Predicting team performance based on past individual achievements using artificial neural networks
Lund University, Faculty of Engineering
Lund University, School of Economics and Management
Ever since the Roman Empire, groups have been composed, compared, and contrasted based on individual performance. However, which are the key determinants of group performance? Is it possible to predict the performance based on certain factors?
Recent research is ambiguous regarding the factors that influence the performance of a group. What if project managers could predict the performance of their teams based on the characteristics and history of prospective team members—would this result in organizations being more successful due to the process of constructing high performance team?
This paper aims to construct an artificial neural network (ANN) model to investigate the linkage between past individual performance and team performance, in order to predict group performance. For this investigation, a case study is performed, where students constitute a relevant target group as they are measured in absolute terms, which simplify the comparison of the individuals and the teams. The performance data consists of registered grades, both from individual achievements and from group performance. In this case, an ANN model is trained and validated using sets of data from the population. The paper confirms the applicability of ANN modeling when constructing project teams in an academic setting; however, it is considered equally applicable in the private sector. Using the ANN model yields the ability to predict 74.3% of the teams' performance, which can be used to optimize the composition of groups.
This study is considered a first step in solving some of the important challenges in the general context of project management with regard to team performance by supporting the utility of using advanced computational algorithms in the research field and providing project managers with an important tool in the process of designing successful teams.
Keywords: team composition, group performance, individual performance, artificial neural network, performance prediction
Predicting Team Performance Based on Past Individual Achievements Using Artificial Neural Networks
Which factors have the greatest impact on a team's performance? Is it possible to predict the performance of a team based on certain key determinants? Ever since the Roman Empire, groups have been composed, compared, and contrasted based on the performance of its individuals. The focus on individual performance has been prevalent in many parts of society up until the Second World War. Steiner (1972) concluded that most post-Second World War research focuses on the area of team composition and on the individual performance in comparison to the team performance. Furthermore, Steiner (1972) argued that the theoretical perspective is important when it comes to constructing teams in order to create high performance groups and maximize their level of output.
Despite decades of research, Guzzo and Shea (1992) and Stewart (2006) argued that there is ambiguity in the field of team composition within organizational research. According to LePine, Hollenbeck, Ilgen and Hedlund (1997), this ambiguity is mainly addressed in research discussing which attributes should be taken into consideration, how the aggregation of the team members' individual performance is related to the group performance, and status differences among the team members.
McGrath, Arrow, and Berdahl (2000) suggested that future team and group research should focus on three main areas: theory development, computational modeling, and empirical research. Concerning the issues stated by LePine et al. (1997) and the suggestions made by McGrath, Arrow, and Berdahl (2000), this paper aims to investigate further the possible linkage between past individual performance and team performance, in order to predict the outcome of a group's performance based on individual contributions using a computational model.
University students in a business administration program were found to constitute a relevant target group for this study as they work in projects in temporary teams with similar requirements as to the ones in the private sector. In addition to these similarities in requirements, student performance is measured in absolute terms using grades that could be compared to the metrics in organizations' performance-based incentive systems such as sales volume, service level, etc. In order to predict the performance of a specific group, an Artificial Neural Network (ANN) analysis has been used.
The result of the ANN analysis aims to explain how project teams among students can be optimally composed to generate a maximum level of output performance.
This study is considered a first step in solving some of the important challenges in the general context of project management with regard to team performance, supporting the utility of using advanced computational algorithms in the research field, but also providing project managers with an important tool in the process of designing successful teams.
1. To investigate the linkage between individual performance and team performance using ANN.
2. To evaluate the applicability of an ANN based approach when designing successful teams.
This section presents relevant contemporary research conducted within the field of team performance. The authors introduce the development of teams throughout the last decades, both literarily and in its organizational context. The concept of team performance is introduced along with different output measurements. Finally, a more extensive review of previous studies that investigates different factors trying to explain the influence on team performance is presented.
Definition of Teams
Organizational research has focused on the concept of groups as an organizational building block ever since Leavitt (1975) suggested that groups, rather than individuals, were the key to designing a successful organization. Before this, organizational and psychological research related to group composition mainly explored the performance aspects of individuals within groups and the individual implications that group membership entails (Steiner, 1972).
Contemporary research has since then replaced “groups” with the idea of “teams” (Guzzo & Dickson, 1996); an augmented concept that often includes the notion of shared purpose and commitment, but also shared and empowered leadership, and a focus on the collective output rather than the individual (Katzenbach & Smith, 1993).
However, in this paper, as in many others, the notion of group and team are used interchangeable. While the authors acknowledge the difference between the two concepts, it is found to be of less importance in the context of the analysis performed in this paper. Literature related to both concepts is found to be of relevance and is therefore used as a theoretical foundation, further attenuating the difference between the two concepts.
What is Team Performance?
Team performance is often defined as the level of output deriving from the coordination and communication processes that teams develop over time (Bowers, Pharmer, & Salas, 2000). The output level measurement is commonly defined in very broad terms, as both Hackman (1987) and Guzzo and Dickson (1996) preferred. The definition does not only include the group-produced outputs (e.g., quantity, quality, speed, service level, etc.) but also the consequences that a group has on its members (individual learning), and the improvement of the group's capabilities to perform its tasks (group learning). While the authors acknowledge the general importance of having a broad definition of team performance, this paper takes a more narrow view of the concept and focuses on the actual group-produced output, in order to facilitate the quantitative analysis that the paper is based upon.
As a natural reference to the building block of the team—the individual—Steiner (1972) defined the performance of the team as that of its individual members by introducing the concept of process loss or process gain. This gain or loss is defined as the difference between the potential team performance and that of the actual, where the potential team performance is the sum of the members' potential individual performances (Steiner, 1972). But what determines this process gain or process loss?
Explaining Team Performance
According to Steiner (1972), there are three factors that influence a team's performance: the nature of the task (which and the amount of resources needed); group resources (resources available); and group processes (actions taken within the group). Related to the subject of which factors influence the performance of a team, a great variety of research has been carried out and it is evident that there are additional factors that contribute to the performance of a team beyond the ones stated by Steiner (1972).
An often-cited study conducted by Cohen and Bailey (1997) (Horwitz & Horwitz, 2007; Kozlowski & Ilgen, 2006; McGrath et al., 2000; Stewart, 2006) presente a framework, which aims to bring understanding to the complex relationships between the factors influencing team performance. The framework implies that team performance is a function of a number of factors including environmental factors, design factors (organizational, task, and group design), internal and external processes, and group psychosocial characteristics. The framework identifies interdependence between internal and external processes, the psychosocial characteristics, and the design factors, resulting in a development of the general accepted “input-process-output” approach (Goodman, Ravlin, & Argote, 1986; McGrath, 1984), by which Steiner's (1972) theory can be categorized.
Environmental factors include those of the external environment, such as the characteristics of the industry and turbulence. Processes refer to communication and conflicts that may arise both internally among the team members as well as with external parties. Examples of psychosocial characteristics are shared norms among team members, cohesiveness, and group effects. Design factors refer to attributes within the organizational context (rewards and training), the task (autonomy and interdependence), and group composition (size and diversity).
Cohen and Bailey (1997) identified the design factors with the largest influence on a team's performance. Contemporary research since then has mainly focused on separate design factors, e.g., organizational context (Salas, Rozell, Mullen, & Driskell, 1999; Koslowski & Bell, 2003), task design (De Dreu & Weingart, 2003), and finally group composition and team diversity (Bowers et al., 2000; Webber & Donahue, 2001; Horwitz & Horwitz, 2007).
Stewart (2006) performed a meta-analysis with 93 studies on the 3 design factors concluded by Cohen and Bailey (1997). From the reviewed studies, Stewart (2006) presented a summary of the most common variables correlated to the design factors that influence a team's performance (Table 1).
Table 1. Design Factors Affecting Team Performance
|Team Composition||Group structure, size, ability, personality, gender, race, demographics, heterogeneity, familiarity, diversity, age.|
|Task Design||Structure, interdependence, self-managing groups, autonomy, task characteristics, team type.|
|Context and Leadership||Leadership, rewards, ecology, automation, supervision, computers, leadership training, team training|
Note. Common group design factors affecting team performance based on Stewart (2006).
As argued by Steiner (1972), team composition is one of the most important design factors as well as one of the factors that have received the most frequent attention in research (Cohen & Bailey 1997; Guzzo & Dickinson, 1996; Stewart, 2006). Team composition refers to the nature and characteristics of group members, and Stewart (2006) identified three important variables in the aspect of group composition: aggregated individual characteristics, team member heterogeneity, and group size. Since this paper aims to investigate previous performance among individuals within a team, the authors mainly concentrate on the group design factor, including the three important variables identified by Stewart (2006), not taking effects of external factors into consideration. Hence, the study refers back to the input-process-output model (McGrath, 1984) as group composition denotes typical input.
Aggregated characteristics refer to whether or not a team's performance will improve linearly according to the member's possession of certain desirable characteristics. Different individual characteristics such as the skill and ability, different personality traits, background, and experience of the team members may underlie the team constructs. The aggregation is based on the assumption that each team member contributes with an amount of talent, and that an increasing amount of talent always is preferable (Stewart, 2006). In the same paper, Stewart concluded a positive correlation between a team's performance and the aggregated individual ability.
Tziner and Eden (1985) conducted a specific study among three-person military crews, which were combined in different teams concerning the individuals' characteristics and abilities to see how this affected the collective performance. The study suggests that crews with high-ability individuals performed better than the expected performance of each individual separately. Correspondingly, the performance of low-ability crews fell below the expected.
Another study also supporting the suggested correlation has been conducted by LePine et al. (1997), who concluded that teams where the majority of the members had a high cognitive ability performed better than teams including their antagonists. It was also found that individuals in high-ability teams had a tendency to help the member or members that had noticeable lower cognitive ability than the other members did.
A meta-analysis by Mount, Barrick, and Stewart (1998) also suggested that there are correlations between how well an individual performs in a team setting and the personal characteristics of that specific individual. Stewart (2003), however, despite the results from previous studies, concluded that high-performing individuals are not guaranteed to perform better collectively.
Member heterogeneity is broadly defined and refers to the mix of demographic lines (e.g., gender, ethnicity), psychosocial characteristics (e.g., personality, intelligence), and background (e.g., education, experience). In contrast to the linear relationship that is suggested to prevail between aggregated characteristics among team members and the collective performance (Stewart, 2006), it is suggested that the specific characteristics of an individual team member depend on the other team members' characteristics (Stewart, 2003). Fundamentally, non-linear relationship exists and the research conducted from this perspective refers to the heterogeneity of team member traits.
Contemporary research on the subject of team member heterogeneity is contradictory in its findings. Some studies favor heterogeneity as this best stimulates creativity and innovation (Bantel & Jackson, 1989; Magjuka & Baldwin, 1991). This, however, is in contrast to Campion, Medsker, and Higgs (1993), who did not find any positive correlation between the heterogeneity of team members' characteristics and team performance. They argued that the reason for this result may be the lack of heterogeneity within the studied population. However, Campion et al. (1993) supported the proposition that homogeneity among team members mitigates conflicts. Stewart's (2006) meta-analysis agreed with Campion et al. (1993) in the matter that team member heterogeneity correlates near zero, but Stewart (2006) also suggested that the correlation varies between different types of teams.
This paper takes a quantitative approach to exploring the linkage between past individual achievements and project team performance. In order to construct the ANN model, students were identified as the target population since quantitative metrics were needed for both individual performance, i.e., through exams, but also group performance, i.e., through projects in teams—represented by the bachelor's degree thesis.
Since the exact relationship between the dependent variable, the group's bachelor's degree thesis grade, and individual grades is unknown, an ANN model has been applied to the problem. To evaluate the applicability of the ANN model, the result is compared to that of a multivariate linear regression analysis.
Identifying Critical Factors for the Predictor Model
With regards to the prior research reviewed in this study, the variables identified as having a presumed influence on the collective performance are categorized accordingly: grades (aggregated characteristics); age and gender (team member heterogeneity); and group size.
The most important aspect when selecting the input variables for this study (Table 2) was to find an aggregated measure for the group as an entirety, which then can be associated back and reflect the past performance of the individuals. The variable categorized according to individual performance relates solely to the absolute measure of the grades received in compulsory courses before their bachelor's degree thesis. When deciding on the aggregated measure, the mean and standard deviation of each individual's grade constitutes the building blocks. Calculating the mean and standard deviation that of the individuals' separate mean and standard deviation then concludes the group's aggregated variables.
Table 2. Identified Input Variables
|F4||Mgrade standard deviation|
|F5||SDgrade standard deviation|
Note. Input factors to the ANN model as identified by the authors. To clarify the variable naming schema; Mgrade standard deviation represents the mean of each group member's grade standard deviation, while SDmean grade represents the standard deviation of the group members' mean grades.
In addition to the individual performance, other variables were identified as having a presumed effect on the outcome and are used to determine the heterogeneity (gender and age) of a group. Finally, group size was also considered as an important input variable.
Verifying Critical Factors
In order to perform a preliminary evaluation of the identified input variables, a Spearman's rank correlation analysis was conducted. This correlation analysis was found to be appropriate due to its ability to measure the correlation between two random variables without making assumptions of the nature of the linkage (Newbold, 1991).
To validate that the identified variables constitute the key determinants that influence team performance, an ANN model has been applied. Using an ANN has been leveraged by a number of studies to predict how different factors influence a specific outcome, e.g., the success of design-build projects in Singapore (Ling & Liu, 2004), the quality performance on Indian construction projects (Jha & Chockalingam, 2009), and the performance on athletes when selecting cricket teams (Yier & Sharda, 2008).
Figure 1. Illustration of a Typical Multilayer ANN
ANN modeling has a built-in learning capability and an ability to adapt to unknown data sets, resulting in the model being able to predict the outcome with minimal error as it adjusts the network parameters according to training algorithms (Zurada, 1992). The model used in this paper is a back propagation-learning algorithm with feed forward network structure, where the inputs are sent forward and the errors are propagated backwards. The model is recommended by Jain, Mao, and Mohiuddin (1996) because it is most suited for general pattern predictions. In its essence, the model is built on mathematical algorithms and is comprised by an input layer, a number of hidden layers, and an output layer, as illustrated in Figure 1.
The input layer presents data to the network, and it is referred to as the number of critical factors; in this paper it is identified through the Spearman's correlation analysis. The input layer is processed in the ANN neurons/nodes and the signal is transmitted between neurons in a hidden layer, which calculates the weighted sum of the inputs. The number of neurons in the hidden layer was defined through trial and error, and the output layer is represented by the team performance in this case.
When configuring the ANN model in this study, the holdout method suggested by Edwards (2007) is used as the data sets are considered to be sufficient. The method uses 80% of the data for training and model validation, and the remaining 20% for evaluating the true performance of the prediction model.
In order to maximize the applicability of potential findings and to increase the study's utility for both the research community and the private sector, the population for the study was selected from students at the Department of Business Administration at the Lund School of Economics and Management, Lund University, Lund, Sweden. This choice is based on the assumption that the field of business administration is the most similar field related to the activities that take place at an organization in the private sector; with a fair balance between activities performed individually, as well as in a project group setting.
The data that constitutes the empirical foundation for the study was collected in two steps. First, performance data was collected from a central student database for each individual in the population. As a representation for the team performance measure, the bachelor's degree thesis was identified as the most profound example of a project group setting. In this case, the work of the bachelor's degree thesis spanned a period of 12 weeks and was performed in a group of 2to 4 students. Besides the team performance data, individual grades for a number of preceding courses were collected for each individual in the population in order to establish the individual's past performance record.
The population consists of four consecutive classes of students at the Department of Business Administration, each containing an average of 239.5 students, yielding 958 individuals, specializing in 5 different subject areas: accounting, entrepreneurship, finance, marketing, organization and strategic management. The entire population had all completed a bachelor's degree thesis and before this had spent at least three years at the university.
In the initial population collected from those students having a bachelor's degree thesis registered during any of the last four semesters, 958 students were present. From those students, a number of individuals were filtered based on the requirement of having a sufficient performance background. Any individual without registered course grades in at least three different courses were removed from the population in order to increase the reliability of the study (e.g., in the case where a student transferred from another academic institution or in the case of any other anomaly). This left 455 students in the population with sufficient performance data present. The filtered population is described in Table 3.
Table 3. Individual Population Descriptive Statistics
Note. The gender variable is a dummy variable representing the gender of the individual where 0 denotes a male and 1 denotes a female.
Along with the individual course grades and the team grade for the students' bachelor's degree thesis, data about the group composition was collected, establishing the constellation of each group. Joining the 2 data sets, 174 complete teams were found to be present in the population. The teams in the population are described in Table 4.
Table 4: Team Population Descriptive Statistics
Note. See Table 2 for a brief explanation of the variable naming schema.
As Table 4 illustrates, a number of dummy variables were generated in addition to those described in the method section of this paper. Both gender and age, and their variance within the team, were calculated automatically from the individuals' social security numbers and were added as complementary data.
The identified input variables are evaluated using a Spearman's correlation approach in order to determine the critical input factors. As Table 5 illustrates, each of the nine input variables are found to have a small correlation with the target variable—the team's performance. Even though some of the factors showed a weaker correlation than others, all nine factors were included in the continued analysis as none were found to potentially complicate the ANN analysis.
Table 5: Correlation Analysis of Input Variables
|F4||Mgrade standard deviation||0.3227|
|F5||SDgrade standard deviation||0.3650|
Note. Correlation calculated using Spearman's rank correlation.
An important observation from the correlation analysis is with the exception of the “mean grade,” the standard deviation for each of the different input factors was found to have a greater correlation with the team's performance than the mean value (F3, F5, F7, and F9 in Table 5). This supports the argument that heterogeneity in team composition has an effect on team performance, even when it comes to performance itself. Age and gender also have an influence on the collective performance.
The relatively strong correlation of the standard deviation of the individual's own standard deviation in grade (F5) is also an interesting observation. This can be interpreted as the team's performance is dependent on whether the team is comprised of individuals that themselves are uneven in their past individual performances.
As for age (F6) and gender (F7), it is observed that the variance of the two (F8 and F9) has a stronger correlation than the variables respective averages, supporting previous research on the effects of heterogeneity as outlined in the literature review section.
While the correlation of the identified input variables are established using the Spearman's correlation, the significance and predictive nature remains to be determined. The constructed ANN model (as discussed in the Methodology section) is used in order to evaluate the predictive nature of the identified variables.
The performance of the ANN model is illustrated in Figure 2. Here, it is observed that the training and validation curves indicate that no over fitting occurred, as the training curve does not diverge significantly from the validation curve. This suggests that the ANN model successfully identifies patterns that can be generalized to a larger population.
Figure 2: Performance Evaluation; Mean Squared Error (MSE) for the ANN simulation
The actual performance of the ANN model is illustrated in Figure 3. The four quadrants represent the three different phases: training, validation, and testing, as well as a summary of the total model simulation. The green sub-quadrants represent the individual teams that are correctly classified according into each of the two output/target classes (where 1 represents a low final thesis grade, and class 2 represents a high final thesis grade for the team), while the red sub-quadrants represent those that were incorrectly classified. The blue quadrants show a summary of the total phase classification performance.
Figure 3. ANN Prediction Evaluation for the Training, Validation, and Test Sub-Populations
Note. The green quadrants represent successfully classified teams for each of the two target classes. Overall prediction ratios are illustrated in the blue quadrants and the total model prediction performance can be observed in the lower left diagram, in the test confusion matrix.
The important subject of analysis in Figure 3 is the test confusion matrix, where the trained and final ANN model is evaluated on a random independent set of the population in order to determine the actual performance. The model successfully predicts 74.3% of the test teams' performance, while erroneously classifying 5.7% as false-positives and 20% as false-negatives.
In order to evaluate the prediction performance of the ANN model, it is compared to a simple multivariate linear regression model, implemented using MATLAB. The coefficients from the regression model are illustrated in Table 6.
Table 6. Coefficients From the Multivariate Linear Regression Analysis
|F4||Mgrade standard deviation||0.0565|
|F5||SDgrade standard deviation||0.1203|
Note. See Table 2 for a brief explanation of the variable naming schema.
As this paper illustrates a classification problem, where a group is categorized as either high or low-performing teams, the residuals from the regression model is used to determine the prediction performance. This is done in order to obtain a comparable performance metric to the ANN model. The classification of a team is considered successful when the calculated output value from the regression model, rounded to the closest integer, matches that of the target class value of the team. This transforms the continuous output of the regression model to the discrete output of the ANN classification model.
Of the 174 teams in the population, 100 teams were correctly classified, resulting in a prediction performance of 57.5% for the multivariate linear regression model. In comparison to the ANN analysis, this multivariate linear regression model results in a significantly lower predicting ability.
Using an ANN model resulted in an ability to predict 74.3% of the teams' performance based on information of the individual team members and past performance. The prediction performance of the model is found to be high, considering the limited number of input variables and population size, but also when compared to other ANN based prediction studies (Ling & Liu, 2004; Jha & Chockalingam, 2009; Yier & Sharda, 2008) with similar types of problems.
When comparing the prediction performance of the ANN model to that of a plain multivariate linear regression model, the performance is found to be impressive, and constitutes a significant improvement in the successful prediction ratio from 57.5% to 74.3%.
The correlation between the input variables and the predicted team performance appears to be low when studied separately, with an exception of the average grade. The result in Table 5 indicates a stronger correlation between the input variables measuring the standard deviation than of those representing a group-average input variable, concluding that heterogeneity have an impact when designing successful teams. However, despite the low correlation between the individual variables and the output, when the input variables are analyzed in their entirety, the ANN based analysis results in a successful prediction probability of 74.3%.
This study confirms the applicability of ANN modeling when constructing project teams in an academic setting. Within organizational project teams, the same approach should be equally applicable, due to the similarities in task characteristics, but would require different performance measures for teams and individuals. Potential input variables could include metrics commonly included in performance-based incentive systems such as sales volume, service level, etc., but could also include softer but quantitative variables from peer reviews depending on what the organization want to measure, control and stimulate.
Potential Improvements and Future Research
This study verifies the applicability of ANN modeling in constructing optimal teams based on individual performance, but it is not in itself a complete solution. While the ANN approach described in this paper can assist an organization in the construction of teams, as well as provide the organizational research community with a tool to continue the exploration of team performance using computational models, the framework fails to provide an analytical solution to the team composition problem.
To explore the relationship between team performance and individual performance from a group composition viewpoint further, other computational models closely related to the ANN concept should be utilized. One approach that could potentially lead to a deeper understanding of the subject is the use of clustering algorithms based on ANN models. As the use of clustering algorithms would classify team performance based on the interdependence of the input variables, the understanding of exactly how the different variables of individual performance affect the output would be greatly improved.
An additional improvement is with regard to the specific combination of input variables that is used in the ANN model, which is of great importance to the prediction outcome. The models applicability varies with the nature of the teams' assignment and additional factors such as environmental issues are not taken into consideration in the analysis.
Concerning the relevance of the target group, it can be questionable if it is possible to extrapolate the performance of the students to that of organizational project teams. The students' grades on individual courses as well as the teams' theses grades may be a non-representative metric as teachers can judge it differently and vary among classes in the population. Furthermore, the grade is not always a fair metric of individuals' actual performance as it could potentially lack other levels of output, such as the individual's learning. The task characteristics of the thesis work may also not completely correspond to the actual requirements and the tasks characteristics of teams in the private sectors.
The ANN approach may not be a complete framework when designing successful teams as it exclude certain types of variables that have an impact on the individuals' performance, but that is not possible to take into consideration strictly using an ANN model. Such variables include social and psychological factors and environmental changes, which has a great influence on the total team performance.
Bowers, C. A., Pharmer, J. A., & Salas, E. (2000). When member homogeneity is needed in work teams: A meta-analysis. Small Group Research, 31(3), 305–327.
Cohen, S. G., & Bailey, D. E. (1997). What makes teams work: Group effectiveness research from the shop floor to the executive suite. Journal of Management, 23(3), 239–290.
De Dreu, C. K. W., & Weingart, L. R. (2003). Task versus relationship conflict, team performance, and team member satisfaction: A meta-analysis. Journal of Applied Psychology, 88(4), 741–749.
Edwards, S. R. (2007). Modeling perceptions of building quality – a neural network approach. Building and Environment, 47, 2762–2777.
Goodman, P. S., Ravlin, E. C., & Argote, L. (1986). Current thinking about groups: Setting the stage for new ideas. In P. S. Goodman & Associates (Eds.), Designing effective work groups (pp. 1–33). San Francisco, CA: Jossey-Bass.
Guzzo, R. A., & Dickson, M. W. (1996). Teams in organizations: Recent research on performance and effectiveness. Annual Review of Psychology, 47, 307–338.
Iyer S. R., & Sharda R. (2009). Prediction of athletes performance using neural networks: An application in cricket team selection. Expert Systems with Applications, 36(3), 5510–5522.
Jain, A. K., Mao, J., & Mohiuddin, K. M. (1996). Artificial neural networks: A tutorial. Computer, 29(3), 31–44.
Jha, K. N. & Chockalingam, C. T. (2009). Prediction of quality performance using artificial neural networks - Evidence from Indian construction projects. Journal of Advances in Management Research, 6(1), 70–86.
Kozlowski, S. W. J., & Bell, B. S. (2003). Work groups and teams in organizations. In W. C. Borman, D. R. Ilgen, & R. J. Klimoski (Eds.), Comprehensive handbook of psychology: Industrial and organizational psychology (pp. 12, 333–375). New York: John Wiley.
Kozlowski, S. W. J., & Ilgen, D. R. (2006). Enhancing the effectiveness of work groups and teams. Psychological Science in the Public Interest, 7(3), 77–124.
Ling, F. Y. Y., & Liu, M. (2004). Using neural network to predict performance of design-build projects in Singapore. Journal of Building and Environment, 39(10), 1263–1274.
McGrath, J. E. (1984). Groups: Interactions and performance. Englewood Cliffs, NJ: Prentice-Hall.
McGrath, J. E., Arrow, H., & Berdahl, J. L. (2000). The study of groups: Past, present, and future. Personality and Social Psychology Review, 4(1), 95–105.
Mount, M. K., Barrick, M. R., & Stewart, G. L. (1998). Five-factor model of personality and performance in jobs involving interpersonal interactions. Human Performance, 11(2/3) 145–165.
Newbold, P. (1991). Statistics for business and economics (3rd ed.). Englewood Cliffs, NJ: Prentice Hall.
Salas, E., Rozell, D., Mullen, B., & Driskell, J. E. (1999). The effect of team building on performance: An integration. Small Group Research, 30(3), 309–329.
Steiner, I. D. (1972). Group process and productivity. New York: Academic Press.
Stewart, G. L. (2003). Toward an understanding of the multilevel role of personality in teams. In M. R. Barrick & A. M. Ryan (Eds.), Personality and work: Reconsidering the role of personality in organizations (pp. 183–204). San Francisco, CA: Jossey-Bass.
Stewart, G. L. (2006). A meta-analytic review of relationships between team design features and team performance. Journal of Management, 32(1), 29–54.
Stewart, G. L., & Barrick, M. R. (2000). Team structure and performance: Assessing the mediating role of intrateam process and the moderating role of task type. Academy of Management Journal, 43(2), 135–148.
Tziner, A., & Eden, D. (1985). Effects of crew composition on crew performance: Does the whole equal the sum of its parts? Journal of Applied Psychology, 70(1), 85–93.
Webber, S. S., & Donahue, L. M. (2001). Impact of highly and less job-related diversity on work group cohesion and performance: A meta-analysis. Journal of Management, 27(2), 141–162.
Zurada, J. M., (1992). Introduction to artificial neural systems. Los Angeles, CA: West Publishing.
© 2010 Project Management Institute