Documente Academic
Documente Profesional
Documente Cultură
Abstract
This paper presents a controlled experiment demonstrating the utility of forming teams
based on Belbin’s team roles. The overall research focuses on the utility of Belbin’s roles
in improving team performance. This experiment explores Belbin’s Plants, who add
innovation and new ideas to teams. The results of this experiment show that Plants
improve team performance. The specific conclusion is that the mean time-to-completion
to solve programming problems is smaller for teams consisting of only Plant members.
This empirical evidence is useful in both formation and evaluation of teams, which can be
useful to managers of programmers as well as educators.
that software engineers must address in meeting the demands of the current software
based on who can work effectively together. Instead of approaching performance issues
performance of teams of programmers. As DeMarco and Lister [10] say, teams can “gel,”
which means the team displays a synergy such that the overall performance of the team is
and viability [25], which correspond to the dual problems of productivity and turnover
costs [4, 7, 10]. This investigation focuses on the issues of performance. More
completion times of teams that develop software solutions to problems given to the teams
in a controlled environment. The teams for this study are formed using Belbin’s team
roles, which has been used previously to form effective management teams [1].
1
2. Background and Motivation
This research project began with the following informal observation, “Why is it
that a team of very gifted individual programmers doesn’t necessarily make a great
similar to some previous work [8, 23, 30]. For example in the 1980s, the realization that
software could not keep pace with hardware improvements “focus(ed) attention on human
factors in the process of system development as well as the performance of the end user
of computer systems” [19]. We approach the performance issue from this type
to those outside of that field, therefore some terms and concepts need to be provided.
First of all, role theory is useful to examine. As a social psychology term, a role is “the
behavior associated with a particular position in a social system” [11]. Further, a social
role “also carries expectations held by others about the behavior that is appropriate for the
occupant of the position” [11]. Role theory “is an interdisciplinary theory in that its
variables are drawn from studies of culture, society, and personality” [21]. Two
important concepts in role theory are functional requisite and role complement, which are
essentially the foundation of Belbin’s work [1] described below. Functional requisites are
“functions that are necessary for the survival or maintenance of a social system” [3]. A
role complement is a set of roles defined for a given context, for example a programming
2
team. These two concepts combine to form the functional requisites that are embodied by
Belbin’s complement of “team roles.” For this investigation, we are interested in whether
His investigation began with a simple idea that different types of people interact in
settings, Belbin identified eight team roles based on these observed “types”: Chairman,
3
Table 2.1: Brief Descriptions of Belbin Roles [1]
Table 2.1 provides a brief description of all of the Belbin roles in terms of their
team role behaviors and personality characteristics. More complete descriptions can be
One of these roles, the Plant, is particularly interesting for this investigation
4
Section 3. Plant team members advance new approaches and ideas with special attention
but “inclined to disregard practical details or protocols” [1]. The Plants are the
brainchildren who must be nurtured and occasionally drawn back into the real world
because they tend to be absorbed in thought. They are considered one of the intellectual
types in a team. The “Plant” received this label because some innovative individuals
were implanted on some teams in controlled experiments in order to discover how such
software.
One should note that Belbin’s view of roles is different than the traditional,
functional view described above in Section 2.1. He has shown that team members can be
viewed with respect to their participation as part of the team, i.e., in terms of their team
roles. For any particular team member, her or his role in the traditional, functional sense
might be a typist on a programming team, whereas her or his team role might be to focus
team role describes how the individual relates to the team and functions for the team, as
investigating how these team roles affect the performance of software development
teams.
As part of his development of the team roles, Belbin also developed an instrument
5
questionnaire with seven sections. The test produces indicators of an individual’s natural
propensity towards filling each role, similar to psychometrics such as the MBTI [18]. For
each of the seven sections, an individual distributes ten points among eight statements,
based on how strongly he or she feels about each statement. Each of the eight statements
corresponds to one of Belbin’s roles. It should be noted that if an individual takes the test
and distributes the points as evenly as possible, each role would have the average of
forming teams based on these roles, a member would need at least an 11 for a given role,
1
8.75 = (7 sections * 10 points/section) / 8 roles
6
Table 2.2: Sample Section of Questionnaire for Belbin Roles [1]
____1. I am not at ease unless meetings are well structured and controlled and generally
well conducted.
____2. I am inclined to be too generous towards others who have a valid viewpoint that
has not been given proper airing.
____3. I have a tendency to talk too much once the group gets on to new ideas.
____6. I find it difficult to lead from the front, perhaps because I am over-responsive to
group atmosphere.
____7. I am apt to get caught up in ideas that occur to me and so lose track of what is
happening.
____8. My colleagues tend to see me as worrying unnecessarily over detail and the
possibility that things may go wrong.
computer science majors in a senior-year software engineering class. These students are
typically among the top of their class and highly motivated. Most of them had had part-
7
Some previous studies have examined how to form programming teams
methods to set up programming teams for software engineering classes [12]. She
presents a heuristic for establishing teams that is based on amount of free time, schedule
conflicts, and grade point average. Scott and Cross also discuss issues in setting up
student programming teams in an effort to make them relatively equivalent [22]. Some of
the issues that they present are academic performance (grades), team and project size, and
they only consider psychological issues such as a team needing an introvert and an
extrovert because their class requires written and oral presentations. “Clearly,” they
conclude, “an Extrovert will be more comfortable with an oral presentation, while an
concern that such research does not directly apply to industry programmers. In addition
to the high quality of the participants of this work, Holt, et al. provides support for using
[5, 15, 17, 23, 24, 26, 27, 28, 29, 31, 32]. One investigation by Thomsett is of particular
8
interest for this experiment since it applies Belbin’s roles to software developers [26].
His study presents a qualitative analysis of software teams in Australia using various
measurement instruments, including the MBTI and the Belbin Self-Perception Inventory.
reported by the senior management of the computing group” [26]. Also, Thomsett
describes some intuitive reasons for the uncommon data distributions of the Belbin roles
and MBTI types: a ‘cloning’ effect. This effect means that managers hire employees
who are similar to themselves because ‘clearly’ they themselves (the managers) are good
employees, and therefore, they should hire other people like themselves. Further, he
observes that “at best, because of the relative lack of Team Worker … and effective
Chairpersons …, computing teams and managers generally lack the required interpersonal
skills,” [26] thus making the teams less effective than they could be.
qualitative and the conclusions are not substantiated quantitatively at all. Therefore,
Thomsett presents an interesting study with some initial findings that need further
investigation. He presents general information that appears very useful but has only
be performed on objective data in order to support his ideas. Such quantitative analysis
9
3. A Controlled Experiment on Innovation on Teams
This paper describes a single experiment that is part of a larger research
investigation. The fundamental issue that the overall research addresses is how to
including some taken from role theory and personality characteristics. This includes
analyzing extant teams and forming new teams. This phase of the project consists of a
Curtis has shown that the initial phase of software development consists of
developing potential solutions to a problem [9]. The innovation necessary in this phase is
obvious, and the natural propensity to innovate is the key characteristic of a Plant.
Therefore, members who score significantly high on the Self-perception Inventory for the
needs to be used to provide a “measure of success.” The performance measure used here
is the time-to-completion for correct solutions. All of the teams were given the same
software problem to solve, and teams that completed a correct solution the quickest were
considered better. The teams were formed such that some were anticipated to perform
well, and other teams were anticipated to perform poorly. After all of the teams finish,
10
the completion-time data was analyzed by comparing the mean completion time of all of
the “good” teams to the mean completion time of all of the “bad” teams.
success, but this measure was used because it is objective, quantitative, and easy to
collect. A solution was not accepted until it was correct, so that each team submitted
solutions until they got it right, and no subjective quality measure was gathered. Thus,
this experiment demonstrates that teams containing the specified role(s) perform better,
i.e. faster, than teams that do not possess that role(s). Specific details on conducting the
supporting Thomsett’s results [26]. Several hypotheses were proposed to focus the
research on issues that could be directly tested. The hypothesis that was ultimately
selected investigates the importance of Plants on a team. Specifically, the null and
must be defined within the context of this work. Teams are the three-member unit under
investigation. Each team is classified as belonging to one of three groups: All Plants,
11
Some Plants, or No Plants. These group names indicate the number of Plants on each
individual team. Each team in the “All Plants” group has three Plants, whereas each
team in the “No Plants” group has none. Each team in the “Some Plants” group has one
Using a table similar to Table 3.1 but including all of the teams, the participants
were formed into teams, and the teams were identified as being in a particular group.
Two teams were formed for the successful group, i.e. the teams anticipated to complete
the programming problems quickest. Three teams were formed for each of the other two
groups, with the anticipation that they would perform worse than the successful teams (in
To mitigate any other factors that could influence the teams’ performance, the
other Belbin roles were made as equivalent as possible across the teams, and the team
members’ scholastic grade point average (GPA) was used as a blocking factor. Such
tactics help distinguish the role under scrutiny as the reason the groups perform
differently.
Table 3.1 shows the information used to accomplish the team formation. Rows in
the table contain team member data. For example, “Team X” consists of three team
members, whose data are in the three rows after “Team X.” In each row, the first column
designates the team name. The second column shows member numbers. The subsequent
columns represent the members’ data for the Belbin roles: Chairman (CH), Shaper (SH),
12
(CW), Team Worker (TW), and Completer-Finisher (CF). The final column shows the
members’ GPA, which is used to make the teams equivalent, mitigating the chances that
Team X
1 c1 - - - - W2 - F9 2.9
2 - s2 p0 - m0 - - - 3.1
3 - s0 p7 - m5 w0 - - 3.6
The data in each cell in the table represents that member's score on the Belbin
Self-Perception Inventory. Cells containing a dash (-) indicate that the score was not
significantly high enough to take into account when forming the teams. As described
and a 20 or higher is very significant. Cells not containing a dash contain a letter and a
digit that indicates the team members’ score on the Belbin Self-Perception Inventory.
The letter represents the role, which is also shown by the column headers abbreviations,
and the digit indicates the score on the Belbin test. Scores in the range 10 to 19 are
indicated with a lower case letter for the role, e.g. c1 indicates a Chairman score of 11.
Scores of 20 and above, which are very high scores, are indicated with a capital letter, e.g.
W2 indicates a Company Worker score of 22. This scheme was used because it removes
extraneous information, making comprehensible the large amount of data that was used to
13
To determine a member’s potential roles even after discarding “insignificant”
1. An individual fills more than one role by having significant scores for multiple
roles, typically two or three roles.
2. Although an individual might have a number that appears to make him or her fill
that role, he or she may not fill that role because he or she has a stronger tendency
to fill other roles. For example, in Table 3.1, member 1 probably does not fill the
Chairman role because the W2 and F9 values are much higher.
3. Other team members may keep an individual from filling a role in the case that the
other member is stronger in the role. This is particularly true if the member has
roles that are significantly stronger. For example, member 3 in Table 3.1 would
not fill the Shaper role because of his or her other stronger roles (p7, m5) as well
as member 2 being such a strong Shaper (s8).
4. Because the scores are relative, not absolute, sometimes a score of ten (10) or
eleven (11) is significant, sometimes not. If a member’s highest scores are 12 in
two roles and 10 in another role, then the 10 would be considered significant,
because the individual has a tendency to assign a few points to all of the
statements in the Belbin Self-Perception Inventory. Member 1 in Table 3.1 is a
good example of when a low score of 11 is not significant; Member 2 is a good
example of when a low score of 10 might be considered significant.
One basic assumption of this research is that some roles can conflict within a
team, such that a team is “better” with only one of that role type on the team. This
appears to be true for certain roles, such as Shaper, where conflicts can occur [13]. As
demonstrated below, other roles appear not to demonstrate this effect; teams with all
Plants perform the best. It is very important to note that the results of this test indicate
individuals who have a natural propensity to these roles. The teams were not formed
based on previous experience or any other factors, and no member was designated as the
leader for each team. Also, the participants had no knowledge of Belbin roles nor of
other qualities that members might possess unless they discussed it amongst themselves.
14
3.2 Conducting the Experiment
The teams and groups were formed prior to the commencement of the
experiments (see Table 3.2), and the participants met in a laboratory where there were
nine computers used in the experiment. The twenty-four (24) student participants had
been formed into eight teams, and each team was limited to one computer apiece. One
computer was used by the experimenter to test and accept problem solutions. Each
participant was given a copy of the problem to be solved that day. Each problem was
intended to be solved in a little over an hour, although some teams ended up requiring up
to two hours. The teams could use any editors, C or C++ compilers, and utilities they
wished. All of the systems were equivalently equipped. Teams emailed their solutions to
the experimenter as soon as they felt that they had a correct solution, and the
experimenter tested the solutions with his own acceptance data, which was not provided
to the participants. A completion time for each team was only recorded once a submitted
solution tested correctly with the acceptance data. Although the problem statements
included some example test data, the participants were informed that they should create
test data themselves in order to have their solution pass all of the acceptance tests.
15
Group Team Member CH SH PL RI ME CW TW CF GPA
3 B
No 1 - - - - m1 w1 t8 - 3.0
Plants 2 - - - - M0 W5 T0 - 4.0
3 c0 s2 - r2 m2 - - - 2.6
F
1 - s1 - - m4 - - - 3.0
2 - - - - - w9 t5 - 3.3
3 c3 - - - - W2 - f9 3.1
G
1 - s2 - - - w2 - - 2.7
2 - - - - - w4 - f1 3.4
3 c0 - - - m4 w1 - - 3.4
2 C
Some 1 - s2 - r2 - w0 - - 2.8
Plants 2 - s3 p7 - m3 w0 - - 2.9
3 c4 - - - - w5 t4 f2 3.5
D
1 - s0 p3 - - w4 - - 2.9
2 - s4 p2 - m0 - - - 3.3
3 - s9 - - m3 w3 - - 3.0
E
1 c0 s4 - - - w1 - f2 3.6
2 - - P0 - - w1 - - 3.5
3 - s3 - - m6 w7 - f0 3.0
The experiment consisted of four problems that were solved in four separate lab
sessions. During each of the four sessions, the teams were given one problem to
complete during that lab session. The means of all four problems for each group was
calculated and compared in the following analysis. Having the teams work together on
four different problems acted like replication, strengthening the results by having more
The use of one computer per team is an important feature of the experiment. The
intent was to force individuals to work as a team, instead of examining individuals. If the
16
members were allowed to separate and work independently, then the experiment would
simply measure which member on each team could solve the problem first.
teams were formed into groups so that some teams had all members who were Plants
(Group 1, All Plants), some teams had one or two Plant members (Group 2, Some Plants),
and some teams had no members who were Plants (Group 3, No Plants). Teams A and H
form Group 1 (All Plants); Teams C, D, and E form Group 2 (Some Plants); and Teams
The initial hope was that these three groups would all be statistically different, but
as shown below in the statistical analysis, only two of the groups could be differentiated.
The mean-time-to-completion for the “All Plant” group is statistically different from the
“Some Plants” group, 54.13 and 100.75 minutes, respectively. Further, since the mean of
the completion times for the “All Plant” group is smaller, we can state that the “All Plant”
group performs better. In fact, “All Plant” teams require only 53.73% of the time to
17
120 100.75
100 86.25
80
Minutes
54.13
60
40
20
0
Some Plants Group1 No Plants Group3 All Plants Group2
More formally, this experiment on innovation on a team has a null hypothesis and
An ANOVA using Proc GLM in SAS [16] shows that the groups are statistically
different by F-test (p=0.0412). The means are 100.75, 86.25, and 54.13 for group 2, 3,
and 1 respectively (See Figure 3.1) with standard deviations of 55.81, 56.77, and 26.43
respectively. The R-Squared is 0.5024. This allows the null hypothesis to be rejected,
indicating that the groups are different. Further, Duncan’s New Multiple Range Test [20]
indicates differences among the group means. For two of the groups, “Some Plants” and
“All Plants,” the data shows a significant difference between the groups.
This raises questions about why the “All Plants” group might appear to perform
better than the “Some Plants” group but not statistically better than teams with “No
Plants.” The easiest aspect to question is the size of the sample, in other words the
18
number of teams in the study. We believe that a larger sample size this would remedy
this, with the “No Plants” joining the “Some Plants” group. With a mean of 86.25, the
“No Plants” group is certainly closer to the “Some Plants” group (100.75) than to the “All
Plants” group (54.13). This idea has some statistical support because, when just the data
for the “All Plant” and “No Plants” are analyzed, the groups are different by F-test
(p=0.1011), but this p-value is not statistically different. Thus, although there is a
were analyzed. The difficulty is that finding large numbers of volunteers to participate in
such experiments is difficult. On the “All Plants” teams, all of the members are
innovators; they have a tendency not only to come up with many different ideas but also
to understand others’ ideas more easily. This is an obvious shortcoming of some of the
who took notes both during and immediately after each lab session, include positive and
negative perceptions. Notes about the teams predicted to perform well include: “No
negative aspects to this team; everyone worked well together.” “The coordination within
the group worked very well; the work was divided up very well. (Person 1) figured out
implementation. (Person 2) would then type it in incredibly quickly, while (Persons 1 and
3) looked for bugs in the algorithm and made sure that (Person 2) wasn’t making typos.”
All three members of this team identify team coordination and “functional roles” as the
19
key to their success. Some comments from, and about, the “unsuccessful” teams include:
“Motivation wasn’t all that … (sic) present;” the team “couldn’t think of many good
solutions.” The “quiet/reserved team member didn’t participate much,” which is very
unfortunate since this person is a very strong Plant. “Sometimes good ideas were ignored
because they ‘almost had it’ even though the other solution would have been better;”
“We thought in different ways. This made coming up with a single, well-understood, and
good solution next to impossible.” Further comments include “no motivation, no division
of labor” and “some ego problems caused good decisions and ideas from all members to
roles. It shows that Belbin’s test can be used to identify characteristics of team members
Obviously, the formal conclusion of this study is the alternate hypothesis stated
above: Teams composed of all Plant team members perform better than teams with some
development team; however, controlled experiments are necessary to prove the fact. As
this research continues, more statements that are definitive will be verified. This research
extant teams to identify deficiencies. The intent for evaluating teams would be to rectify
deficiencies and to take full advantage of the positive aspects of the team. It should be
20
noted that this was a limited controlled study. Results may be different for long term
projects simply because human interactions over an extended period tend to be different
from a controlled experiment, other factors can and indeed do affect the performance of
teams. To say otherwise would be ridiculous and misleading. It should also be obvious
that not all development teams can be composed of all Plants, because there are only a
limited number of Plants. Further, teams composed of all Plants are not guaranteed to
perform great because other factors affect team performance; statistically speaking,
teams composed of all Plant members are more likely to perform better. The significance
of this work is that the existence of Plant roles on a team does in fact affect team
performance and can be used as part of the formation of teams. This result should be
determine which of the other roles are important to software development teams. Some
conclusions about the other Belbin roles have already been reached [13]. Some of the
roles intuitively appear to be less significant than others, for software development teams.
Some of the roles may in fact have a negative impact on a team; some of them may not
have much of an impact because the role is extremely common in the computer science
field. The basic concept that Belbin developed, his complement of roles for management
21
teams, needs to be empirically tested for the computer science domain. Possibly, a
information directly to hire “computer science” types, the literature indicates that team
concepts of team roles, with other important factors such as the need to have diverse
teams. The above analysis demonstrates the utility of Belbin’s team roles in forming
extension. Further, if teams are properly formed using roles, then the team members are
happier and more likely to remain with an employer, thus increasing the viability of the
team. All in all, applying Belbin’s role concepts to software development teams can be
used to improve both the performance and the viability of teams and, therefore, team
effectiveness.
22
Bibliography
[1] Belbin, R M, Management Teams, John Wiley & Sons, New York, 1981.
[3] Biddle, B J, Role Theory: Expectations, Identities, and Behaviors, Academic Press,
New York, 1979.
23
[14] Holt, R W, Boehm-Davis, D A, and Schultz, A C, Mental Representations of
Programs for Student and Professional Programmers, in: Empirical Studies of
Programmers: Second Workshop, G M Olson, S Sheppard, and E Soloway, eds.,
Washington, DC, December 7-8, 1987 (Ablex Publishing Corporation, Norwood,
New Jersey, 1987) 33-46.
[15] Jeffery, D R, The Relationship Between Team Size, Experience, and Attitudes and
Software Development Productivity, in: Proceedings of COMPSAC87, Tokyo,
Japan, October 7-9, 1987, (IEEE Computer Society Press, New York, 1987) 2-8.
[16] Littell, R C, Freund, R J, and Spector P C, SAS System for Linear Models, 3rd
Edition, SAS Institute, Inc., Cary, North Carolina, 1991.
[17] Mills, H D, Software Productivity, Little, Brown and Company, Boston, 1983.
[18] Myers, I B, Introduction to Type, Consulting Psychologists Press, Inc., Palo Alto,
California, 1987.
[20] Ott, R L, An Introduction to Statistical Methods and Data Analysis, Duxbury Press,
Belmont, California, 1993.
[21] Sarbin, T R, Role Theory, in: Handbook of Social Psychology, Gardner Lindzey,
ed., (Addison-Wesley Publishing Company, Reading, Massachusetts, 1954) 223-
258.
[22] Scott, T J and Cross, J H II, Team Selection Methods For Student Programming
Teams, in: Software Engineering Education: Proceedings of 8th SEI CSEE
Conference, New Orleans, Louisiana, March 29 - April 1, 1995 (Springer-Verlag,
New York, 1995) 295-303.
24
[27] Trower, J K and Straub, D W, Jr., Improving the Performance of Technologists and
Users on Interdisciplinary Teams: An Analysis of Information Systems Project
Teams, Computer Personnel (November, 1991) 62-72.
[28] von Mayrhauser, A, Task Analysis and the Selection of Software Development
Team Structures, in: Proceedings of COMPSAC84, Chicago, Illinois, November 7-
9, 1984, (IEEE Computer Society Press, New York, 1984) 120-126.
25