Application of Social Network Theory to Software

Development: The problem of task allocation

Chintan Amrit

Erasmus Research Institute of Management

F1-1, Postbus 1738, RSM Erasmus University

3000 DR, Rotterdam

Abstract. To sy

stematize software development, many process models have

been proposed over the years. These models focus on the sequence of steps

used by developers to create reliable software. Though these process- models

have helped companies to gain certification and attain global standards, they

don’t take into account interpersonal interactions and various other social as-

pects of software development organizations.

In this paper we tackle one crucial part of the Coordination problem in Soft-

ware Development, namely

the problem of task assignment in a team. We pro-

pose a methodology to test a hypothesis based on how social networks can be

used to improve coordination in Software Industry.

In a pilot case study based on 4 teams of Masters Student working in a globally

distributed environment (Holland

and India), the social network structures

along with the task distribution in each of the teams were analyzed.

In each case we observed patterns, which could be used to test many hypothe-

ses on team coordination and task allocation between them.

1 Introduction

“Organizations which design systems are constrained to produce designs which are

copies of the communication structures of these organizations”.

Though Conway said this back in 1968, little has been done to align product

ar-

chitecture to the company communication structure, or the Social Network as we call

it nowadays.

Software has been plagued by many problems and there seems to be a great chasm

bet

ween the theoretical models and the actual implementation in the industry. In a

recent article by Laplante and Neill (2004) found that as many as 1/3 of their survey

respondents followed the waterfall model for Software Development. Though this

model was introduced in 1970 when computer systems were archaic and user needs

were very basic, the fact that it is still in use by a large section of the software devel-

opment industry shows that many of the best practices in software development are

being ignored by the industry (Laplante & Neill, 2004).

Amrit C. (2005).

Application of Social Network Theory to Software Development: The problem of task allocation.

In Proceedings of the 2nd International Workshop on Computer Supported Activity Coordination, pages 3-17

DOI: 10.5220/0002576500030017

 SciTePress

There is also a huge difference between the design and implementation of soft-

ware and as mentioned in one report (The Standish Group, 2003), on an average only

52% of required features and functions make it to the released product.

While there is no single cause for the problems in Software Development, a major

factor is the problem of coordinating activities while developing large software sys-

tems (Kraut & Streeter, 1995). Kraut and Streeter (1995) mention scale of software

projects, inherent unpredictability of software specifications and tasks as well as the

interdependence of software components as some of the factors that lead to the neces-

sity of efficient co-ordination between the different work groups involved in the de-

velopment process.

In this paper we tackle one crucial part of the coordination in Software Develop-

ment, namely the problem of task assignment among team members of a software

development team. The aim of this paper is to come up with insights on the method-

ology by which one can use social network analysis to improve the coordination in

the Software Development Process of an IT company.

In this paper we have come up with a hypothesis and a couple of propositions

based on previous work done on coordination in teams. We conducted a pilot survey

on teams of students who worked on software design tasks. We then observed the

social network of the teams concerned, along with the distribution of the software

design tasks among the team members. We then used the propositions and hypothesis

to predict the performance of the team members and then compared it with the actual

performance of the teams; thereby demonstrating the methodology by which more

such hypothesis and propositions can be verified.

2 Network Theory and Groups

2.1 Social Network and Metrics

The problem of coordination can be better explained by first illustrating the concepts

of social networks and centrality measures. A social network consists of a set of ac-

tors (“nodes”) and the relations (“ties” and “edges”) between these actors

(Wasserman & Faust, 1994).

Researchers have developed a variety of metrics for quantifying the differences in

network structure. Among the many frequently used metrics at the actor level are

degree centrality (the extent to which actors send or receive ties) and betweeness

centrality (the extent to which actors have ties with others who are not directly con-

nected). Metrics used to describe networks include: Density (the ratio of the pairs of

nodes that are mutually reachable to the total number of pairs of nodes) and Centrali-

zation (Difference between the centrality scores of the most central actor and those of

all other actors in a network is calculated, and used to form the ratio of the actual sum

of the differences to the maximum sum of the differences).

How is a group defined, from the network perspective? The construct of a group,

when used in the social network literature has had two primary meanings: (a) a struc-

tural feature of a network, or (b) an exogenously determined or imposed category.

According to the first meaning, groups (cliques, a maximally complete subgraph) are

subsets of fully connected, or almost fully connected, nodes within some population

(Katz et al. 2004).

2.2 Internal Networks Ties

The problem of task allocation among team members is closely related to the network

structures formed between them. Several researchers have asked the question “What

is the optimal network for group performance?” Many have broadened the scope of

investigation by moving from the laboratory to the field.

Sparrowe, Liden, Wayne and Kraimer (2001) demonstrated a relation between

network structure and both in-role and extra-role performance in a field setting. They

replicated earlier findings; with a complex task, and found that groups with decentral-

ized communication patterns perform better than groups with centralized communica-

tion patterns. Cummings and Cross (2003) also found that groups with decentralized

communication patterns outperformed more centralized groups.

Other researchers have focussed on the number rather than the pattern of com-

munication links among group members. Baldwin, Bedell and Johnson (1997) and

Reagans and Zuckerman (2001) established in field studies that groups with more ties

performed better than groups with fewer ties.

2.3 External Network Ties

External networks deal with ties to particular strategic others, as well as the overall

structure of those ties (as measured in quantity and pattern), where the ties may be to

other groups or to the environment (Katz et al. 2004). Understanding the context in

which the group is embedded as well as its relationship with key players in its envi-

ronment improves our understanding of how the group functions.

Ancona and Caldwell (1992) show that the pattern of external activity is a better

predictor of group performance than simply the frequency of communication. They

also try to understand the types of external activities that are needed for team effec-

tiveness.

Baldwin et al. (1997) find no relationship between a team’s external ties and its

performance. They suggest that due to the nature of the group’s task the configuration

of their internal ties is more important than their external ties. In other words the

group’s need for external resources is defined by the nature of the task.

Reagans et al. (2004) compare two approaches to team formation, one based on

the member’s demographic characteristics and the other based on the members’ social

networks. They hypothesize that demographic diversity not only decreases the team’s

internal density but also increases the team’s amount of range in its external network,

and that both of these variables have a positive effect on team performance.

3 Team Formations and Task Allocation

Teams are the basic building block for many contemporary business organizations.

We focus on how we can improve coordination in software development projects

using the concepts of coordination between and among teams keeping task assign-

ment as a moderating variable. Coordination refers to team-situated interactions

aimed at managing resources and expertise dependencies (Faraj and Sproull 1995).

Research on software development teams has found that team performance is linked

with the effectiveness of teamwork coordination (Kraut and Streeter 1995).

Faraj and Sproull (1995) take two perspectives on coordination: administrative

coordination and expertise coordination. They claim that administrative coordination

(management of tangible and economic resource dependencies) is good for simple

routine tasks, while for complex non-routine intellectual tasks, expertise coordination

(the management of knowledge and skill dependencies) become more important.

Through expertise coordination the team can recognize and access expertise when it’s

needed.

Grinter and Herbsleb (1999) suggest the chief motivation for the assignment of

tasks (involving a search for experts) in R&D projects to be:

Functional Area: an expertise of distant systems,

Product Structure: an understanding of the internals of components built at remote

sites,

Process: knowledge about what happens during other processes,

Customisation: knowledge of core or the knowledge of how the core is custom-

ized depending on which site is involved.

Though their analysis models and solutions seem good for R&D projects, they

have not used the concepts of social networks to improve their model.

Stewart and Barrick (2000) build on organization-level findings and show that

differences in how responsibilities are apportioned and coordinated correspond to

variance in performance at the team level. They also show that the effect of these

social elements is moderated by technical demands (tasks), consistent with socio-

technical systems theory.

Hansen et al. (2001) distinguish between exploration and exploitation tasks

among teams. They show that teams engaged in exploratory tasks complete their

projects more quickly if they have a social network structure composed of many

strong external ties that are non-redundant. In contrast, teams pursuing tasks that

exploit existing expertise take longer to complete if they have this type of social net-

work structure, mainly because external ties have to be maintained but are not much

needed for the task.

Sparrowe et al. (2001) hypothesize that centrality in a work group’s advice net-

work will be positively related to an individual’s job performance. Where centrality

in the advice network reflects an individual’s involvement in exchanging assistance

with co-workers and engaging in mutual problem solving. An individual who is cen-

tral in the advice network is, over time, able to accumulate knowledge about task-

related problems and workable solutions (Baldwin et al., 1997). While the central

individual develops problem solving capability and serves as a valued resource for

future exchanges with co-workers, those individuals who are in peripheral positions

in the advice network find it difficult to develop expertise and competencies for high

levels of performance (Sparrowe et al., 2001). Hence, Sparrowe et al. (2001) hy-

pothesize that centralization in a work group’s advice network is negatively related to

group performance.

Yang and Tang (2004) try to analyse the relation between team structure and ISD

performance using a social network approach. They show how the structural proper-

ties of the work groups fluctuate during the various phases of Software Development,

and how group cohesion and centrality are related to the final ISD performance.

Though Yang and Tang (2004) do show how social research methods can be used to

tackle “group process” factors, they do not deal with task allocation nor do they illus-

trate how one can solve the problem of task allocation among team members.

Though these studies indicate how coordination can be improved in industries,

there has been not much work done in the analysing the use of social networks in

improving coordination through better task allocation in the Software Development

industry.

Sparrowe et al. (2001) hypothesize that centralization in a work group’s advice

network is negatively related to group performance. But a group with a high centrali-

zation in the advice network can still perform well if the central individual contributes

in all the phases of development. Adding the task component to the hypothesis 1 by

Sparrowe et al. (2001) we come up with the following proposition:

Proposition 1: When the centralization in the advice network is high then the team

performance increases if the central person(s) contributes in all the different tasks.

Proposition 2: When the density in advice networks is high then the team perform-

ance increases when the tasks are evenly distributed among the team members.

Hypothesis 1: Performance of a team is positively related to the density of the task

network, when the density in the advice network is high.

4 Empirical Test

4.1 Project Questionnaire

A pilot survey was conducted on 4 globally distributed teams of Masters Students

consisting of 8, 8, 7, 7 students respectively. Approximately half of the members of

each team consisted of Dutch students located in a Dutch university and other half

were Indian from a university located in India. The students were asked to select a

topic for a design-based project, and complete four design tasks involving the crea-

tion of a vision document, activity diagram, use case and class diagram for the se-

lected project topic. The data was collected with the help of a questionnaire, in which

among other questions we asked:

Rate your contribution (relative to the average team member) in creating the Vi-

sion Document?

Rate your contribution (relative to the average team member) in creating the activ-

ity diagram?

Rate your contribution (relative to the average team member) in creating the Use

Case diagram?

Rate your contribution (relative to the average team member) in creating the class

diagram?

Mark your team members from whom you regularly sought information and ad-

vice to help in your project work.

Mark your team members, whose advice you did not seek, during the course of

the project.

Mark your team members who in your opinion are very dependable in executing a

crucial part of the project.

The first 4 questions had options out of a scale of 5, while the questions 5 to 7 had

the list of all the team members to choose from (multiple members could be checked

for each answer). From the answers to questions 1 through 4 the relative contributions

to the particular task were obtained. This helped in drawing the 2-mode task network.

Where we can see the team members assigned to the tasks (making the Use case dia-

gram, activity diagram etc.) with the links having a particular weight corresponding

to the amount of effort each team member has put in performing the task. The contri-

butions ranged from 1(no relative contribution) to 5 (full contribution). The contribu-

tion was based on team member perception, so it was possible for all the members of

the team to think that they had done the task themselves (all of them to fill 5). From

the answer to questions 5 and 6 the advice network of the group members was ob-

tained. The answer to question 6 confirmed the network obtained from the answer to

question 5. From the answer to question 7 the discussion network among the mem-

bers of the group was obtained.

4.2 Measures

The Network measures were calculated using the tools available in UCINET. The in-

degree centrality scores were computed for each individual (Borgatti, Everett & Free-

man, 1992). Where in-degree centrality (the number of ties received by a vertex) is a

form of degree centrality that counts only those relations with a focal individual as

reported by other group members, and it does not suffer from the limitations of self-

reports, as does out-degree centrality (Sparrowe et al. 2001).

4.3 Network density

In binary network data, density is the proportion of actual nominations among the

total possible number of nominations (Wasserman &Faust, 1994). This was computed

by using the density function of UCINET for networks, using total number of ties

present divided by the total number of all possible ties.

4.4 Network Centralization

Network centralization is the sum of the observed differences in individual centrality

scores (computed by finding the differences between the largest individual centrality

score and the scores of all the other individuals in the network) divided by the maxi-

mum possible sum of differences (Wasserman & Faust, 1994). This network centrali-

zation was computed using Freeman’s (1979) definition in the UCINET IV software

package (Borgatti, Everett & Freeman, 1992).

4.5 Task Density

The task network is a 2-mode network (Borgatti & Everett, 1997). In order to find the

density the weighted task network was first dichotomised using the standard di-

chotomise routine in UCINET IV software package (Borgatti, Everett & Freeman,

1992). The cut-off value was considered to be 3, as the contribution of less than 3 on

5 was considered negligible (also because 3 is the median on a scale of 1 to 5). So the

dichotomization rule was as follows:

y(i, j) = 1 if x(i , j) >= 3, and 0 otherwise.

Then the density of this dichotomised 2-mode matrix was calculated using the density

routine of UCINET IV (Borgatti, Everett & Freeman, 1992).

4.6 Team Performance

The final performance of the team was rated according to the following metrics:

Time taken for project completion

Documentation and its revision history

Quality of deliverables

Relevance of alternative solutions suggested

The final grade scored by the individual teams was considered (out of 10).

Table 1. Statistics of some of the network measures

Team

Centraliza-

tion of Ad-

vice Network

Density

of Advice

Network

Density

of task

network

Team 1 52.38% 0.3929 0.7188

Team 2 57.14% 0.3036 0.8125

Team 3 53.33% 0.5000 0.9643

Team 4 20.00% 0.4048 0.8571

5 Results

Though we cannot really test the validity of the hypothesis using the statistical data

we have, we can see how this statistical data can be used to predict the actual per-

formance of the teams. On the basis of the results in table 1 we can estimate the per-

formance of the teams according to our propositions. We expect the team with lower

centralization to perform better than the team with higher centralization (Sparrowe et

al, 2001). Thus, according to the centralization of the Advice network we expect the

teams to have performed in the order: Team 4 > Team 1> Team 3> Team 2. We ex-

pect a team with a higher density in the Advice network to perform better than a team

with a lower density (Sparrowe et al., 2001). Thus, according to the density of the

advice network we expect the teams to have performed in the order: Team 3 > Team

4 > Team 1 > Team 2. Finally according to our hypothesis of the density of the task

network we expect then that a team with the higher density in the task network to

perform better than a team with a lower density, when their density in the advice

network is high. Thus according to our hypothesis we expect the teams to be ranked

as: Team 3 > Team 4 > Team 1 > Team 2, as Team 1 has a higher density in the ad-

vice network as compared to Team 2.

We obtained the final ranking by evaluating the quality of the deliverables, which

was: Team 3 > Team 1 > Team 4 > Team 2 (table 2).

Table 2. Evaluation of the Teams

TEAM Grades

Evalua-

tion of

Quality

Overall

Rank

Team 1 7.5 10 2

Team 2 7 8 4

Team 3 7.5 15 1

Team 4 7 10 3

6 Discussion

The results do not entirely support our hypothesis, as the survey was only a pilot

survey conducted on 30 students. On the other hand, this study shows that we can use

social network analysis in order to test hypothesis and propositions related to team

performance in a Software Development project. The results suggest that our hy-

pothesis is not a very bad predictor when it comes to finding the team that performed

the best, or the worst. Also according to the results of this pilot survey we see that the

density of the task network is almost an equal predictor of performance as the density

of the advice network.

The Propositions (1 and 2) suggest that tasks in a team must be assigned in accor-

dance to the centrality and density of the advice networks of the team. When the

centrality of the advice network is large, then the team wouldn’t fare well (Sparrowe

et al., 2001) unless the central individual contributes in every aspect of the develop-

ment process, as an individual who is central in the advice network is, over time, able

to accumulate knowledge about task-related problems and workable solutions (Bald-

win et al., 1997). Then again, when the density of a team’s advice network is large

then the team performs well, when the tasks are distributed more evenly among the

team members.

From figures I to VIII (see appendix at the end), we can see that when the density

in the advice network is high (Fig. V), along with the density of the Task Network

(Fig. VI), the performance of the team is very good (table 2). While when the density

of the advice network is low (Fig III) and the density of the corresponding task net-

work is also low (Fig IV) then the performance of the team suffers (table 2).

In all the figures I to VIII we see that the advice networks consists of a structural

hole (Monge & Contractor, 2003) on either side of which represents the Indian com-

ponent (a connected subgraph) and the Dutch component (another connected sub-

graph) of the globally distributed team. The structure of these advice networks further

shows the presence of gatekeepers, who are persons involved in communicating with

their global (Dutch or Indian as the case maybe) counterparts. In the case when one of

the components (Indian or Dutch) in the advice networks is not well connected we

see that the performance of the groups suffers. This is evident in figure III, where the

Indian component is sparsely connected. This can be further be used to show the

differences in predicting the performance of Team 1 and Team 4. Though, the differ-

ences can be due to several factors, as is common in most Software Development

Projects (Kraut & Streeter, 1995) from a network perspective we can say that the

visible difference in their advice networks might have contributed to the difference in

performance. We find that the advice network of the Indian component in Team 1

(figure I) is not so well connected as the advice network of the Indian component in

Team 4 (figure VII). From this difference in structure one can say that the lack of a

well-connected advice network of a component of a globally distributed team might

have had an impact on the performance of the team, causing Team 1 to perform worse

than Team 4.

7 Limitations

The primary limitation is that the empirical data is weak, as the sample size is too

small for any kind of statistical analysis. The idea behind this paper is to illustrate

how this methodology can be used for future analysis on larger samples.

The students in the Dutch University were without much industrial experience

while the students in the Indian University had some industrial experience. Further,

the teams were composed of relative equals and had limited existence, making them

unlike most real world teams. In future research, more case studies/surveys need to be

conducted on people working in the software industry.

The tasks, which were a part of their course project, were at best small tasks at the

design stage of software development. Also, the number of modules was limited to

four, while in a normal software development project the number of modules runs

into thousands and corresponding number of tasks per person can be much higher.

So, in future such a study can be conducted in an IT company with more demanding

project requirements.

Although sociometric techniques were used with small groups in experimental re-

search conducted in the 1950s (Shaw, 1964), it should be noted that contemporary

research typically studies networks larger in size than the groups studied here.

8 Conclusion

This study adds to the growing body of theory in software development methodology,

using social networks. Though there have been many papers written on the develop-

ment of architecture in Software Engineering, there have been few studies on improv-

ing the development of Software using social networks with better task allocation

strategies. This study adds the task component in judging group performance through

advice networks, which was dealt with in the paper by Sparrowe et al. (2001) and

shows a method of verification of such hypothesis.

In future research, more surveys/case studies can be conducted on larger groups in

the Software Industry to test the hypothesis presented as well as other related hy-

potheses on task allocation and group structure. Further, such a test can be conducted

in a longitudinal manner at different stages of the software development project. Add-

ing the time component can throw more light on the way social networks change and

develop during the course of a project. It could also be used to understand the differ-

ent task allocation strategies needed to make a project successful.

Software engineering can only advance towards being an engineering discipline

by moving away from its current dependency upon advocacy and analysis, and by

employing more systematic empirically-based approaches to developing an under-

standing of what works, why and under what conditions. This paper is a first attempt

towards such an empirically based approach. It’s an attempt at opening the black box

of the complex development process, which goes into a software development pro-

ject.

Acknowledgements

Jos Van Hillegersberg, RSM Erasmus University, my supervisor, without whose

guiance the survey wouldn’t have been possible, Sridhar V, MDI, Gurgaon, who

helped in initiating the survey,Dhruv Nath, MDI Gurgaon, who contributed in con-

ducting the survey

References

1. Ancona, D.G., & Caldwell, D.F. (1992). Bridging the boundary-external activity and per-

formance in organizational teams. Administrative Science Quarterly, 37, 634-665

2. Baldwin, T.T., Bedell, M.D., & Johnson, J.L. (1997) The social fabric of team-based MBA

program: Network effects on student satisfaction and performance. Academy of Manage-

ment Journal, 40, 1369-1397.

3. Borgatti, S.P. & Everett, M.G. (1997). Network Analysis of 2-mode data. Social Networks,

19, 243-269.

4. Borgatti, S.P., Everett, M.G. & Freeman, L.C. (1992). UCINET IV version 1.0 reference

manual. Colmbia, SC: Analytic Technologies

5. Cummings, J.N., & Cross, R. (2003) Structural properties of work groups and their conse-

quences for performance. Social Networks, 25, 197-281

6. Faraj Samer and Spruoll Lee (2000), Coordinating Expertise in Software Development

Teams, Management Science, Vol 46, No 12.

7. Freeman, L.C. (1979), Centrality in social networks, Conceptual clarification. Social Net-

works, 1: 215-239.

8. Grinter E. Rebecca and Herbsleb D. James (1999), The Geography of Coordination: Deal-

ing with Distance in R&D Work, Proceedings of the international ACM SIGGROUP con-

ference on Supporting group work

9. Hansen, M.T., Podolny, J.M. & Pfeffer, J. (2001). So many ties, so little time: A Task

contingency perspective on corporate Social Capital in organizations. Social Capital of Or-

ganizations, Elsevier Science Ltd., Vol 18, pages 21-57.

10. Katz, N., Lazer, D., Arrow, H. and Contractor, N. (2004), Network Theory and Small

Groups, Small Group Research, Vol. 35, No 3.

11. Kraut, E. Robert & Streeter, A. Lynn (1995). Coordination in Software Development.

Communication of the ACM, Vol. 38, No. 3.

12. Laplante, A Phillip & Neill, J. Collin (2004). The Demise of the Waterfall Model is immi-

nent and other Urban Myths. ACM Queue, (February 2004)

13. Monge R. Peter & Contractor S. Noshir (2003). Theories of Communication Networks.

Oxford University Press.

14. Reagens, R., &, Zuckerman, E.W. (2001). Networks, diversity and productivity: The social

capital of corporate R&D teams, Organization Science, 12, 502-517.

15. Reagens, R., Zuckerman, & E., McEvily, B. (2004) How to make the Team: Social Net-

work vs. Demography as Criteria for Designing Effective Teams. Administrative Science

Quarterly, 49, 101-133

16. Shaw, M.E., 1964, Communication Networks, In L. Berkowitz(Ed.), Advances in experi-

mental social psychology: 111-147. New York: Academic.

17. Sparrowe, R.T., Liden, R.C., Wayne, S.J., & Kraimer M.L. (2001) Social Networks and the

performance of individuals and groups. Academy of Management Journal, 44, 316-325

18. Stewart L. Greg & Barrick R. Murray (2000). Team Structure and Performance: Assessing

the mediating role of intrateam process and the moderating role of task type. Academy of

Management Journal, Vol 43, No. 2

19. The Standish Group, http://www.standishgroup.com/press/article.php?id=2 (2003)

20. Wasserman, S & Faust, K (1994). Social network analysis. Cambridge, MA: Cambridge

University Press.

21. Yang, Heng-Li & Tang, Jih-Hsin (2004). Team structure and team performance in IS

development: a social network perspective. Information & Management, 41, 335-349

Appendix

Fig. 1. Advice Network of Team 1

Fig. 2. Task Network of Team 1

Fig. 3. Advice Network of Team 2

Fig. 4. Task Network of Team 2

Fig. 5. Advice Network of Team 3

Fig. 6. Task Network of Team 3

Fig. 7. Advice Network of Team4

Fig. 8. Task Network of Team 4