mean between students with positive teamwork experiences and negative teamwork experience and confidence interval

Variable | ∆ | Bootstrapped C.I. (95%) |
---|---|---|

Extraversion | 1.674797 | (0.218, 5.976) |

SpeakUp | 0.6747967 | (−1.3437, 3.7271) |

BT_BelongingConcern | −1.658537 | (−5.416, −0.157) |

From Table 8, it is observable that there is a difference in mean in the “Extraversion” and “BT_BelongingConcern” variables between students with positive teamwork experiences and negative teamwork experiences. Surprisingly, the result from this test is similar to the ones obtained from Hypothesis 2a. The boot-strapped confidence intervals for “Extraversion” and “BT_Belongingness” do not include 0 in them. Students with past positive experiences score 1.7 points higher in terms of extraversion, meaning that they identify as often speaking in groups, and 1.7 points lesser in terms of BT_BelongingConcern, meaning that they expect to fit right into the course when the course has just started. I believe that past positive teamwork experiences have a similar effect as many teamwork experiences in terms of giving students the confidence to express themselves in groups and reduce their fear towards teamwork projects in the new course. Once again, I believe that Medaille and Usinger’s explanation can be applied here to explain why there is no clear mean difference in the “SpeakUp” variable between the two types of students (243). Thus, I will reject the null hypothesis in favor of the alternate hypothesis that there is indeed a difference in the mean score for “Extraversion” and “BT_BelongingConcern” among the two types of students. However, since the confidence interval does include 0 for the “SpeakUp” variable, there is no clear evidence for me to reject the null hypothesis that there is a difference in mean score for the “SpeakUp” variable among the two types of students.

Average Loss and cut-off points for “SpeakUp” and “BT_BelongingConcern” variable

Variable | Average Loss | Average Cut-off point |
---|---|---|

SpeakUp | 0.4544 | 6.0303 |

BT_BelongingConcern | 0.3823 | 3.0606 |

As mentioned in the introduction, assuming that the lecturers are unable to access the “Extraversion” score directly due to some reasons, then a variable with the least average loss must be chosen as the predictor to predict whether a student is expected to be quiet (Extraversion score is 4 or below) or talkative (Extraversion score is 5 or above). By computing the average loss for both of the variables with the highest correlation coefficient with “Extraversion”, it is observable that “BT_BelongingConcern” has a lesser average loss in predicting whether a student is quiet or not, even though the semantic relatedness with “SpeakUp” is high. Moreover, since the average cut-off point is 3.06, it means that students scoring a point of 4 or more in BT_BelongingConcern have a (1–0.3823)*100% = 61.77% likelihood of being a quiet student since “BT_Belongingness” and “Extraversion” are negatively correlated. However, if “SpeakUp” is used as a predictor, then only a student who scores a point of 7 out of 7 has a (1–0.4543708)*100% = 54.5629% accuracy of being a talkative student. Therefore, the result shows that “BT_BelongingConcern” is a better predictor in predicting whether a student is quiet or not. I believe that this statement makes sense as if a student feels that he or she does not fit into a class, then the student might choose to be quiet in it. (Medaille and Usinger 254)

Lastly, hierarchical clustering is used to plot the distribution of students’ responses.

Model coefficients of each cluster and the probability of students being quiet if categorized into each cluster

Cluster | Model coefficients | Probability |
---|---|---|

1 | (Intercept) xTRUE |
0.1648 |

2 | (Intercept) xTRUE |
0.6917 |

Dendrogram of the hierarchical clustering performed on the data.

Scatterplot of the data points separated into 2 cluster.

From the results, it can be observed that there are correlations between “SpeakUp” and “Extraversion” and “BT_BelongingConcern” and “Extraversion”. This suggests that among the students who responded to the Beginning of Term survey, those students who identify as often holding back ideas or preferences to keep a group happy are also usually the students who would identify as tending to listen more than speak. Moreover, from the survey, I learnt that if students feel that they do not fit into the course even before the term has begun, these students are also likely to listen more than they speak in a team discussion. Nonetheless, even though the aforementioned correlations between variables exist, the correlation coefficient is not large enough to show a strong correlation between the variables. I suggest that in the future, the survey forms can include more questions on students’ background so that future researchers can determine if there are other variables other than “SpeakUp” and “BT_BelongingConcern” that are affecting a student’s “Extraversion” score.

From the second hypothesis, it is noticeable that among the students who responded to the survey, if a student has many teamwork experiences or has past positive teamwork experiences, then the student is expected to be more talkative and more likely to fit into a new course that contains teamwork projects. I believe that past experiences gave students the courage and confidence to express themselves in teams and they no longer feel scared to communicate with others in the team. Owing to this, they expect themselves to fit into the new course more easily than other students who had no experiences. Nonetheless, there is no significant evidence showing that students with more experiences will speak up about their ideas or preferences even if it disrupts the group. I believe that for some students, the thought of interrupting other students’ ideas is rude so they choose to hold back their ideas. This explanation is similar to Medaille and Usinger’s statement of “silence in teams is the result of personal, social, academic, cultural and contextual constraints” (243).

It is important to note several limitations of this study. First, the result of the survey might not be a good representation of the students themselves. This is because sometimes the students might not accurately categorize themselves. For example, a high achieving student might have imposter syndrome and thus feel that he is doing badly and does not deserve to be on the team; or a student thinks that he or she is actually talkative, but in reality, he or she is quiet. Therefore, it is recommended to use an End-of-Term survey that includes both the students’ evaluation of themselves and their peer evaluations on them.

Secondly, although all the respondents are students and all of their courses contain teamwork projects and discussion, the nature of the teamwork projects and discussions might not be the same across different disciplines. For example, a team discussion in humanities class might be interesting or relaxing while a team discussion in engineering class might be boring or stressful. As a result, students from different disciplines might have different attitudes towards the idea of teamwork and collaborative learning. Future research can be more precise by focusing on investigating whether the above results still hold in each discipline (Humanities, Social Sciences, Engineering, etc.)

Thirdly, the results and predictions obtained in this study are only applicable at the beginning of each course term. When the courses begin, there are even more factors throughout the semesters that may change a student’s attitude such as the quality of the lecturers, the course’s syllabus, and the quality of the peer discussions. It is suggested that future research can also focus on investigating how different factors that happen throughout the semester might change a student’s personality (such as from being quiet to talkative).

Lastly, it is to be noted that this survey focuses on students’ data collected during Fall 2020 and the result cannot be used to predict future students’ personality who enrolled in those classes unless a new survey form is filled out by the students and the same analytical method is performed. This result is useful in helping instructors to predict a student’s personality in the beginning of the semester, but it does not guarantee that the prediction is always accurate as the student’s personality changes throughout the semester. Therefore, an instructor should always observe any changes in students throughout the semester and make suitable changes to the group arrangement if necessary.

The author would like to thank Dr. Robin Fowler from the Technical Communication Program for the advice given throughout the writing of this paper, Dr. Cait Holman and the Center for Academic Innovation for data access, and Professor Mark Fredrick for the advice on the usage of statistical packages and formulae.