Social and Engineering Aspects of Construction Site Management using Simulation and Social Network Analysis

TABLE OF CONTENTS

LIST OF TABLES

LIST OF FIGURES

CHAPTER 1 INTRODUCTION
1 1 Introduction
1.2 Research Objectives
1.2.1 Research Objective 1: To Develop the Jobsite Social Network of Construction Crews and Quantify its Impact on Performance (a Case Study)
1.2.2 Research Objective 2: To Investigate the Impact of Social Conformity on the Performance of Construction Crews
1.2.3 Research Objective 3: To Identify the Improvement Direction (Benchmarks) for Inefficient Construction Crews
1.2.4 Research Objective 4: To Investigate the Cost/Benefit of Low or High Reliable Construction Crews and Develop a New Educational Version of Parade Game
1.3 Research Scope
1.4 Research Significance
1.5 Research Outline

CHAPTER 2 RESEARCH BACKGROUND
2.1 Social Network Analysis
2.1.1 Social Conformity: Tendency to Match Group Norm
2.1.2 Social Learning: Learning from Others
2.1.3 Construction Research Pertaining to Social Network Analysis
2.2 Benchmarking through Data Envelopment Analysis (DEA)
2.2.1 DEA in Construction Industry: Identify Performance Benchmarks (Role Model)
2.3 Production Capacity Variability/Reliability and Construction Work Plan Variation/Reliability
2.3.1 Unstable Work Flow, Consequence of Work Plan Variation and Production Capacity Variability
2.3.2 Parade Game: Demonstrate the Impact of Variability in Construction Project
2.3.3 Investment on Planning to Improve Construction Work Plan Reliability and Consequence
2.4 Identified Gaps in the Body of Knowledge
2.4.1 Identified Gaps in Construction Research Pertaining to SNA Application
2.4.2 Identified Gaps in Construction Research Pertaining to Benchmarking via DEA
2.4.3 Identified Gaps in Construction Research Pertaining to Capacity Variability/Reliability and Work Plan Variation/Reliability
2.4.4 Addressing the Identified Gaps

CHAPTER 3 RESEARCH METHODOLOGY
3.1 Methodology to Achieve Research Objective 1, To Develop the Jobsite Social Network of Construction Crews and Quantify its Impact on Performance (a Case Study)
3.2 Methodology to Achieve Research Objective 2, To Examine the Impact of Social Conformity on the Performance of Construction Crews
3.3 Methodology to Achieve Research Objective 3, To Identify the Improvement Direction (Benchmarks) for Inefficient Construction Crews
3.3.1 Conventional DEA Benchmarking
3.3.2 Social Network based DEA Benchmarking
3.4 Methodology to Achieve Research Objective 4, To Investigate the Cost/Benefit of Low or High Reliable Construction Crews and Develop a New Educational Version of Parade Game

CHAPTER 4 RESEARCH EXECUTION AND ANALYSIS RESULTS
4.1 Analysis Results and Discussion pertaining to Research Objective 1, To Develop the Jobsite Social Network of Construction Crews and Quantify its Impact on Performance (a Case Study)
4.1.1 Summary and Conclusion (Objective 1)
4.2 Analysis Results and Discussion pertaining to Research Objective 2, To Examine the Impact of Social Conformity on the Performance of Construction Crews
4.2.1 Results of First Case Study (Case Study 2)
4.2.2 Results of Second Case Study (Case Study 3)
4.2.3 Summary and Conclusion (Objective 2) 109 4.3 Analysis Results and Discussion pertaining to Research Objective 3, To Identify the Improvement Direction (Benchmarks) for Inefficient Construction Crews
4.3.1 Results of Conventional DEA benchmarking
4.3.2 Results of Social Network based DEA benchmarking
4.3.3 Summary and Conclusion (Objective 3)
4.4 Analysis Results and Discussion pertaining to Research Objective 4, To Investigate the Cost/Benefit of Low or High Reliable Construction Crews and Develop a New Educational Version of Parade Game
4.4.1 Production Output
4.4.2 Cost Output
4.4.3 Summary and Conclusion (Objective 4)

CHAPTER 5 SUMMARY
5.1 Intellectual Merit and Broader Impact

CHAPTER 6 FUTURE RESEARCH

REFERENCES

APPENDICES
APPENDIX A - ADJACENCY MATRICES OF CASE STUDY
APPENDIX B - ADJACENCY MATRICES OF CASE STUDY
APPENDIX C - ADJACENCY MATRICES OF CASE STUDY
APPENDIX D - INTRODUCTION TO DATA ENVELOPMENT ANALYSIS

ABSTRACT

ABBASIAN HOSSEINI, SEYED ALIREZA. Social and Engineering Aspects of Construction Site Management using Simulation and Social Network Analysis. (Under the direction of Dr. Min Liu and Dr. Simon Hsiang.) The crews/actors/subs during a construction project make relationship and communicate with each other on the jobsite primarily when they work in a task sequence or when they work in the same working area at the same time. These interdependencies can have various impacts on their performance, the decisions their supervisor make and their action from both engineering and social aspects. The main focus of the past research is on the project parties’ relationship based on the information exchange and formal communication, while the research pertaining to the interpretation and investigation of the construction crews/trades’ interdependencies during the construction project is very limited. How are the construction jobsite actors connected in a construction jobsite? How do the existing interdependencies among them affect their performance? And how can understanding these interdependencies be beneficial for construction site managers? The primary goal of this research is to better understand the existing interdependencies among the construction crews/trades/subs and its impact. Particularly, the objectives of this research are to: 1) develop the jobsite social network of construction crews/trades and quantify its impact, 2) investigate the impact of social conformity on the performance of construction crews/trades, 3) identify the improvement direction (benchmarks) for inefficient construction crews/trades, and 4) investigate the cost/benefit of low or high reliable construction crews/trades and to develop a new educational version of Parade Game.

First, social network analysis (SNA) is implemented to develop a technique to construct the dynamic jobsite social network of crews/trades in a project and quantify its impact through the network centrality analysis. The results of a case study are presented. Then, SNA and social norm analysis are combined as a method to measure conformity, one of the main social network influences types that results in a change of performance/behavior in order to fit in a group, at construction crew/trade level and demonstrate how it can play role in the performance of crews/trades/subs particularly in their work plan reliability through two case studies. Then, inspired by social learning phenomenon, data envelopment analysis and SNA is combined to develop a procedure that can identify the improvement direction for the inefficient crews/trades/subs in a construction project. At the end, the research concentrates on the engineering aspects of the jobsite interdependencies by developing a simulation model, as a new educational version of Parade Game, that uses different variability levels and the corresponding costs at different work stations to investigate the relationship between the interdependencies and crews/trades’ variability/reliability.

Results demonstrate that the performance of construction crews/trades is under the influence of the social aspect of the interdependencies as well as the engineering aspect. They show that there is an association between influences a crew/trade/sub receives from the network and his/her performance. Results of case studies show that the subcontractors follow the performance norm in the project and their tendency to follow the norms of their neighborhood is higher than their willingness to follow the project norm. Parade Game simulation results also show that the production will enhance if the reliability increases and the investment made to improve reliability will return in most of the scenarios.

This research is significant and valuable as it looks at construction jobsite interdependencies from an exclusively analytical perspective, which has not been done previously. Previous research also did not investigate the social aspects of the construction crews/trades/subs interdependencies. Construction personnel at every level of management are constantly planning and trying to figure out how best to manage and coordinate the construction crews/trades/subs. A better understanding of the existing jobsite interdependencies will help project managers to control it through better planning and leadership, consequently increasing jobsite productivity.

© Copyright 2015 by Seyed Alireza Abbasian Hosseini All Rights Reserved

Social and Engineering Aspects of Construction Site Management using Simulation and Social Network Analysis

by

Seyed Alireza Abbasian Hosseini

A dissertation submitted to the Graduate Faculty of North Carolina State University in partial fulfillment of the requirements for the Degree of Doctor of Philosophy

Civil Engineering

Raleigh, North Carolina

APPROVED BY:

illustration not visible in this excerpt

Dr. Min Liu Committee Co-Chair

illustration not visible in this excerpt

Dr. Michael L. Leming

illustration not visible in this excerpt

Dr. Simon M. Hsiang Committee Co-Chair

illustration not visible in this excerpt

Dr. Edward J. Jaselskis

DEDICATION

This work is dedicated to my wife, Soha. Without her love, sacrifice, patience, and support during the past four years, this would not have been possible. This work is also dedicated to my parents and my brothers. It is their unconditional love and support that motivates me to set higher targets.

BIOGRAPHY

Seyed Alireza (Ali) is the second of three children born to Hashem and Nasrin Abbasian Hosseini in Mashhad, Iran in 1986. He attended Allameh High school for talented students and graduated in 2004. He entered Azad University of Mashhad in 2004 and earned a Bachelor’s degree in Civil Engineering in 2008. Mr. Abbasian Hosseini studied Construction Engineering and Management as a Master of Science student at Iran University of Science and Technology in 2008 under supervision of Prof. Parviz Ghoddousi. He focused on the application of lean construction practices in construction projects through computer simulation. He has experience including working for two of the largest construction companies in his hometown as a project estimator and controller. In August 2011, he enrolled in North Carolina State University to pursue his Doctorate of Philosophy in Construction Engineering and Management under the guidance of Prof. Min Liu.

ACKNOWLEDGMENTS

There are several individuals I would like to acknowledge. First and foremost, I want to acknowledge my two advisors, Dr. Min Liu and Dr. Simon M. Hsiang for their guidance, patience, motivation, encouragement, and immense knowledge not only on my research but also on my life and mental growth. Secondly, I would like to thank my committee members, Prof. Michael L. Leming, and Prof. Edward J. Jaselskis for their invaluable recommendations. Part of this dissertation research was conducted with the financial support of the Lean Construction Institute (LCI). LCI is a non-profit organization operates as a catalyst to transform the industry through lean project delivery with the goal of increasing owner and construction supply chain satisfaction. I wish to acknowledge LCI and its founder, Prof. Gregory Howell for their support and collaboration. Next, I would like to thank the other staff and faculty members of the North Carolina State University Civil, Construction, and Environmental Engineering Department who have taught and assisted me in various ways during my studies. I also extend my thanks to the current and former colleagues: Dr. Marion (Mark) Russell, Ashtad Javanmardi, and Rabia Omer. I also appreciate the help from my friends in Mann Hall, Dr. Amirhossein Norouzi, Behrooz Keshavarzi, and Mehdi Mashayekhi. Many others―family, friends, colleagues, fellow students, students that I have taught―have helped and motivated me to progress through this academic journey—thank you.

LIST OF TABLES

Table 1.1. Tasks and responsible crews

Table 2.1. Summary of previous construction research pertaining to social network analysis

Table 3.1: Schematic adjacency matrix (ASN) for relationships of subcontractor crews

Table 3.2. sub-scenarios based on the Arrangement and Cost Scenarios for Parade Game model

Table 3.3. A brief description of ARENA modules

Table 4.1. Centrality analysis results for the jobsite social network in Week 18

Table 4.2. Work plan variation Analysis results for the Week 18

Table 4.3. Correlation analysis results - weekly Centrality Ranks (CRs) versus weekly Variation Ranks (VRs)

Table 4.4. Correlation analysis results - monthly Centrality Ranks (CRs) versus monthly Variation Ranks (VRs)

Table 4.5. WPRI measurements for drywall subcontractor crew - Case Study 2

Table 4.6. Adjacency matrix of Month 4 - Case Study 2

Table 4.7. Correlation analysis results for the relationship between DFNN/DFPN and project time - Case Study 2

Table 4.8. Correlation analysis results for the relationship between DFNN/DFPN and project time - Case Study 3

Table 4.9. Summary of the Raw Data

Table 4.10. Eigenvalues and corresponding eigenvectors of the selected PCs

Table 4.11.The New Set of Data based on PCA

Table 4.12. Conventional benchmarks for the inefficient crews

Table 4.13. ASN of the specialty crews

Table 4.14. Results of Conventional DEA versus SN-based DEA Benchmarking

Table 4.15. Comparison between the nine Parade Game Arrangement Scenarios - Completion Time (week)

Table 4.16. Comparison between the nine Parade Game Arrangement Scenarios - Number in Buffer (unit)

Table 4.17. Comparison between the nine Parade Game Arrangement Scenarios - Lost Capacity (capacity)

Table 4.18. Summation of the Crews’ Total Cost for the 81 Parade Game sub-scenarios

Table 4.19. GC Total Cost for the 81 Parade Game sub-scenarios

Table 5.1. Summary of Research Objectives, Methods, and Conclusions

LIST OF FIGURES

Figure 1.1. A typical building project at construction phase (Tekla 2015)

Figure 1.2. Dividing the project into different working areas and assign the tasks

Figure 1.3. Interdependencies among crews based on tasks sequence (engineering aspect)

Figure 1.4. Interdependencies among crews based on sharing the working area (social aspect)

Figure 1.5. Research design flow chart

Figure 3.1. Methodology steps to achieve Research Objective 1

Figure 3.2. Schematic jobsite social network based on working space

Figure 3.3. Methodology steps to achieve Research Objective 2

Figure 3.4. Methodology steps to achieve Research Objective 3

Figure 3.5. Methodology steps to achieve Research Objective 4

Figure 3.6. Triangular PDFs used to reflect various production capacity reliability (variability) levels in the model

Figure 3.7. 9 Arrangement Scenarios for the Parade Game model

Figure 3.8. Parade Game (new version) simulation model by ARENA

Figure 3.9. The graphic animation screen of the Parade Game model

Figure 3.10. The graphic animation screen of the Parade Game model (Zoom in to Stations 1 and 2)

Figure 4.1. Jobsite social network in Week 18 based on frequency

Figure 4.2. Jobsite social network in Week 18 based on severity

Figure 4.3. Jobsite social network topology over the course of the project (28 weeks)

Figure 4.4. Week-by-week consistency between Centrality and Variation Ranks for the “Painting” Crew

Figure 4.5. Week-by-week consistency between Centrality and Variation Ranks for the “Electrical” Crew

Figure 4.6. Month-to-month consistency between Centrality and Variation Ranks for the “Painting” Crew

Figure 4.7. Month-to-month consistency between Centrality and Variation Ranks for the “Electrical” Crew

Figure 4.8. WPRI4 and ProjectNorm4 over time for drywall subcontractor crew - Case Study 2

Figure 4.9. Social networks of the subcontractor crews over the course of the project - Case Study 2

Figure 4.10. Deviation from project norms over time - Case Study 2

Figure 4.11. Deviation from neighborhood norms over time - Case Study 2

Figure 4.12. Trends of DFPN and DFNN over time - Case Study 2

Figure 4.13. Social networks of the subcontractor crews over the course of the project - Case Study 3

Figure 4.14. Trends of DFPN and DFNN over time - Case Study 3

Figure 4.15. The specialty crews organized by social network

Figure 4.16. Steps of SN-based DEA benchmarking procedure for the “Finishing” crew (FI): (a) FI sub-SN; (b) FI sub-SN showing Relationship Score (RS); (c) FI sub-SN showing efficiency level; and (d) FI sub-SN showing Relative Efficiency Distance (RED)

Figure 4.17. Practical benchmarks for the “Finishing” crew (FI)

Figure 4.18. Parade Game simulation model results - Replacing first station by a crew with a low/high reliable capacity

Figure 4.19. Parade Game simulation model results - Replacing middle station by a crew with a low/high reliable capacity

Figure 4.20. Parade Game simulation model results - Replacing last station by a crew with a low/high reliable capacity

Figure 4.21. Parade Game simulation model results - Replacing all the stations by crews with a low/high reliable capacity

CHAPTER 1 INTRODUCTION

1.1 Introduction

Research on the decision making process in construction has focused at the project level, where the contract documents specifying the responsibility and risk of each participant are the main basis for consideration. Although this formal decision making usually comes with clear objectives and boundaries, in practice, strict adherence to the formal procedures and chains of command would not always be possible without an unacceptable expenditure of time and money. In reality, it is the informal day-to-day or moment-to-moment decision making at the site level, usually led by a foreman/supervisor (the leader of a specialty trade, crew or subcontractor), that plays a key role in keeping a project’s tasks going without undue delay or the generation of undesirable cost.

Current dominant construction project scheduling techniques focus on the project tasks (or activities). They rely on development of a logical network of tasks that represent the dependencies among them (such as deterministic methods including Critical Path Method (CPM) or probabilistic methods including Program Evaluation and Review Technique (PERT)). The structure is a topological map of discrete tasks joined by logical relationships. It indicates who should work on what task and in which area. Figure 1.1 shows a typical building project during construction. To minimize the jobsite interference and plan the work better, project managers usually divide the overall project into different working areas and manage each area as a mini-project (as shown in Figure 1.2). Then, a task schedule is created for each of the areas describing the project tasks, their logical relationships and the crews responsible to perform them. Note that “crews” in this document refer to construction working crews that can belong to general contractor (GC), specialty trades, working parties or subcontractors of project. Table 1.1 and Figure 1.2 demonstrate a simplified version of this process. The interdependencies among the trades come from the task dependencies in each working area. These interdependencies among crews can be based on the tasks sequence, when they work in a task sequence with the production output of one being a prerequisite to the activity of another, that is the engineering aspect (see Figure 1.3), or based on sharing the working area, when they are supposed to work in the same working area at the same time, that is the social aspect (see Figure 1.4).

illustration not visible in this excerpt

Figure 1.1. A typical building project at construction phase (Tekla 2015)

Table 1.1. Tasks and responsible crews

illustration not visible in this excerpt

illustration not visible in this excerpt

Figure 1.2. Dividing the project into different working areas and assign the tasks

illustration not visible in this excerpt

Figure 1.3. Interdependencies among crews based on tasks sequence (engineering aspect)

illustration not visible in this excerpt

Figure 1.4. Interdependencies among crews based on sharing the working area (social aspect)

The interdependencies of construction crews can impact their performance, the decisions their supervisors make and their action in various ways. For instance, when one crew fails to complete a task based on the schedule, follow-on crews in the task schedule are delayed in starting their portions of the project (Thomas and Flynn 2011). Or when two or more crews are supposed to work in the same working area at the same time, there would be a potential of conflict due to inadequate working spaces or work area access, overcrowding jobsite, safety hazards, etc. and subsequently loosing productivity (Gou 2002, Thomas et. al. 2006, Watkins et al. 2009). This impact can easily go beyond the project task and space dependencies in large and complex construction projects, especially when there are several independent subcontractors working on a jobsite at the same time. For instance, when a subcontractor finds a project unreliable (for example when he/she sees that other subcontractors are not going to finish their tasks on time), he/she may decide to take his/her resources to another job, since subcontractors work often on multiple projects simultaneously to keep their workload at the maximum level in order to have a supply of available work and maximize profitability at any given time (Sacks 2004, Thomas and Flynn 2011, Freeman and Seppanen 2014).

Social Network Analysis (SNA), introduced by Moreno (1960) is a methodology to determine the conditions of social structures by investigating the relations and interrelationships of a set of actors (De Nooy et al. 2005). The interdependencies between crews in a construction project constitute a social network, which can be analyzed and interpreted via SNA (refer to Figure 1.4). It is arguable that understanding the jobsite social network can help the construction site managers to coordinate the crews more effectively and succeed in the challenging environment of projects; however, achieving this skill takes years of experience and few superintendents could articulate it (Wambeke et al. 2012 & 2014).

Although project management research gradually turns to the social sciences to interpret the issues related to construction, application of SNA in construction research is still rather limited (Pryke et al. 2011). Previous studies mostly focused on the project parties’ relationship based on information exchange and formal communication. Analysis and interpretation of the dependencies among the construction crews during construction has not been well documented in industry handbooks or academic research work. Specifically, 1) How are the construction site actors connected in a construction jobsite? 2) How much do the existing interdependencies among them affect their performance? 3) Which crews influence the others more and which are more under the influence of the others? 4) How can knowing the jobsite social network be beneficial for construction site managers? Answering these questions requires understanding the interdependencies of construction crews and analyzing their underlying jobsite social network during construction projects. This research contributes to the body of knowledge as it uses systematic and applicable approaches/procedures to quantify the jobsite social network impact and demonstrate how the interdependencies play a role in the performance of the construction crews over the course of a project. The results of a few case studies are also discussed.

1.2 Research Objectives

The primary goal of the research is to better understand the existing interdependencies among the construction crews in construction phase and the impact of that on their performance from both engineering and social aspects. Guiding this research are four individual research objectives that support the overall framework of the research. The first two objectives focus more on the identification and detection of the interdependencies and its impact. The Objectives 3 and 4 concentrate on making the implementation strategy:

- Research Objective 1: To develop the jobsite social network of construction crews and quantify its impact on performance (a case study)
- Research Objective 2: To examine the impact of social conformity on the performance of construction crews.
- Research Objective 3: To identify the improvement direction (benchmarks) for inefficient construction crews.
- Research Objective 4: To investigate the cost/benefit of low or high reliable construction crews and develop a new educational version of Parade Game.

1.2.1 Research Objective 1: To Develop the Jobsite Social Network of Construction Crews and Quantify its Impact on Performance (a Case Study)

Studies showed that the same subcontractors have better efficiency/productivity (about 30%) on similar scope in areas where only one crew owns the working space, i.e., no influence received from others (Thomas et. al. 2006, Seppänen 2009). The construction industry lacks a systematic and applicable approach to quantify the influence of jobsite social network, based on the working area dependencies, on the crews’ performance over the course of the project. To achieve the first research objective, using SNA, an analytical approach is proposed to show how the jobsite social network is developed and quantify the influence crews receive from the network at any time of a project. Then, the impact of jobsite social network on the crews’ performance, particularly on their work plan variation (defined as the time difference between what was planned and what occurred), is explored through a case study.

1.2.2 Research Objective 2: To Investigate the Impact of Social Conformity on the Performance of Construction Crews.

It is not an uncommon situation that a subcontractor makes decision based on the other subcontractors’ behavior. In a construction project, it can be argued that there is usually little motivation (no reward or profit) for a subcontractor to have a better performance than the other subcontractors. When a subcontractor finds other subcontractors make excuses to start or complete their tasks with delay and there is no strict enforcement by the GC, he/she realizes that it is not urgent to make his/her best effort in completing the tasks on time based on the planned schedule. Thus, he/she learns from the other subcontractors to develop excuses for their tasks’ delay and utilize the minimum possible resources for completing the tasks (Freeman and Seppanen 2014). Or, as mentioned previously, when a subcontractor finds the project unreliable, he/she may decide to take his/her resources to another job to maximize his/her profitability at any given time (Sacks 2004, Thomas and Flynn 2011, Freeman and Seppanen 2014). On the contrary, a subcontractor tends to have a more satisfactory level of reliability (i.e. start and complete the tasks on-time) when he/she finds all the other subcontractors perform their tasks nearly based on the planned time schedule. It is firstly because he/she finds the project profitable when he/she is able to rely on the reliability of the project work plan, so he/she attempts to be as reliable as the others to keep the project profitable for himself/herself. The second reason is that no one prefers to be the most unreliable party in the project because of fear of being punished (or sued) by the GC or other subs. To summarize, it can be argued that a subcontractor usually decides to slow down or speed up their pace based on the interdependencies he/she has with other subcontractors, and therefore, as the project proceeds, he/she tries to sync up with the project norm.

Conformity, also known as herding behavior, is one of the influences of being in a social network. It refers to the tendency of changing behavior (or performance) in order to fit in a group and converge to the group norms (Asch 1951). To reach the second objective, this research develops an analytical approach to 1) measure the performance norm (particularly their work plan reliability (WPR) norm) of a project and neighborhood combining SNA and conformity concept, 2) demonstrate how conformity impacts the performance of crews through their relationships, and 3) identify the role of social network in conformity occurrence in construction projects. The results of two case studies are also discussed.

1.2.3 Research Objective 3: To Identify the Improvement Direction (Benchmarks) for Inefficient Construction Crews

Social learning is another important social network influence that refers to the transmission of learned information from one agent (can be an individual, organization, group etc.) to another through a number of mechanisms (Kendal et al. 2009). Agents can learn not only directly from their own experience, but also indirectly from observing the behavior of others and the outcomes of those behaviors. For example, a car manufacturing plant may improve its performance by applying a new practice developed at its sister plant. A hotel can utilize the experience of other hotels to increase its knowledge of effective customer service (Argote et al. 2000).

Learning from others can happen in any small or large social context (such as organization or workplace) through observation and socialization; a construction project is not an exception. Besides the inherent tendency of humans to learn from others’ experiences (Bikhchandani et al. 1998), two more specific reasons can be identified for the construction crews to sometimes learn from each other’s practices. First, they need to continually improve their management skills and performance to exist and emulate in the dynamic market of construction (Luu et al. 2008). Second, to achieve the highest performance level by a crew, learning the best practices from the best-performing crews can be a quicker, simpler, and often more reliable way compared with the traditional way (utilizing its own experience, identification of problems, analyzing the available solutions, investigating the right possible practices). Learning from others reduces the cost of individual learning and avoids the risk of false identifications of problems and effort and risk of trial-and-error learning.

Interviews with experienced project and site managers and construction trades supervisors indicate that the interdependencies among the crews provide many learning opportunities in a jobsite. They believe that although the nature of the works and operations may be different from one crew to another, they follow the same objectives of maximizing the productivity, minimizing the cost, and delivering the work on time, with an acceptable quality, and without injuries or accidents. Thus, all of them deal with many similar challenges every day such as developing a safe work place, monitoring the laborers’ performance, making a weekly work plan, controlling cost and time estimation, training the workers, and preparing reports. The similarities of the crews’ objectives and challenges in one hand and the previously mentioned reasons on the other hand bring about the opportunity for the crews to observe, learn from, and implement the other crews’ practices in their own decisions and practices. Selecting a proper method for safety enhancement or implementing a low-cost strategy to meet the equipment emission standards of the EPA can be the examples of those decisions made based on the practices of other good performers.

Based on social learning phenomenon, this research proposes a procedure that combines DEA (assessing the relative efficiency of DM units) and SNA to identify the improvement direction (called benchmarks in this research) for the inefficient crews in a construction project. The procedure is called Social Network based Data Envelopment Analysis (DEA) Benchmarking Procedure (SDBP). The implementation of the proposed procedure is shown through a case study.

1.2.4 Research Objective 4: To Investigate the Cost/Benefit of Low or High Reliable Construction Crews and Develop a New Educational Version of Parade Game

Work plan variation (also called variation in this research), defined in this research as the time difference between what was planned and what occurred, results in unstable work flow in construction projects (a flow with unpredictable production). Previous studies demonstrated that how the unstable work flow can affect the performance of the construction crews negatively and reduce their productivity (for example, see Thomas et al. 2002, González et al. 2010, Liu et al. 2011, Wang et al. 2012). A reliable work plan is a work plan without variation. A work flow can be managed effectively by promoting work plan reliability through better planning strategies (Tommelein et al. 1999, Ballard 2000, Thomas et al. 2002, Liu et al. 2011).

Variability in this study is defined as the deviation of production capacity from an expected average. When a production capacity is reliable, it means that it is within the expected range (no variability). The Parade Game is a construction management educational tool demonstrating the effects of interdependency and variability in construction projects. This research extends the original version of the Parade Game by developing a simulation model that uses probability distribution functions (PDFs) instead of the die to determine the variability level, is able to get different variability levels at different stations, and analyze the outcome from cost standpoint (in addition to the production standpoint). The aim is to first identify the consequence of having crews with high or low reliable capacity at early, middle or late stations and then, explore how the capacity reliability improvement (variability reduction) can benefit the production and cost of the trades and project.

To summarize, by the first objective, the study focuses on development of the jobsite social network and its immediate area-sharing impact on the crews. The research concentrates on social aspects of existing jobsite interdependencies through the second and third objectives where it explores the impact of two main social network influences, i.e. conformity and social learning, on construction crews. The fourth research objective is to zoom in to the task-sequence dependencies of crews to investigate the engineering aspect of the relationships of the crews with different variability levels and its impact on project production and cost.

1.3 Research Scope

In construction there exist multiple participants from the beginning to the completion of a project. Owners, architects and engineers, general contractors, sub or trade contractors, suppliers and all their project managers, superintendents, and foremen/supervisors contribute to complete a project. During construction, many official/unofficial relationships are made among these actors. In addition, the complex nature of construction processes involves the actors in many inevitable interdependencies. Although they are all important areas for research, the concentration of this research is on the existing interactions/relationships/interdependencies among the various specialty crews in a construction jobsite.

The existing interactions/relationships/interdependencies between two construction crews can be based on some personal or work-related reasons (personal such as “friendship”, work-related such as “sharing the material or space”). This research concentrates on the work-related interactions/relationships/interdependencies, although it is acknowledged that the personal reasons would be the basis of some existing relationships between crews. This is because exploring and analyzing personal types of relationships is a huge research area in the field of behaviorism and sociology and out of the scope of this research.

1.4 Research Significance

Illustration and analysis of social network of construction crews during construction has not been well documented in industry handbooks or academic research work. This research looks at interdependencies and relationships of construction crews from an exclusively new analytical perspective. Understanding the underlying social network of the construction site will help project managers to control it through better planning and leadership, consequently increasing the jobsite productivity. The process of network development and analysis is explained step by step, so it can be repeated in any project.

The construction industry lacks a sound understanding of the social aspect of work plan reliability. All the previous research focused on the task-dependency impact of the reliability. This research, for the first time, proposes social network based approaches to quantify two of the most important social network influences, i.e. social conformity (Objective 2) and social learning (Objective 3).

Some research exists on benchmarking in the construction industry, most of the researchers attempted to find the best improvement direction at the level of construction companies (companies learn from each other) while less attention has been paid to the topic of benchmarking at the level of construction site crews in a single project. The combination of the Data Envelopment Analysis (DEA) and SNA is discussed for the first time (Objective 3) as a network-based benchmarking procedure for construction crews.

This study, through Objective 4, enhances the educational value of the Parade Game. The simulation model, as a user-friendly educational tool, can be used by construction educators/students to help them comprehensively understand all the aspects of variability, variation and reliability in a construction production system. The simulation model inputs (like cost and variability inputs) are adjustable, so the model provides the opportunity to the construction educators/students to create and examine their own scenario(s) and explore the outputs.

1.5 Research Outline

The flow chart in Figure 1.5 demonstrates an overview of the research. The literature review served to identify what research has been done in terms of social network concepts, analysis and application in construction industry. Construction performance benchmarking through DEA, production capacity variability/reliability, work plan variation/reliability and the research pertaining to the Parade Game enhancement are also reviewed comprehensively. Four different construction projects were studied to test the applicability of the proposed approaches. General project information, crews’ performance report and work plan progress report in case studies were collect from the projects. The overall research methodology was divided into four separate sections, corresponding to the four research objectives. More details about the analysis techniques used in the methodology and also the case studies are discussed in the Research Methodology and Results sections.

illustration not visible in this excerpt

Figure 1.5. Research design flow chart

CHAPTER 2 RESEARCH BACKGROUND

2.1 Social Network Analysis

SNA is a mixture of sociology, IT, and mathematics (Pryke 2012). SNA, introduced by Moreno (1960), has been known as a methodology to determine the conditions of social structures by investigating the interactions, relations and interrelationships of a set of actors (nodes) (De Nooy et al. 2005, Park et al. 2011). A social network refers to a pattern of ties that exist among different entities (nodes) such as countries, states, organizations, etc. The ties matters in social network studies, because they can affect the entities (nodes) as they transmit behavior, information, attitude, etc. SNA provides the methodology to analyze the relationship and conceptualize the social networks (De Nooy et al. 2005). While classic SNA research has concentrated on sociological networks, it has been applied to many research fields (such as aerospace equipment, automotive bodies, and computer and office equipment) with the goal of investigating various relationships among organizations and individuals (Park et al. 2011).

The social network concepts of cohesion, density, distances, and relationships are currently being applied by researchers in many diverse and distinct domains. Classic SNA research focuses on sociological networks involving individuals in the workplace and their exchange of information to complete tasks (Chinowsky et al. 2011). In the context of specific domains, social networks are being analyzed in many areas such as academic institutions, learning and innovation, and political connections (Chinowsky et al. 2010). Work in network visualization techniques is providing researchers with the ability to isolate relationships and visualize network principles such as dominance, centrality, and egocentricity, and graphically present results that were previously limited to mathematical matrices (Chinowsky et al. 2011).

2.1.1 Social Conformity: Tendency to Match Group Norm

Conformity, also known as herding behavior, is one of the important influences of being in a social network. It refers to the tendency of changing behavior (or performance) in order to fit in a group and converge to the group norms (Asch 1951). The group norms emerge out of interaction with the others in the social network. A wide variety of research shows that the behavior of others in our social environment shapes our response to a situation. Indeed, watching others in the social network provides information about the normal behavior in different novel or ambiguous situations (Cialdini and Trost 1998, Javarone 2014, Buechel et al. 2015).

In recent years, much further attention has been paid to conformity in different fields. For instance, Wang et al. (2014) studied college students’ classroom behavior from conformity point of view. Or Javarone (2014) studied the effect of social conformity in opinion dynamics by making a model based on the majority rule voting. They found that Conformity strongly affects opinion dynamics and conformist agents play the role of stabilizers in fully-connected networks. In another research, Buechel et al. (2015) demonstrated how conformity affects opinion leadership and the quality of information aggregation and wisdom.

According to the best of our knowledge in the relevant literature, there is not any research found on exploring the social conformity impact in construction project.

2.1.2 Social Learning: Learning from Others

Social learning, as one of the important social network influences, focuses on the learning that occurs within a social context. It considers how people learn from one another, encompassing such concepts as observational learning, imitation, and modeling (Vives 1996, Banerjee 1992, Gubanov et al. 2011). Indeed, people do learn from other people in the network, most particularly from their actions to avoid the costs, in terms of effort and risk, of trial-and-error learning and leapfrog directly to adaptive behaviors (Vives 1996, Henrich and Boyd 2001). Buying the most popular brands, patronizing well-attended restaurants by tourists, and purchasing bestsellers by readers are examples of social learning (Vives 1996).

The researchers, from various fields of study, discussed how learning by observing the past decisions of others (social learning) can help explain puzzling phenomena about human behavior. For instance, Ellison and Fudenberg (1993) studied how economic agents decide which of the two technologies to use when the relative profitability of the technologies is unknown. Bala and Goyal (1998) developed a general framework to understand how the structure of neighborhoods in a society affects the generation of information as well as its social dissemination. Bikhchandani et al. (1998) discussed the application of social learning in various fields such as laboratory experiments, business strategy, consumer marketing, crime and enforcement, politics, and medical practices. Argote et al. (2000) discussed social learning and knowledge transfer in organizations. Thornton and Thompson (2001) studied learning from other’s experiences in World War II shipbuilding. Lamberson (2010) analyzed a model of social learning in a social network in the case of deciding whether or not to adopt a new technology. McFadden and Train (1996) investigated the implications of learning from others on the sales of new products and the impact of advertising.

According to the best of our knowledge in the relevant literature, there is not any construction research found on social learning in construction projects.

2.1.3 Construction Research Pertaining to Social Network Analysis

SNA has become important within the engineering and construction field recently due to significant attention to some concepts such as trust and communication between project participants (Chinowsky et al. 2008). Wambeke et al. (2012) believed that an underlying social network of trades exists in a construction project and its recognition can contribute to project success by helping the construction site managers to coordinate the trades effectively. Pryke et al. (2011) believed that although project management research increasingly turns to the social sciences for the interpretation of issues related to construction, SNA’s use in construction research is still rather limited.

Application of SNA in construction project management research is mostly dedicated to the project team information exchange and communication. Chinowski et al. (2008) attempted to reduce the uncertainty during construction by network modeling of the information passed through the team members. Wong et al. (2010) investigated the differences between global and domestic projects through robust project network designs. In other research endeavor, SNA is suggested for use with learning dynamics and building information modeling (BIM) (Taylor et al. 2009). Chinowski et al. (2010), through some case studies, demonstrates the need to introduce the social network model into project organization development. Chinowski et al. (2011) used SNA to assess project effectiveness by focusing on the alignment of actual stakeholder knowledge exchange with knowledge exchange requirements defined by task relationships. Park et al. (2011) investigated the formation and impact of construction firms’ collaborative networks for performing international projects, using an SNA approach. Dogan et al. (2013) attempted to assess the coordination performance of a construction project based on the centrality measures of e-mail communication network.

The research pertaining to the jobsite social network, where actors are construction crews or subcontractors and they are connected (have relationship) in the network if they physically work in the same area(s) at the same time, is very limited. Wambeke et al. (2012) outlined a procedure to identify the organizational social network of construction trades and determine its key members. They provided a detailed description of the steps to create a social network of trades, as well as how to determine the key members of the network using degree and eigenvector centrality. In 2014, Wambeke et al. coupled the variation analysis with the associated social network of trades to create a decision making system. They determined the causes of variation that pose the greatest risk of impacting project performance, then analyzed the social network of trades to develop a decision support system to target trades in an effort to reduce variation. Lin (2014) studied the underlying job-site management problems and potential technology interfaces by analyzing job-site social networks and found that the order-management network has the highest degree of social density. Priven and Sacks (2015) explored how Last Planner system strengthens the project social network by building relationships among construction teams.

Table 2.1 summarizes the main social network research done in construction project management field.

Table 2.1. Summary of previous construction research pertaining to social network analysis

illustration not visible in this excerpt

2.2 Benchmarking through Data Envelopment Analysis (DEA)

Positive aspect of learning from others’ experiences (social learning) can be interpreted as benchmarking and learning from the better performers (benchmarks). Benchmarking is defined as a process of assessing and comparing to determine ways to improve processes and reach higher performance. It provides an objective evaluation for an organization to measure the performance of its processes and see if and how they can be improved (Lai et al. 2011). One of the common techniques of benchmarking, introduced by Charnes et al. (1978), is DEA, which measure the productivity by the ratio between a weighted sum of inputs and a weighted sum of outputs. It uses operations research techniques to automatically calculate the weights assigned to the inputs and outputs of the production units being assessed (Kabnurkar 2001). DEA generates a surface, called “empirical frontier”, which connects the relatively best decision making units (DMUs) (the most efficient ones). The empirical frontier can be used as a reference for efficiency improvement.

2.2.1 DEA in Construction Industry: Identify Performance Benchmarks (Role Model)

DEA has been utilized in construction in the last decade, mostly to explore optimum direction and performance enhancements. El-Mashaleh (2003) used DEA to evaluate the firm-level performance of construction contractors. Pilateris and McCabe (2003) developed a DEA model for financial benchmarks of the construction industry. McCabe et al. (2005) combined their improved DEA with contractor prequalification benchmarks for the “practical frontier”. Vitner et al. (2006) compared project efficiency in a multi-project environment based on DEA. El-Mashaleh et al. (2006) made use of DEA to quantify the impact of information technology on contractors’ performance. El-Mashaleh et al. (2007, 2010) advanced the DEA trade-off analysis based on the STs’ performance metrics and reallocated resources according to overall performance.

2.3 Production Capacity Variability/Reliability and Construction Work Plan Variation/Reliability

This section review the previous research studies on the production capacity variability and work plan variation.

2.3.1 Unstable Work Flow, Consequence of Work Plan Variation and Production Capacity Variability

Production capacity reliability simply means to make sure that adequate production capacities (resources such as labor, equipment and material) are available at the right time and the right place based on the planned task schedule (Thomas et al. 2002). A reliable production capacity means that there is no variability. When the production capacity of a crew is not reliable, the variability spreads out through the project work flow in the form of variation and increase instability of the work flow. Studies showed that unstable work flow has negative impact on labor productivity, schedule control, and project cost (Ballard 1993; Tommelein et al. 1999; Alarcón and Ashley 1999; Thomas et al. 2002; Linhard 2014 among others). Unforeseen conditions, lack of labor, equipment, or material, or under/overestimate the amount of work that is required to complete the tasks are common factors reducing the reliability of a work plan (increasing the variation) (Thomas and Flynn 2011, Hsieh 1998).

Reliability in production (eliminating variability) is an essential element for improving project performance (González et al. 2010, Thomas et al. 2003). Previous research showed that reliability in production and work plan will result in a project productivity increase. Ballard (1994), by focusing on reliability improvement of construction work plans, created Last Planner System® (LPS), which is successfully implemented in construction project (Ballard 2000, Ballard and Howell 1998, Kim and Jang 2005, Liu and Ballard 2008)). Howell et al. (2001) suggested that productivity can increase up to 30% when reliability of work plan improves from 50 to 70%. Thomas et al. (2003) examined whether improving the reliability of the work plan improves the construction productivity. They concluded that more effort devoted to improving the reliability affects cost and schedule performance positively. González et al. (2007) conducted a research to understand how changes in work plan reliability levels impact the project performance during construction phase. They, through statistical analysis, found that performance will improve when reliability is enhanced. Liu and Ballard (2008), through a pipe installation case study, demonstrated that reducing variability enhances labor productivity significantly.

2.3.2 Parade Game: Demonstrate the Impact of Variability in Construction Project

The existence of parades of crews is common in construction projects where a large number of specialty trades usually perform their tasks in a repeating and continuing work sequence. The Parade Game, created by Greg Howell and presented by Tommelein et al. (1999), is used as a lean construction educational tool to demonstrate the impact of variability on the performance of construction crews and their successors when they work in a task sequence and the production output of one crew is prerequisite to task of the next crew. The original version of the Parade Game was played based on the seven players, lining up in sequence. Each player, representing a subcontractor’s crew, is supposed to perform an activity of processing units repetitively (totally 100 units need to be processed). Players are given a similar die, and the capacity of each crew in each week is determined by rolling the die. After the capacity of a crew is determined by rolling the die, it will pass the number of units equal to the number on the rolled die. If the crew does not receive enough units from its predecessor (i.e. player does not have enough units in his/her buffer), it passes whatever exists there (the smaller number of units passed along). The project (game) is complete when all 100 units pass the last player. The game shows that it is possible to shorten project duration and increase the throughput by reducing the variability in production capacity of crews.

During the past years, a few studies have been done to enhance the original version of the Parade Game. Han and Park (2011) believed that the Parade Game will be more realistic if it includes the managerial actions taken during construction to alleviate the impact of variability. Based on this, they developed a Parade Game model that allows users to work overtime or have higher production capacity (if needed) as the two managerial actions. They concluded that their model can help in better understanding of variability and interdependency in the dynamic environment of construction processes. Lindhard (2014) used simulation modeling to examine the characteristics of variability in two designs of production flow: linear sequence and network. He concluded that the design of production flow has a great impact on the consequence of variability. He also concluded that waste emerges between task or activity handoffs and the impact of variability depends on the duration of the activity in the project. In addition to these two research studies on the Parade Game enhancement, a few studies were also conducted attempting to increase the usability and increase the features of the Parade Game using computer modeling techniques (for example see Senior 2011).

2.3.3 Investment on Planning to Improve Construction Work Plan Reliability and Consequence

Shapira and Laufer (1993) defined three stages of planning for construction projects: pre-bid, pre-construction, and during-construction planning stages. Planning during construction can be defined as the production of budgets, schedules, and other detailed specification of the steps to be followed and the constraints to be obeyed in the execution of the project (Ballard and Howell 1998). In large construction projects, delivery is usually handled by numerous project teams or subcontractors with different task packages. Thus, the planning process during construction is complex and involves many participants. Several research efforts have examined the relationship of planning and outcomes. For instance, Dvir et al. (2003) examined the relationship between project planning efforts and project success. They considered definition of requirements, development of technical specifications, and project management processes and procedures as three aspects of planning. They found that the project success is positively correlated with the investment in requirements’ definition and development of technical specifications. Dvir and Lechler (2004) found that the quality of planning has a positive effect on the project efficiency and customer satisfaction. Wang et al. (2008) studied the pre-project planning of industrial and building construction projects and found positive relationships between that and project cost/schedule growth. In another research using artificial neural networks, Wang and Gibson (2010) found that projects with better pre-project planning are more likely to have a better project performance at completion. Most of studies focused on the impact of planning at the preconstruction stage and the research addressing the consequence of additional planning efforts carried out during construction is limited.

2.4 Identified Gaps in the Body of Knowledge

2.4.1 Identified Gaps in Construction Research Pertaining to SNA Application

The following gaps have been identified in using SNA to resolve the construction related issues:

1. The basic definition of the term “relationship” in sociology refers to information exchange; most of the previous construction research with regard to social network focused on the information exchange or communication (see Table 2.1). There is also some research focused on the formal (contractual) relationships of the parties in the project (for example, West 2014). However, the research pertaining to the spatial social network of the crews in the jobsite during construction, as an informal underlying structure of construction project, is very limited. The construction industry lacks an appropriate method to quantify the influence of jobsite network on the project actors.
2. In the previous studies of jobsite social network, the relationships among the crews over the course of the project have been presented by a single constant social network. Since the frequency and severity of their relationship fluctuate over the course of the project, there is a need to consider the dynamic aspect of social network. These gaps are addressed through Research Objectives 1 and 2.

2.4.2 Identified Gaps in Construction Research Pertaining to Benchmarking via DEA

The following gaps have been identified in construction research pertaining to benchmarking via DEA:

1. Although there are some research studies on “benchmarking” in the construction industry, most of the researchers attempted to find the best benchmarks at the level of construction companies (project manager level) while less attention has been paid to the issue of benchmarking and finding the improvement direction at the level of construction crews.

2. In reality, there are two difficulties for an inefficient crew to learn from/follow the assigned benchmark(s) which have not been explained in construction benchmarking publications. They are “availability” and “achievability” of the benchmarks. The real benchmarks should be “available” in such a way that the required learning environment for inefficient crews is provided. In other words, it should be considered that an inefficient crew is only able to learn from/follow the benchmarks effectively, when it has enough frequent relationships with them. It is also important for inefficient crews to set their targets not too far to reach, so that they will be “achievable”. In other words, it should be considered that since it is not feasible for an inefficient crew to achieve the highest level of efficiency by a single movement, it is much easier for it to traverse this way step by step and through setting reasonable targets.

These gaps are addressed through Research Objective 3.

2.4.3 Identified Gaps in Construction Research Pertaining to Capacity Variability/Reliability and Work Plan Variation/Reliability

The following gaps have been identified in construction research pertaining to capacity variability/reliability and work plan variation/reliability:

1. The previous research examined the impact of different levels of work plan reliability on project performance, mostly measuring the impact on the subsequent crews in the project task sequence (for example, see Tommelein 1999, Ballard 2000, Thomas et al. 2003, Gonzalez 2007 and 2010). The research pertaining to the social aspect of crews’ work plan variation/reliability is very limited. The impact of social network influences such as conformity and social learning on the work plan reliability and variation of crews in the scope of a construction project was not found in industry handbooks or publications.
2. There is limited research pertaining to the enhancement of the original version of Parade Game. Previous studies (either the original version of Parade Game or the following enhancement versions) assumed that variability is the same for all the crews (i.e., players play with the same die) in order to reduce the complexity of the game and conclusion. However, in reality, variability can be varied from one crew to another, depending on how much their production capacity is reliable (Ballard 2000, Howell and Liu 2012). Furthermore, the consequence of variability reduction, or in other words, production capacity reliability improvement, by making additional effort on planning has not been studied in the previous research.

These gaps are addressed through Research Objectives 2 and 4.

2.4.4 Addressing the Identified Gaps

This research provides a unique and numeric perspective by addressing the gaps in the body of knowledge identified above. A systematic approach is describing to quantify the influence of jobsite social network on the crews’ performance from different aspects, exclusively looking at the interdependency and variability of construction crews over projects.

CHAPTER 3 RESEARCH METHODOLOGY

This chapter of the dissertation details the design of the work effort undertaken to deliver those objectives. Different case studies, data collection, analysis and modeling techniques are used to achieve each of the objectives of this research. Therefore, this chapter is divided into the four sections corresponding to the four research objectives.

3.1 Methodology to Achieve Research Objective 1, To Develop the Jobsite Social Network of Construction Crews and Quantify its Impact on Performance (a Case Study)

Figure 3.1 shows the different steps of the methodology to achieve the first objective. The research uses a case study (Case Study 1) to show the applicability of the proposed approach. A general contractor (GC) overseeing 43 subcontractor crews, involved with the construction of a 150,000 square foot data center participated in the case study. A GC with numerous subcontractor crews was chosen because the research is focused on the social network that exists among the various crews, thus there was a desire for a study a project with more than just a few subcontractors. The $50M project entailed the build-out of an existing warehouse building into a data center and white space computer labs. Construction ran from February through September 2010 and the project was studied from the beginning of March through completion at the end of September. There were nearly 1200 tasks performed by the 43 various crews working on the data center during the course of this 28- week study.

illustration not visible in this excerpt

Figure 3.1. Methodology steps to achieve Research Objective 1

1. Construct jobsite social network of subcontractors based on working space: The social networks are constructed based on spatial proximity due to the desire to examine which crews were physically working together in a shared physical space. A social network consists of a set of vertices and ties between them. The crews are defined as the vertices and the ties among them indicate that if they are going to work in the same space at the same time. The required data for constructing a social network is collected and tabulated by an adjacency matrix (ASN). Two subcontractor crews are called adjacent/neighbor if they are connected by a tie in the network. Table 3.1 shows a schematic ASN. ASN represents the existing relationships, where the elements “aij“ and “aji“, called “line value”, indicates how frequent the relationship or how strong the social network tie is between Subcontractor Crews i and j. ASN can be symmetric (aij = aji) or asymmetric (aij ≠ aji) depending on the purpose of social network development. In the methodology implemented for the first objective, the networks are assumed to be asymmetric, because it is assumed that the influence of Crew i on j may not be necessarily as the same as the influence Crew j has on i.

Table 3.1: Schematic adjacency matrix (ASN) for relationships of subcontractor crews

illustration not visible in this excerpt

In the proposed method, instead of making a single social network for all the existing spatial relationships of the project, several networks are created; each represents the crews’ interferences for a time interval. Since the construction managers usually make their work plan on a weekly basis, a network is made for each week of the project to track the interdependencies with more accuracy. Thus, if two crews are supposed to work in the same working area in a week, there will be a tie between them in the jobsite social network created for that week.

As noted, the social networks used for the first objective of the research are asymmetric, since the influence of one on another (sending influence) in the network is not necessarily the same as the influence it gets (receiving influence). Directed reciprocal lines were used to establish the social networks for the proposed method. Two different networks are created, one based on the “frequency” and the other based on the “severity”, to accurately estimate the impact of the networks. In the frequency network, the numbers on the ties (weights) show the number of tasks each of the crews perform, while in the severity network, the weights indicate the summation of tasks’ duration for each crew. Figure 3.2 depicts a schematic example of the jobsite social network. Assuming it has been created based on the work plan of the first week of a project, it shows that four crews (A, B, C, and D) out of a total of six crews in the project are going to perform their tasks in the first week of the project. A tie between the two crews shows that they are going to share the working area to perform their tasks at the first week of the project. Figure 3.1.a shows the frequency network where the weights indicate the number of tasks each of them performs. For instance, the tie between A and C shows that the Crews A and C perform 1 and 2 tasks respectively in the same working area in Week 1. In the severity network (Figure 3.1.b) the weights indicate the summation of the tasks’ duration. For instance, the “3” and “10” on the tie between A and C show that Crew A’s task take 3 days to complete while the summation of the tasks’ duration for the Crew C is 10 days in Week 1. The tie weights represent the magnitude of influence the crews are going to have on each other in that time interval. It should be noted the tasks performed by various crews may not be equal (there may be different operations, and therefore, different equipment, workforces and materials). However, it was assumed all the tasks to have the same magnitude of influence in this method. There are usually some boundaries for construction crews for their task breakdown. For instance, they may break their tasks down into the activities with the maximum duration of one week and the maximum cost of $10,000. These kinds of boundaries/scopes in defining the tasks lighten, if not eliminate, the inequality impacts of the crews’ operation. When there are no boundaries/scopes for making work plan, the weights can be adjusted according to the tasks’ characteristics to balance the inequality of the crews’ operation.

illustration not visible in this excerpt

Figure 3.2. Schematic jobsite social network based on working space

2. Conduct centrality analysis to calculate network frequency and severity centrality ranks: “Centrality” measures the relative importance of the vertices within a network. There are various ways to measure the centrality such as degree, betweenness and closeness centralities (De Nooy et al. 2005). In this research, Weighted In-degree Centrality is used to measure the receiving influences by each crew. Degree centrality of a node in the network is the number of its ties; in-degree centrality is the number of its incoming (receiving) ties (Equation 3.1); and the Weighted In-degree Centrality is the summation of the weight of its incoming ties (Equation 3.2) (Opsahl et al. 2010):

illustration not visible in this excerpt

In Equation 3.1, [illustration not visible in this excerpt] is the node in-degree centrality for a binary network (the strength of all ties is assumed to be equal), i is the focal node, j shows all other nodes, N is the total number of nodes in the network, and x is the adjacency matrix of a binary network (ݔି௝ is 1 if there is a tie from Node j to i).

illustration not visible in this excerpt

In Equation (3.2), [illustration not visible in this excerpt] is the weighted in-degree centrality for a weighted network, and w is the weighted adjacency matrix [illustration not visible in this excerpt] is greater than 0 if here is a tie from Node j to i and the value indicates the tie strength).

The weighted In-degree centrality of a crew represents the frequency and severity of the tasks performed by the other crews (neighbors) in the working area(s) the crew works (i.e., workload in the neighborhood). It indicates how much influence the crew receives from the network. For the frequency network shown in Figure 3.2.a, the weighted in-degree value is 2 for Crew A and 4 (=3+1) for Crew B. These numbers are 10 and 7 (=5+2) for the severity network (Figure 3.2.b) respectively.

3. Measure work plan variation of subcontractors: Work plan variation (also called variation in this research) is defined as the time difference between what was planned and what occurred. It represents the reliability of work plan. A reliable work plan is a work plan without variation. Variation is one of the root causes of productivity loss (Howell et al. 1993, Ballard et al. 2005, Wambeke et al. 2011, Hajifathalian et al. 2012). Four variables (Equations 3.3 to 3.6) are measured for each of the crews at each week over the course of the project as the indices of work plan variation (showed by “V” in the equations). By measuring these four variation variables the performance level of crews was measured from both variation frequency and severity standpoints.

illustration not visible in this excerpt

4. Calculate the variation ranks for each of subcontractors: Each subcontractor crew is given a rank based on his/her variation variables’ value (totally four ranks based on the four variation variables).

5. Run correlation analysis between variation ranks and centrality ranks: In order to investigate the impact of jobsite social network on the performance of the crews, the consistency, using correlation analysis, between the influences each crew receives from the networks and its work plan variation is evaluated over the course of the project. The jobsite social network impact on the crews is measured by “centrality” (see Step 2). Work plan variation is used to indicate the crews’ performance (See Step 3). The correlation analysis is conducted to calculate the degree of association between the two (jobsite social network centrality and work plan variation).

3.2 Methodology to Achieve Research Objective 2, To Examine the Impact of Social Conformity on the Performance of Construction Crews

Figure 3.3 shows the different steps of the methodology implemented to achieve the second objective. The proposed approach will demonstrate the impact of conformity, as one of the important social network influence, on the crews’ performance.

This research studied two projects to show how the proposed approach can be implemented in a real project. Results will demonstrate how the subcontractor crews’ performance over the course of the project can be interpreted and visualized through social conformity influence. First case (Case Study 2) involved a large general contractor working on a two level, $211 million, 350,000 square foot facility center over six main subcontractor crews and several small subcontractors (for short period and small tasks) on the job. The project was ground up with a steel structure, precast walls, and complete build out of data hall suites and administration areas. The required data were collected from the six main subcontractor crews of the project (Mechanical, Electrical, Plumbing, Drywall, Painting, and Fire Protection) over 15 months. Second case (Case Study 3) is the same as the project studied for the first objective, but the focus here is on only the main subcontractor crews of the project. As mentioned in the previous section, the project was a $50 million, 150,000 square foot construction project entailed the build-out of an existing warehouse building into a single level data center and white-space computer labs. The project included a general contractor overseeing multiple specialty subcontractors. For the purpose of Objective 2, seven main subcontractor crews of the project (Mechanical, Electrical, Concrete, Drywall, Painting, Steel Fabrication and Fire Protection) and their performance and interdependencies were studied. Construction ran from February through September 2010 and the project was studied from the beginning of March through completion at the end of September. The first three weeks of the project were not included in the study due to the limited scope and number of subcontractors involved during that period.

illustration not visible in this excerpt

Figure 3.3. Methodology steps to achieve Research Objective 2

In both cases, the study were limited to the main subcontractor crews (instead of including all the subs in the project) because 1) they continually perform the work, so enough data was available, and 2) they have enough relationships and interdependencies with each other to constitute a social network and be affected by the conformity influence. Both cases were selected from non-residential/commercial construction category because 1) GCs in commercial project usually subcontract most of the work due to the project size and complexity (Costantino et al. 2001, McCord 2010); thus, the number of subcontractors and their relationship are higher, which helps in better demonstration of the analysis and results, 2) the subcontractors in commercial construction are usually big enough to bring their own labor, purchase their own materials, and run their own equipment; therefore, they have the power of controlling/adjusting their production capacity when they find the project unreliable, and 3) they usually work on multiple projects concurrently; thus, they can move their resources among their projects to increase their profitability (Sacks 2004, Thomas and Flynn 2011, Freeman and Seppanen 2014). However, the approach can be implemented in any other project type. In the following paragraphs the approach will be explained step by step.

1. Measure Work Plan Reliability Indices of Subcontractors: As discussed, a reliable work plan is a work plan without variation. Variation is defined as the time difference between what was planned and what occurred. One of the common ways to measure the WPR is percent plan complete (PPC), a ratio between actual completed activities and planned activities of Work Plans (Ballard 2000). In this research, the basic measurement of the WPR is expanded to include the reliability measurement of both tasks’ start and completion. Four variables (Equations 3.7 to 3.10) are measured for each of the subcontractor crews at each month over the course of the project as the work plan reliability indices (WPRIs). WPR has inverse relationship with variation (Equations 3.3 to 3.6); the more variation means less WPR. By measuring these four indices the WPR level of the subcontractor crews is quantified from both frequency and severity standpoints. Note that sometimes the delays happen due to something out of the control of the subcontractors; those delays that have the legitimate excuses were excluded from the calculation to have a true measurement of WPR.

illustration not visible in this excerpt

2. Identify WPRI Norm of Project: As explained in Research Background section, norms are behaviors (can be performance, beliefs or any features) that become standards in a society, organization or group of people. Norms in a group gradually formed based on the behaviors of the group members and can change over time (Fehr and Fischbacher 2004, Lapinski 2005). In this study, two versions of norm were defined and studied, first is based on the global environment (project) and then based on the local environment (neighborhood). The first version (project or global norm) does not take the social network of the subcontractor crews into account, while the second version (neighborhood or local norm) includes only the neighbors of the subcontractor crews (their social network) in calculation of the norms. The goal of defining two versions of norm is to first identify the impact of social network of subcontractor crews on their conformity, and second realize which one can explain better the conformity phenomenon in the project case studies. WPRI norm of a project at Month m is the average of the WPRI of all the subcontractor crews from beginning of the project (Month 1) to Month m-1 (see Step 4 for the WPRI norm of neighborhood):

illustration not visible in this excerpt

In Equation 3.11, WPRIPk,j is the average of the kth WPRI of all the subcontractor crews of the project at Month j, WPRIk,j,i is the kth WPRI of the ith subcontractor crew at Month j, and Nj is the total number of the subcontractor crews working in the project at Month j. In Equation 3.12, ProjectNormk,m is the project norm (global norm) for the kth WPRI of the subcontractor crew at Month m (it takes the WPRI of all the subcontractor crews into account from beginning of the project to Month m-1). m is the project duration in month. The WPRI measurement of the current month is not included in the norm calculation, because WPR norm for a subcontractor crew is a result of his/her observation of the past (and not current) behavior of other subcontractor crews.

3. Construct Social Network of Subcontractors: Construction of social network was explained in Section 3.1. The social network construction method is as the same as the one used for the first objective (Section 3.1), however, since our purpose here is to measure the level of interdependency between subcontractor crews (and not their reciprocal influence on each other), undirected symmetric network is used to represent the social network. The subcontractor crews are defined as the vertices and the ties among them indicate their interdependencies/relationships. The assumption is that there is a social tie between two subcontractor crews if they work at the same time in the same area in the project. The element “aij”, or “aji” (symmetric ASN) represents the number of times Subcontractor Crews i and j perform their tasks in the same area at the same time (number of times they share the working area). Each social network represents the relationships among the subcontractor crews in a given month of the project.

4. Identify WPRI Norm of Neighborhood: As explained earlier, two versions of norm were studied in this research. The Project norm, as explained in Step 2, includes all the subcontractor crews of the project in the norm calculation (global view). Neighborhood or local norm includes only the neighbors of the subcontractor crew in calculation of the norm. Since the neighbors may change from one sub to another, the neighborhood norm for each subcontractor crew is unique at each month and is calculated separately. The WPRI neighborhood norm for a subcontractor crew at Month m is the average of the WPRI of its neighbors from beginning of the project subcontractor crew:

illustration not visible in this excerpt

In Equation 3.13, WPRINk,j is the weighted average of the kth WPRI of the crew’s neighbors at Month j, WPRIk,j,p is the kth WPRI of the pth neighbor of the subcontractor crew at Month j, wp,j is the line value between the subcontractor crew and its pth neighbor at Month j, and P,j is the total number of the crew’s neighbors at Month j. The influence a subcontractor crew gets from his/her neighbors is highly dependent to the strength and frequency of the relationship he/she has with them. Thus, the social network line values are included in the formula to adjust the neighborhood norm by enhancing the power of the frequent neighbors and decreasing the role of the infrequent neighbors. In Equation 3.14, NeighborhoodNormk,j is the neighborhood norm (local norm) of the kth WPRI of the subcontractor crew at Month m (it takes the WPRI of all the subcontractor’s neighbors into account from beginning of the project to Month m-1). m is the project duration in month. In contrast to the project norm, the neighborhood norm is unique for each subcontractor crew since the social network of each subcontractor crew is unique at each month. So, at each month, a single project norm and N neighborhood norms are developed corresponding to N subcontractor crews of the project.

5. Measure deviation from project norm (DFPN): To realize how much a subcontractor crew conforms to a norm the similarity between his/her behavior (indices) and the norm should be quantified. In sociology, any behavior that is against the social norms is called “deviance” (Cialdini and Trost 1998). Identically, in this research “Deviation” is defined as the difference between each of the subcontractor crew’s WPRI and its corresponding norm in order to measure the conformity level of a crew to the WPRI norm. Deviation is calculated monthly for both project and neighborhood norm. Deviation from project norm (DFPN) for each subcontractor crew at Month m is calculated as follows:

illustration not visible in this excerpt

Where DFPNk,m is the deviation from project norm for the kth WPRI of the given crew at Month m, ProjectNormk,m is the project norm for the kth WPRI at Month m, and WPRIk,m is the kth WPRI of the crew at Month m.

6. Measure deviation from neighborhood norm (DFNN): DFNN, the difference between the norm of the neighborhood and the crew’s WPR, for each subcontractor crew at Month m is calculated as follows:

illustration not visible in this excerpt

Where DFNNk,m is the deviation from neighborhood norm for the kth WPRI of the given subcontractor crew at Month m and NeighborhoodNormk,m is the neighborhood norm for the kth WPRI at Month m.

7. Conform to local or global?: DFPN and DFNN in each month indicate the deviation of a given subcontractor crew from norm in that particular month. To identify if the behavior of a crew for the entire project is based on the conformity, the trend of the deviation is examined over the course of the project. In Step 7, statistical analysis is conducted to investigate how the deviation from norm (both DFPN and DFNN) exists and changes over the course of the project. The objective is to find out if there is any meaningful trend for the deviation over the project time. If a subcontractor crew has a descending trend, i.e., negative correlation, it indicates that he/she reduces/increases his/her WPR to minimize the deviation from norm and sync with others as the project proceeds. In other words, the subcontractor crew conforms to the norm (conformist sub). Ascending or irregular trends show that the crew’s behavior is not particularly under the influence of the norms (non-conformist sub). Correlation analysis (Pearson, Kendall Tau, and Spearman) is implemented to quantify the relationship between the deviation from norm and the project time. Additionally, DFPN and DFNN are compared to identify the role of social network in conformity occurrence and to find that if subcontractor crews in this study conform to their neighbors more than the others.

3.3 Methodology to Achieve Research Objective 3, To Identify the Improvement Direction (Benchmarks) for Inefficient Construction Crews

Figure 3.4 shows the different steps of the methodology to achieve the first objective. To demonstrate the applicability of the proposed procedure, a case study was conducted (Case Study 4). The selected project is a typical residential construction project ($2.5M, duration: 6 month) located in North Carolina. 12 major specialty crews working on the project were studied. They are coded as CO: Concrete; WP: Wood & Plastic; FI: Finishing; PL: Plumbing; ME: Mechanical; TM: Thermal & Moisture Protection; DW: Doors & Windows; MA: Masonry; EL: Electrical; ML: Metal; SP: Special Work; and LA: Landscaping. A fast-track residential construction project was selected to illustrate how to implement the proposed approach. The methodology can be applied to other types of construction projects as well, as long as it involves coordination of multiple crews. The data can be collected at any time when the project is ongoing and the analysis results can be used as improvement directions in the rest of the project or in the similar projects in the future. The interview and the data collection in this research were done at the Week 10 of the project.

illustration not visible in this excerpt

Figure 3.4. Methodology steps to achieve Research Objective 3

The procedure includes eight steps. First, the input and output data needed for running DEA are collected, summarized and prepared. Then, the basic DEA is performed, including all the crews, to find out the benchmarks in the scale of the entire project (conventional DEA benchmarking where the existing relationships/interdependencies are not considered). Afterwards, the proposed procedure, Social Network based DEA Benchmarking Procedure (called SDBP) was implemented. It utilizes the DEA and SNA concurrently to tackle the limitations of conventional benchmarking (refer to Section 2.4) and identify the practical benchmark(s).

1. Determine Data of Inputs/Outputs: Since measuring the efficiency level of crews is needed, the input and output variables should be selected in such a way that they represent the performance level of each crew. Level of the seven precondition readiness variables are measured before execution of each task as the inputs. In construction planning, 7 types of preconditions need be established for a task to be executed smoothly (Koskela et al. 2002, Bertelsen et al. 2007). They are: weather conditions, equipment availability, labor availability, material availability, prerequisite work readiness, space sharing adequacy, and design and working method clarification. For the outputs, the performance of crews regarding work plan reliability and cost are collected.

2. Principal Component Analysis (PCA) on the DEA Inputs and Outputs: The aim of PCA is to simplify data by reducing the dimensionality from n variables to something much less, without much loss of information. Indeed, PCA, via doing eigen-decomposition (calculate the eigenvectors and eigenvalues) of the covariance matrix, takes a cloud of data points, and rotates it to fit in a new coordination system such that the maximum variability is visible. PCA is used to remove the correlation among the variables and reduce the dimensions of the variables in an optimal way such that the differences and similarities amongst the data are highlighted without much loss of information (for more information about PCA, see “Jolliffe 2004”). It helps provide better DEA results through simplifying and organizing the data using a few dimensions.

3.3.1 Conventional DEA Benchmarking

3. Run Basic DEA: The original method of DEA is known as CCR (Charnes, Cooper, and Rhodes (1978)), which maximize the ratio of weighted outputs to weighted inputs in such a way that the similar ratio in every DMU is less than or equal to one. Two main alternative approaches are available for the basic DEA model: input-oriented and output- oriented. The following DEA model used is an output-oriented Constant-Return to-Scale (CRS) DEA model in which the outputs are maximized and the inputs are kept at their current levels (Zhu 2009):

illustration not visible in this excerpt

Where DMUo represents one of the n DMUs under evaluation, and yr0 and xi0 are the rth output and ith input for DMUo, respectively. θ is the uniform proportional increase in the DMUo’s outputs and its maximum amount (θ*) is known as the DMUo’s efficiency score.

Technically speaking, the presented model states that if there is a weighting vector λ that solves the Equation (1), and θ* > 1, then it is concluded that the DMUo is inefficient. The optimal value to (1) is θ*=1, which shows that the current output levels cannot be increased proportionally and the DMUo is efficient and the input-output combination is optimized. Appendix D presents an example of DEA.

Using DEAfrontier© Software, the output-oriented DEA (Equation 3.17) is performed to identify the best-performing crews in the scale of the entire of the project. The DEA assigns an efficiency score to each of the 12 crews based on its performance level relative to its peers. Thus, one can determine how far each inefficient crew is from the bestperforming crews (benchmarks) with regard to the performance measurement.

4. Identify DEA benchmarks. Among those efficient crews, DEA assigns one or more benchmarks for each of the inefficient crews. The conventional DEA is done to compare the benchmarks assigned by the conventional method with the benchmarks obtained from the implementation of SDBP.

3.3.2 Social Network based DEA Benchmarking

The main goal of SDBP is to show how the existing relationships/interdependencies among the crews affect the benchmarking process in the construction jobsites. It is an analytical method to show who could be the effective leaders and who could be their followers in a project. As discussed in Section 2.4, there are two difficulties or limiting factors with the conventional DEA benchmarking method: “availability” and “achievability”. “Availability” states that it may not be possible for an inefficient crew to learn from/follow an assigned DEA benchmark(s), because its interaction/interdependencies are limited to only a few of the crews in the site and thus the required relationships with those benchmark(s) may not be made. Sometimes, even if the desired relationship exists, it may not be close enough to provide a learning environment for the inefficient crew. The second difficulty is “achievability” which declares that it is not always possible for an inefficient crew to achieve its target (being efficient) in a single step, when the crew is far from the assigned DEA benchmark(s). The proposed procedure, SDBP, was designed to overcome the two mentioned difficulties. In addition to utilizing the conventional form of DEA, it takes the crews’ relationships/interdependencies into account by establishing the underlying social network of the crews. The SDBP was established to assign the most practical (“available” and “achievable”) benchmarks to the inefficient crews. The SDBP includes four main steps:

5. Construct Social Network of Crews based on Communication: In the proposed method, the crews are defined as the vertices and the ties among them represent the frequencies of communications between the crews in the jobsite. The social network is symmetric and “aij” in ASN represents the number of times (frequency), the Crew “i" communicates weekly with Crew “j” based on possible reasons. Indeed, aij shows the strength of the relations between i and j, which is indicated by “Line Value” on the top of each tie (connection) in the network.

6. Measure the Relationship Score (RS) based on sub-SNs: To measure the “availability” of a crew to another, the focus is on the relationships/interdependencies of the crew under evaluation. To do so, a sub-social network (sub-SN) is drawn from the entire social network by considering only the relationships/interdependencies of the crew under evaluation. Then, to assess the relative availability of each crew to the crew under evaluation, the tightness of the relationships is measured. The more frequent relationships/interdependencies means the higher “availability” level. The relative “availability” of each crew is assessed in the sub-SN by calculating the RS:

illustration not visible in this excerpt

Where RSi, j is the Relationship Score between the crews of i and j, and (Line Value)i, j is a number showing the strength of the relationships between them. By obtaining RSs in the sub-SN, one can determine who has more/less frequent relations with the each of the inefficient crews.

7. Measure the Relative Efficiency Distance (RED) by DEA on the sub-SNs: In order to measure the “achievability” of a crew to another, the differences between their efficiency levels is calculated. The DEA (Equation 3.17) is performed for the established sub-SN of the inefficient crew and the relative efficiency score for each crew in that sub-SN is obtained. Then, the “achievability” level of each crew to the inefficient crew is assessed by calculating the difference between their efficiency scores in percentage. Thus, the original value of the efficiency scores is translated to the RED for each of crews in the sub-SN of the crew under evaluation:

illustration not visible in this excerpt

Where REDi, j is the efficiency distance of j relative to i. It can be inferred that the RED ranks the crews in a sub-SN based on their “achievability” for the inefficient crew under evaluation.

8. Identify the practical benchmark(s): Practical benchmarks are those benchmarks “available” and “achievable” enough for an inefficient crew, so it can learn from their practices. The greater the value of the RS indicates the more frequent relationship between the crews which provides a more reliable learning environment (the more “availability). The greater the value of RED indicates the target which is less achievable for the inefficient crew. However, if the RED is too small, there may not be enough things to learn, since the smaller the RED indicates the less difference between the two crews. The criteria to determine the effective RED may vary in each situation depending on the objectives. In this research, the RED value of +15% is set as the most effective RED level between the two crews based on the discussion with the project managers of the given case.

Now, the crews can be ranked in the sub-SN of each inefficient crew, based on their level of RS and RED. The crew with the highest RS level and the RED value closer to +15% can be the best practical benchmark for the inefficient crew under evaluation. To make it easier to identify the practical benchmark(s) a two-dimensional plot will be developed, which visualizes the concept.

3.4 Methodology to Achieve Research Objective 4, To Investigate the Cost/Benefit of Low or High Reliable Construction Crews and Develop a New Educational Version of Parade Game

Figure 3.5 shows the different steps of the methodology to achieve the first objective. Parade Game, as explained in Section 2.3.2, is used as a lean construction educational tool to show the impact of variability on the performance of construction crews and their successors when they work in a task sequence and the production output of one crew is prerequisite to task of the next crew. In Parade Game, each player, representing a subcontractor crew, is supposed to perform an activity of processing units repetitively (totally 100 units need to be processed). The fourth objective of this research is to first develop a new version of Parade Game that is able to get different variability levels at different stations, and analyze the outcome from cost standpoint (in addition to the production standpoint). Secondly, this research aims to identify the consequence of having crews with a high or low reliable capacity at early, middle or late stations and explore how the capacity reliability improvement (variability reduction) can benefit the production and cost of the crews and project. Instead of rolling a die (Original Parade Game is played with dice), the production capacity of each crew in each week is determined by predefined triangular PDFs with the average of 10. Using continuous PDFs makes the production capacity (and subsequently variability) determination closer to the reality since they do not have the irregularities of discrete functions such as die.

illustration not visible in this excerpt

Figure 3.5. Methodology steps to achieve Research Objective 4

1. Determine the Parade Game Model Inputs:

Variability Inputs: As discussed, variability in this study is defined as the deviation of the production capacity from an expected average (expected production average for each crew is assumed to be 10 units per week). No variability means the production capacity is reliable. Three triangular PDFs with the same average but different standard deviation (i.e. different variability) were defined to represents low, medium and high level of production capacity reliability. They respectively represent the crews with a low, moderately, and high reliable capacity. After reviewing related literature, left-tailed triangular PDFs are selected to make the model closer to the reality. Low, medium and high variability level are shown by TRIA (9,10,11), TRIA (7,11,12), and TRIA (5,11,14) respectively. The crew production capacity in each week is determined by the one of these PDFs. Since the value must be an integer, the model rounds the given number to the nearest integer. The triangular distributions used are common for construction tasks’ durations (AbouRizk and Halpin 1992). They are continuous, limited between two positive time intercepts, and with a unique mode in its defined range. Thus, they satisfy the basic prerequisites of PDFs used for construction simulations (Fente et al. 2000).The specifications of the triangular PDFs are shown in Figure 3.6. Standard deviation (SD) is a measure that is used to quantify the amount of variation. Relative standard deviation or coefficient of variation (CV) is a normalized measure of dispersion of a probability distribution and it can be calculated by dividing the SD by mean. Both SD and CV value represent how much variability exists in a production capacity.

illustration not visible in this excerpt

Figure 3.6. Triangular PDFs used to reflect various production capacity reliability (variability) levels in the model

Cost Inputs: Construction costs are usually categorized as indirect and direct costs. They constitute the main financial components of any construction item (Enshassi et al. 2008, Becker et al. 2014). Both subcontractor’s and GC’s cost consist of these components. Direct costs are the costs attributed to a single task of construction work, which is usually associated with a construction crew performing a task using specific materials and equipment. Indirect cost is simply all those costs which are not direct costs and cannot be attributed to a single task of construction. They include overhead, profit etc. (Knutson et al. 2009, Becker 2012). Planning cost, the efforts spent on making work plan and maintaining reliability, is also part of the indirect cost. Different levels of variability are result of different amount of efforts on planning.

The cost input data required for the model development were gathered through existing literature and face-to-face interviews with construction practitioners (such as foremen, project and field engineers and project managers) of local construction companies. Three cost-related items that were added to the model are:

a) Direct Cost: As noted, the Parade Game assumption is that the project is to be done 100% by the subcontractors, which are the seven crews in the work stations (zero percent self-perform). Direct cost (including labor, equipment, and material (LEM)) of processing a unit was assumed to be $C for any of the crews. In other words, every time a crew wants to process a unit, it needs to spend $C to have LEM available for the work. Note that the GC pays the subcontractor crews only for the units they process. Thus, the direct cost for a crew is calculated based on the number of production capacities he/she brings to the project to get the job done (based on capacity), while the direct cost for the GC (the amount GC pays to the crew) is calculated based on the number of units the crew processed in that week, no matter how many capacities he/she lost (based on unit). Note that the trade also includes his/her profit and indirect cost in the invoice to the GC. Therefore, direct cost for the subcontractor crews is going to be $C per capacity and for the GC $C per completed unit plus a constant percentage as the subcontractors’ profit (assumed 10% in the model) and indirect cost.

b) Indirect Cost: Literature reported that the percentage of the indirect cost to total cost ranges from less than 20% to as much as 40% depending on the type, complexity and location of the project (Ahuja and Campbell 1988, Assaf et al. 2001, Enshassi et al. 2008, Becker et al. 2014). The same range was obtained from the face-to-face interview with a few project engineers/managers working for local construction companies. To cover all the possibilities, the model is run with the three scenarios of 20, 30, and 40 percentage of the indirect cost to total cost.

c) Planning Cost: “Planning” has various levels in a construction project. The scope of planning in this research includes efforts toward scheduling and organizing the wherewithal within control and coordinating with other parties (in the project) for shared resources and/or interaction. The assumption basis is that more efforts spent on planning results in variability reduction (production capacity reliability improvement). What is needed as the model input for the planning cost are 1) the percentage of planning cost to the total project cost and 2) the cost associated with each of the production capacity PDFs which were defined according to the different variability levels (refer to Figure 3.6). According to the best of our knowledge, no research has been done to quantitatively address the relationship between the planning efforts carried out during construction and its outcome regarding to the variability. Thus, to use realistic data for the planning cost inputs rather than arbitrary numbers, a short survey was conducted to answer the following questions: 1) what is the percentage of planning cost to total project cost for construction subcontractors? and 2) what is the relationship (if any) between variability level of a subcontractor and his/her planning effort (planning cost)?

The survey was conducted through face-to-face interviews with practitioners (such as foremen, superintendents, and project engineers/managers) of different specialty subcontractors working for general contractor construction companies mainly located in North Carolina, US. In addition to the general information of project, the respondents were asked to specify the amount of efforts (in terms of man-hour per week) they usually spend on planning. Responses varied depending on the type and size of the project tasks. Then, they were asked to specify their production capacity in terms of minimum, most likely, and maximum rate, if they plan to complete 10 typical tasks per week (identical to production of 10 units in the game). In other words, they were questioned to identify their regular production capacity variability based on a 10 units/week (in average) production commitment. Then, using the minimum, most likely, and maximum values, a production capacity triangular PDF was built for each of the respondents indicating his/her crew’s variability. Examining the planning efforts versus the CV of the respondents’ PDF, the relationship between the planning efforts (cost) and variability were developed.

The respondents of the survey were site managers of construction projects (foremen, superintendents, and project managers) working on commercial construction projects for general contractors (43% of respondents) and trade contractors (57% of respondents) mostly in North Carolina. Based on the 81 collected survey responses, the percentage of the planning cost to the total project cost for a typical subcontractor ranges from 5 to 15% of the total cost, depending mainly on the type and size of the project and company. To cover all the possibilities, the model is run with the three scenarios of 5, 10, and 15 percentage of the planning cost to total cost.

Survey results generally verified that the variability of the subcontractors is lower when they spend more efforts on planning. Based on the developed relationship between the planning efforts and the variability, it was concluded, and subsequently assumed in the simulation model, that a trade with a moderately reliable capacity (medium variability) spends 21% less efforts than a trade with a high reliable capacity (low variability) and 33% more than a trade with a low reliable capacity (high variability).

2. Design Arrangement and Cost Scenarios:

Design of Arrangement Scenarios: By allocating different PDFs to each of the seven work stations, 2187 (3^7) different scenarios can be made. According to the research objectives, 9 scenarios were made based on the arrangement of the crews and analyzed. First, crews with a moderately reliable capacity are allocated to all the seven work stations (Arrangement Scenario 1 “M M M M M M M”). This scenario is used as the benchmark for the result comparison. The second and third scenarios are locating a crew with a low and high reliable capacity at the first station respectively. In Scenarios 4 and 5, the middle station and in Scenarios 6 and 7, the last station is replaced by a crew with a low and high reliable capacity. Scenarios 8 and 9 are where all the stations replaced by crews with a low and high reliable capacity respectively. Figure Scenarios.

illustration not visible in this excerpt

Figure 3.7. 9 Arrangement Scenarios for the Parade Game model

Design of Cost Scenarios: Our model presumptions for the direct, indirect and planning cost (as part of indirect cost) were explained in details in the previous section. The model first calculates the direct cost for each crew based on the $C per capacity rate assumption, then computes the other costs (like indirect and planning cost for crews or direct and indirect costs for GC) based on the explained cost presumptions.

As discussed, the ratio of the indirect cost or planning cost to the total project cost may vary from one project to another. In order to reach a certain conclusion in this study, all the possibilities are covered by running the model with the three presumptions of 20, 30, and 40% ratio for the indirect to total cost (shown by “I/T”) and three presumptions of 5, 10, and 15% ratio for the planning to total cost (shown by “P/T”) (totally 9 Cost Scenarios). Running each of the 9 Arrangement Scenarios (refer to Figure 3.7) with these 9 Cost Scenarios resulted in running the model with 81 sub-scenarios (shown by “S”). Table 3.2 shows the 81 sub-scenarios, where the two numbers of the array subscripts represent the features of the sub-scenarios. The left number shows the Arrangement Scenario, while the right number indicates the Cost Scenario.

3. Develop the Simulation Model using ARENA: ARENA® (Takus and Profozich, 1997), a generic discrete-event simulation language with a graphical interface, was used in this study to simulate the Parade Game. ARENA® consists of module templates, constructed around SIMAN language patterns, and augmented by a visual front end (Altiokand and Melamed 2007).

The simulation model includes seven crews, lining up in sequence, and the task of each crew is to complete the production of 100 units and they are expected to complete 10 units per week. Using the PDFs, three levels of variability were defined to represent the crews with different production capacity reliability levels in the system: crews with 1) low, 2) moderately and 3) high reliable capacity. Figure 3.8 shows the simulated model of the Parade Game with ARENA. Various kinds of modules in ARENA were implemented to push the model closer to what happened in the real world process. A brief description of the main modules of ARENA simulation is gathered in Table 3.3. Some extra modules or linkages were also used to meet the logical aspects of the way the process was completed. Some modules, such as the Assign module, were only utilized for producing data required in output analysis. However, for a complete explanation of the ARENA simulation modeling, see Altiok and Melamed (2007).

Table 3.2. sub-scenarios based on the Arrangement and Cost Scenarios for Parade Game model Cost Scenarios 1 2 3 4 5 6 7 8 9

illustration not visible in this excerpt

Figure 3.8. Parade Game (new version) simulation model by ARENA

Table 3.3. A brief description of ARENA modules

illustration not visible in this excerpt

Figure 3.9 shows the graphic animation screen of the model, has been built to enable us to watch the game during the model running (during playing). User can allocate any variability level to each of the station before run the model. They can select the variability level by clicking on the Assign module above each station (see Figure 3.10). User can watch the game from the beginning to end. The output is shown on the screen as the game play. User can increase or decrease the speed of the game or pause at any time to watch the process better.

illustration not visible in this excerpt

Figure 3.9. The graphic animation screen of the Parade Game model

illustration not visible in this excerpt

Figure 3.10. The graphic animation screen of the Parade Game model (Zoom in to Stations 1 and 2)

4. Analyze Production Output Results: The developed simulation model enables user to run the game with any scenario and track different outputs related to time, cost and productivity. In this study, the results were divided into two categories: Production and Cost.

In Production category, our focus is to demonstrate the results based on the production-related outputs:

Project Completion Time (week): Project Completion Time is the total time of the project. In other words, it is the number of weeks required for all the seven crews complete the production of all the 100 units.

Crew Completion Time (week): Crew Completion Time is the time taken by each of the crews to process all the 100 units. It is calculated for each crew separately.

Crew Number in Buffer (unit): Crew Number in Buffer is the summation of the number of units remained unprocessed each week in the buffer between any two subsequent work stations (it is also called inventory). It is calculated for each crew separately. Units remained in buffer when the number of receiving units from the previous station is more than the crew’s capacity.

Total Number in Buffer (unit): Total Number in Buffer is the summation of Crew Number in Buffer of all the seven crews.

Crew Lost Capacity (capacity): Crew Lost Capacity is the summation of the capacities remained idle in any week for each of the crews during the game. It is calculated for each crew separately.

Total Lost Capacity (capacity): Total Lost Capacity is the summation of Crew Lost Capacity of all the seven crews.

5. Analyze Cost Output Results: In Cost category, the sub-scenarios are analyzed and compared and the simulation model results are presented based on the following outputs:

Total Crews’ Cost: Total Crews’ Cost is summation of the total cost of all the seven crews. The total cost for each crew is calculated separately based on his/her direct and indirect costs. Note that the planning cost is included in the calculation of the indirect cost.

GC’s Total Cost: GC’s Total Cost is summation of the direct and indirect costs of GC. As noted, the assumption is that the project is all done by subcontractors.

CHAPTER 4 RESEARCH EXECUTION AND ANALYSIS RESULTS

4.1 Analysis Results and Discussion pertaining to Research Objective 1, To Develop the Jobsite Social Network of Construction Crews and Quantify its Impact on Performance (a Case Study)

In this section, the analysis results of the methodology proposed to achieve the first objective are presented and discussed through a case study (Case Study 1).

The case study was a $50M, 150,000 square foot data center. Subcontractor crews were supposed to make their work plan based on the master schedule developed by the GC. They had been asked to breakdown their tasks to the activities with the maximum duration of 1 week and the maximum cost of $10,000. These rules in defining the tasks of the work plan make the crews’ work plan variation comparison more reliable. In this section, the application of the proposed approach is explained step by step and the results will be discussed.

1. Construct jobsite social network of subcontractors based on working space: For each of the 28 weeks of the case study project, two jobsite social networks were developed, one based on the frequency and one based on the severity (overall, 28*2=56 networks). For instance, the jobsite social networks depicted in Figures 4.1 and 4.2 show the interdependencies of the trades from both frequency and severity standpoints in the 18th week of the project. 11 crews were active in Week 18. The existing ties among the crews indicate which crews worked together (share the space) in the same working area during Week 18. The weights of a tie between two crews indicate the frequency (Figure 4.1) and severity (Figure 4.2) of the influences they sent to each other in that time interval. For example, the weights of 2 and 1 for the tie between the “Ceiling Tile” and “Painting” crews in the frequency network indicate that the “Ceiling Tile” and “Painting” crews performed 2 and 1 tasks respectively in the same area in Week 18. These weights are 8 and 1 in the severity network, which shows the influence of the “Ceiling Tile” on the “Painting” crew is 8 times more when you consider the duration (severity) of the tasks. Appendix A presents all the adjacency matrices of crews over the 28 weeks of the project.

illustration not visible in this excerpt

Figure 4.1. Jobsite social network in Week 18 based on frequency

illustration not visible in this excerpt

Figure 4.2. Jobsite social network in Week 18 based on severity

Figure 4.3 shows all the 28 networks of the project; to make the networks less confusing, the weights have not been shown in the figures. It demonstrates the dynamic aspect of the jobsite social network. It visualizes how the interferences change among the 43 subcontractor crews over the course of the project. There are two facts about the dynamic aspect of jobsite social network can be explained by the Figure 4.3:

1. The network complexity can be represented by the network density. Network density simply shows how congested a network is. The “network weighted density”, was calculated for each of the network (can be seen under each network), is the ratio of the sum of the weights of ties versus the maximum possible ties in the network (Liu et al. 2009). The network density fluctuates week by week but gradually increases as project proceeds until it reaches its peak in the Month 6 (Week 21 to 24). The density decreases in the last two weeks of the project. This trend reflects the workload of a typical construction project. The workload is low at the beginning and then it increases gradually until a few weeks before the project completion.

2. The impact of jobsite social network on the crews depends on their position and situation in the network. As the network topology changes in each week, the position and situation of a crew changes. Thus, the influence it receives from the social network varies from one week to another. For instance, two crews (“Electrical” and “Painting”) are highlighted in the Figure 4.3, so their position and situation can be tracked in the networks over the course of the project. The one shown by a rectangular is “Electrical” crew and the other one indicated by a circle is “Painting” crew. Painting crew started its work from Month 3 (Week 9), while the Electrical crew performed its tasks from the beginning of the project. As can be seen in Figure 4.3, the position of them changes in each week as the topology changes, i.e., they experience a different neighborhood and different relationships in each week. Thus the network influence on each of them fluctuates over the course of the project. For instance, comparison between the social networks of Week 13 and 14 indicates that the Painting crew (highlighted by a circle) gets less influence (i.e., less interferences) from the network of Week 13 than Week 14, while the situation of the Electrical crew (highlighted by a rectangular) remains almost the same.

Figure 4.3. Jobsite social network topology over the course of the project (28 weeks)

illustration not visible in this excerpt

Note 1: The node labels used in the drawings represent the crews as follows: A: Ceiling, B: Inspection, C: Door, D: Flooring, E: Water, F: Unistrut Hangers, G: CAT, H, Fire Stopping, I: Ceramic Tile, J: Interior Fixtures, K: Drywall, L: Energy Company, M: City, N: Mechanical, O: Fire Protection, P: Specialty Concrete, Q: Owner, R: Flooring/Finishing, S: Global Inspection, T: Bathroom Fixture, U: Wall Finishing, V: Electrical, W: Roofing, X: Door Supplier, Y: Hardware Distributor, Z: Concrete, AA: Controls, BB: Fiber Installation, CC: Generator, DD: Painting, EE: Electrical Specialty, FF: Floor Cleaner, GG: Security Cage, HH: Steel Fabricator, II: Architect, JJ: Owner/Arch, KK: Substation Install/Testing, LL: Steel Cladding, MM: Misc, NN: Unistruct Installer, OO: Caulking, PP: Fire Proofing, and QQ: Utilities.

Note 2: To make the networks less confusing, the weights have not been shown in the figures, so there was no need to separately show the frequency and severity network.

2. Conduct centrality analysis to calculate network frequency and severity centrality ranks: Centrality analysis was conducted and two centrality values (one based on the frequency and one based on the severity network) were calculated for each crew in each week (based on Equations 3.1 and 3.2). Table 4.1 shows the centrality analysis results for Week 18 (refer to the jobsite social networks in Week 18 shown in Figures 4.1 and 4.2). The second and third columns represent the values of the weighted in-degree centrality for the frequency and severity network respectively. The fourth and fifth columns are the ranks of the crews based on their centrality values (called centrality rank (CR)). Results show that the Electrical crew received the most influence from the frequency network, and the “inspection” and “Hardware Distribution” crews suffers more than the others from the jobsite social network based on the severity. The centrality values quantify the influences of the jobsite social network on the crews and ranking highlights those crews get more influences from the networks. CRs are measurements help the project managers and superintendent to identify the crews with high conflict potentials. The crews with high frequency and severity centrality ranks (high FCR and SCR) have probably the higher chance to be impacted by the jobsite interferences. Project management team can work with the site managers or the crew leaders to take proper actions in order to lighten the impacts. One action could be adjusting the planned schedule to decrease the pressure on the high-ranking crews. More discussion will be provided later in this section.

The same analysis was done for each of the 28 weeks of the project (overall 28 FCRs and 28 SCRs recorded for each crew).

Table 4.1. Centrality analysis results for the jobsite social network in Week 18

illustration not visible in this excerpt

3. Measure work plan variation of subcontractors & 4. Calculate the variation ranks for each of subcontractors: In this research, work plan variation means the time difference between what was planned and what occurred. The four work plan variation variables (V1 to V4, refer to the Equations 3.3 to 3.6), were recorded as performance indices for each crew over the course of the project. The crews were ranked based on their work plan variation variables’ value in each week (V1R, V2R, V3R, and V4R, four ranks for each crews in each week refer to the four measured variables). For instance, the work plan variation variable ratios recorded for the crews in Week 18 are shown in the Table 4.2. Column 1 shows the crews performed their tasks in Week 18. Columns 2 to 7 are the data recorded according to the project work plan and schedule in Week 18. The remaining columns show the work plan variation variables calculated based on the Equations 3.3 to 3.6 along with their ranks in Week 18.

Table 4.2. Work plan variation Analysis results for the Week 18

illustration not visible in this excerpt

5. Run correlation analysis between variation ranks and centrality ranks: From the schedule performance control standpoint, the key crews for the project managers to focus on are those with the higher work plan variation, because when one crew experiences variation, other crews can be impacted. From jobsite interference network standpoint, the key crews are those with the higher centrality, because they receive higher impacts from the network. To investigate the impact of jobsite social network on the crews’ performance, it is tested to reveal how much the network impact is correlated with the crews’ work plan variation indices. To do so, the consistency between the CRs (FCR and SCR) and VRs (V1R, V2R, V3R, and V4R) of the crews was measured. Correlation analysis was conducted to quantify the consistency and find out how significant the relation between the CRs and VRs is. Before presenting the correlation analysis results, to have a better understanding of the relation, the consistency between the CRs and VRs has been depicted for two of the crews, “Painting” and “Electrical”, in Figures 4.4 to 4.7 (Figures 4.4 and 4.6 are for “Painting” and Figures 4.5 and 4.7 are for “Electrical” crew). They show how the VRs change when the CRs of the crews change over the course of the project. Figures 4.4 and 4.5 show the consistency week-by- week, while month-to-month consistency has been depicted in Figures 4.6 and 4.7. Painting crew started its work on Week 11, while Electrical crew performed tasks from the beginning to the end of the project. A sensible consistency between the weekly CRs and VRs can be inferred for the “Painting” crew based on the charts of Figure 4.4. The relation between the CRs and each of the VRs became more noticeable when their consistency was evaluated on a monthly basis (see Figures 4.6 and 4.7). A monthly rank is the average of four weekly ranks. For instance, rank of Month 1 is the average rank of the first four weeks (Weeks 1, 2, 3 and 4).

The impact of social network on work plan variation can be different from one crew to another. It highly depends on the crew’s work type (the type and amount of labor, equipment, space, and material). Figures 4.4 to 4.7 show how the consistency between the CRs and VRs is different between the Painting and Electrical crews. The more consistency level was found for Painting indicates that it suffered more (compare to Electrical) from the interferences of the jobsite social networks. In other words, Painting crew was more sensitive to the interferences than Electrical in the studied case. The case by case examination will help the project management team and site managers to identify the subcontractor crews suffer more from the jobsite interferences, so they can focus more managerial efforts on these crews to alleviate the impact.

illustration not visible in this excerpt

Figure 4.4. Week-by-week consistency between Centrality and Variation Ranks for the “Painting” Crew

Note: FCR is Frequency Centrality Rank and SCR is Severity Centrality Rank

illustration not visible in this excerpt

Figure 4.5. Week-by-week consistency between Centrality and Variation Ranks for the “Electrical” Crew

Note: FCR is Frequency Centrality Rank and SCR is Severity Centrality Rank

illustration not visible in this excerpt

Figure 4.6. Month-to-month consistency between Centrality and Variation Ranks for the “Painting” Crew

Note: FCR is Frequency Centrality Rank and SCR is Severity Centrality Rank

illustration not visible in this excerpt

Figure 4.7. Month-to-month consistency between Centrality and Variation Ranks for the “Electrical” Crew

Note: FCR is Frequency Centrality Rank and SCR is Severity Centrality Rank

Tables 4.3 and 4.4 summarize the correlation analysis between CRs and VRs. The data of all the 43 crews were used in the analysis. The correlation results between weekly CRs and VRs are presented in Table 4.3, while Table 4.4 shows the correlation between the monthly ranks. The most common methods of measuring a monotone association, Spearman, and Pearson correlation analysis, were conducted. Spearman correlation is a non-parametric measure of correlation showing how well a monotonic function can describe the relationship between the CRs and VRs. Pearson correlation measures the strength and direction of the linear relationship between them. The results (both weekly and monthly analysis) show that there is a positive monotonic association between CRs and each of the VRs. It indicates that there is a consistency between the impacts of the jobsite social network (from both frequency and severity standpoints) and the variation ratios, i.e. the more centrality, the more variation. By comparing Tables 4.3 and 4.4, it can be inferred that the relationship between the CRs and VRs is more noticeable when their association was evaluated on a monthly basis. It should be noted that part of this coefficient improvement is because of sample size reduction when the data of four weeks were combined as a single monthly data point (smoother data). Even though the correlation coefficients do not show a strong correlation between the CRs and VRs, they confirm the positive moderate association between the two. It was not surprising that a strong correlation will not be obtained, because there are several uncertainty sources in a construction project other than job site interferences influence the variation ratios. Our focus in this research was on evaluation of the jobsite social network impact, and the moderate positive (linear) correlation obtained for the association appears to be a tangible result.

Table 4.3. Correlation analysis results - weekly Centrality Ranks (CRs) versus weekly Variation Ranks (VRs)

illustration not visible in this excerpt

Note 1: Number of data points was 345.

Note 2: All the correlations are statistically significant at the 0.05 level (P-value < 0.05).

Note 3: All the variables are normally distributed.

Table 4.4. Correlation analysis results - monthly Centrality Ranks (CRs) versus monthly Variation Ranks (VRs)

illustration not visible in this excerpt

Note 1: Number of data points was 140.

Note 2: All the correlations are statistically significant at the 0.05 level (P-value < 0.05).

Note 3: All the variables are normally distributed.

4.1.1 Summary and Conclusion (Objective 1)

The focus of the first research objective was first to illustrate the process of developing the jobsite social network over the course of the project and quantify the impact of network on the crews’ performance by evaluating the consistency between the characteristic of a crew in the network (centrality indices) and its work plan variation.

The applicability of the approach was shown through a case study, which was construction of a 150,000 square foot data center involving a GC overseeing 43 various crews. 1200 tasks were performed during the course of this 28-week study. 56 networks (28 frequency and 28 severity networks) were developed corresponding to the 28 weeks of the project and centrality analysis was conducted for each week. Results showed that how the topology of jobsite social network and accordingly the position and situation of crews change over the course of the project. The proposed approach uncovers the influences sent and received between the crews at any time of the project. The results, as expected, indicated that there is a positive monotonic association between each of the centrality indices and each of the variability indices: the more influences a crew receives form the network, the more work plan variation (less performance) it will have. The results were predictable since the more interference in the jobsite increases the chance of conflict occurrence. The value of the proposed approach is that it helps the project management team to quantify this impact. However, it should be noted that the impact of jobsite social network on the crews differs from one crew to another. Although the correlation analysis was based on the data of all the crews, for better understanding the approach, the consistency between the network centrality and work plan variation was depicted separately for two selected crews (Painting and Electrical crews). The results showed a sensible consistency for the “Painting” Crew, while the results for the “Electrical” crew demonstrated that it suffers less by the influences of the social networks.

The research introduced a new approach for analyzing the jobsite social network and quantifying its impact on the specialty crews for the first time. It helps the construction managers to understand better the underlying network of the crews in a project. The process of network development and centrality analysis was explained step by step, so it can be repeated in any project based on the existing work plan. Additionally, the research has the following benefits for the project/construction managers and superintendents:

1) The proposed approach helps the project managers and superintendents to identify the most critical crews with regard to the influences they get from the jobsite social network (those suffer more from the interferences), so they can take proper actions to reduce the conflict potentials. The jobsite social network for each week can be developed based on the existing work plan of that week before the task execution (in our case, one week ahead). Thus, the centrality values of a crew in that week indicate how much influence it is going to receive from the network at that particular period of the project (in our case, next week). Therefore, the influence of jobsite social network on each crew at any time of the project can be predicted based on the centrality values. Then, in order to alleviate the impact, proper actions can be taken by the site managers or crew leaders with regard to those crews under the strong influence. One common way is to adjust the planned schedule. Easy application of approach enables project managers to run the analysis each time they adjust the plan, so they can select the best alternatives (with the lowest impact on the crews). In some cases where schedule adjustment is not feasible (like performing tasks on the critical path of the schedule), project managers can set extra meeting with the leaders of the critical crews to clarify the difficulties they are going to face and find the best solution to reduce the interferences/conflicts.

2) Although the applicability of the proposed approach was shown for an on-going construction project in this study, it can be used at the preconstruction stage of the project, i.e., prior to starting the project, where GCs make their work plan/task schedule. GCs can implement this approach to evaluate the developed work plan/task schedule with regard to the jobsite social network influences on the crews. Therefore, they can adjust the task schedule or modify the work breakdown structure to achieve the schedule with the least interference/conflict potentials.

4.2 Analysis Results and Discussion pertaining to Research Objective 2, To Examine the Impact of Social Conformity on the Performance of Construction Crews

In this section, the analysis results of the proposed approach for the second objective are presented and discussed through the two projects studied.

4.2.1 Results of First Case Study (Case Study 2)

The first case was a two-story, 350,000 square foot facility center. The project included construction of several data hall suites and administration areas. Steel frame and precast walls were used for the structure frame of the building. The project lasted 24 months with the total cost of $211 million. The project was split into two phases: Phase A/B and Phase C/D. The data were collected through access to the work plan schedule, site observation, and interview with project engineers, superintendents, and crew foremen. The research focused on the main six subcontractor crews of the project: Mechanical (Mech.), Electrical (Elec.), Plumbing (Plum.), Drywall (Dryw.), Painting (Paint.), and Fire Protection (Fire P.). Data were collected on a total of 640 activities over the 15 months of the project. The data collection was started when the main subcontractor crews became active to perform their portion of work in the project. In the following paragraphs, the results of the first case study are discussed based on the methodology steps.

1. Measure Work Plan Reliability Index of Crews/Trades/Subs: The first step is to measure the WPR of the subcontractor crews over the course of the project. On Time Start Frequency Ratio (WPRI1), On Time Start Severity Ratio (WPRI2), On Time Completion Frequency Ratio (WPRI3) and On Time Completion Severity Ratio (WPRI4) were calculated based on Equations 3.7 to 3.10 for each of the subcontractor crews in each month. As an example, the WPRI measurements for the drywall crew are shown in Table 4.5. The number of tasks, total duration, and delays in start and completion time are shown in the table based on the performance of the drywall crew at each month of the project. As can be seen, his/her WPR fluctuates over the course of the project but has more stability in the second phase of the project from Month 11 to 15.

Table 4.5. WPRI measurements for drywall subcontractor crew - Case Study 2

illustration not visible in this excerpt

2. Identify WPRI Norm of Project: Using Equations 3.11 and 3.12, WPRI norm of project was calculated for each of the indices at each month (totally four ProjectNorms corresponding to the four WPRIs for all crews). For instance, Figure 4.8 shows the project and neighborhood norm of WPRI4 over the course of the project. The figure also includes the measurements of WPRI4 for the drywall crew to show an example of comparing a WPRI and norm. As can be seen, the WPRI4 of the drywall crew is lower than the project norm in most of the first phase of the project, but it is higher than that in the second phase. The difference between the WPRI4 and the project norm (Deviation) decreases as the project proceeds in each phase. DFPN will be discussed in details in Step 5.

illustration not visible in this excerpt

Figure 4.8. WPRI4 and ProjectNorm4 over time for drywall subcontractor crew - Case Study 2 3. Construct Social Network of Subcontractor Crews: In an effort to minimize the crew interference, the project manager divided the overall project into 23 different working areas. Fifteen adjacency matrices were created corresponding to the months of the project. Due to space limitations, only ASN respective to the 4th month of the project is shown (see Table 4.6). Arreys of the ASN indicates how many times the subcontractor crews have worked together in the same area at the same time in Month 4. For example, the number “2” between mechanical and plumbing crews in the matrix means that they have worked together twice in Month 4. Appendix B shows all the adjacency matrices of the project.

Table 4.6. Adjacency matrix of Month 4 - Case Study 2

illustration not visible in this excerpt

Fifteen networks were created based on the developed adjacency matrices; each represents the social network of subcontractor crews at each month. Figure 4.9 shows the monthly social networks of subcontractor crews. There was less work from Month 6 to 10 due to the transition from the first to the second phase of the project. Note that when there is a node without any connecting tie, it means that that subcontractor crew was there in the project performing his/her job in that particular month but he/she did not have any neighbors (did not share the working area).

Figure 4.9. Social networks of the subcontractor crews over the course of the project - Case Study 2

illustration not visible in this excerpt

4. Identify WPRI Norm of Neighborhood: In Step 2, the norm was calculated from a global view (called ProjectNorm). In this step, using social network analysis, only the neighborhood of the subcontractor crews is included in calculation of the norm. As discussed in the methodology section, the neighborhood norm for each crew is unique since his/her neighborhood is unique at each month. Using Equations 3.13 and 3.14, WPRI norm of neighborhood was calculated for each of the indices and for each subcontractor crew separately in each month (totally four NeighborhoodNorms corresponding to the four WPRIs for each crew). The trend of the neighborhood norm of WPRI4 for drywall crew is shown in Figure 4.8. Looking at Figure 4.8, it can be inferred that the trend of WPRI4 for the drywall crew is more close to his/her neighborhood norm (local) than the project (global). The comparison between the ProjectNorm and NeighborhoodNorm was one of the goals of this research and will be discussed in detail in the next steps.

5. Measure deviation from project norm (DFPN): A visual illustration of the DFPN is shown in Figure 4.8 for the drywall crew (DFPN4,2 in Figure 4.8 refers to the deviance of the WPRI4 from the project norm in the second month of the project). Using Equation 3.15, DFPN was calculated for each subcontractor crew in each month to quantify the similarity/dissimilarity of the crews to the project norm. Figure 4.10 compares the deviation of the WPRIs from project norms among the crews over the course of the project. As expected, the trend of DFPN varies from one subcontractor crew to another and also varies from an index to another. Although DFPN fluctuates several times over the course of the project, its trend is descending for most of the crews. The statistical relationship between any of the DFPNs and the project time will be discussed in Step 7.

illustration not visible in this excerpt

Figure 4.10. Deviation from project norms over time - Case Study 2 6. Measure deviation from neighborhood norm (DFNN): A visual illustration of the DFNN is shown in Figure 4.8 for the drywall crew (DFNN4,2 refers to the deviance of the WPRI4 from the neighborhood norm in the second month of the project). Similarity/Dissimilarity of the crews’ WPRI to the neighborhood norm was measured through DFNN (based on Equation 3.16). Figure 4.11 demonstrates DFNN of subcontractor crews for the four WPRIs over the course of the project. Similar to DFPNs, the DFNNs fluctuate over the course of the project with a general descending trend except DFNN1. DFNN1, as shown in Figure 4.11.a, is less than 20% for all the crews in most of the time of the project. This indicates that the subcontractors of the project synced well to their neighborhood norm related to the WPRI1. The other three DFNNs (see Figures 4.11.a, 4.11.b, and 4.11.c) are higher at the beginning of the project for most of the subcontractor crews, but it gradually reduces as the project proceeds. This descending trend indicates that the subcontractor crews of the case study start the project with different levels of WPR, but as the project proceeds, their WPR becomes more similar to their neighborhood norm.

illustration not visible in this excerpt

Figure 4.11. Deviation from neighborhood norms over time - Case Study 2

7. Conform to local or global?: The last step is to conduct a correlation analysis between the deviation from norm (DFPN and DFNN) and the project time to 1) identify if there is any meaningful trend for the tendency of subcontractors to the norms over the course of the project and 2) examine the role of social network in conformity of subcontractors by comparing the DFPN and DFNN trend over time. Analysis includes all the subcontractor crews of the project. The results of Pearson, Kendall Tau, and Spearman correlation analysis are tabulated in Table 4.7. Figure 4.12 also demonstrates the trend of DFPNs and DFNNs over the project time. Results show that there is a significant negative correlation between each of the deviation from norms and time except for the DFPN3. Results also demonstrate that the DFNN has a stronger negative correlation with project time than the DFPN. This also can be inferred from Figure 4.12 where the slope of DFNN linear line is steeper than DFPN for all the four WPRIs. This indicates that the tendency of the subcontractors of our case to follow the norms of their neighborhood (local) is higher than the one to follow the project norm (global). This reveals how WPR of the subcontractor crews in the project is under the influence of their social network over the course of the project.

Table 4.7. Correlation analysis results for the relationship between DFNN/DFPN and project time - Case Study 2

illustration not visible in this excerpt

Note 1: Number of data points was 65.

Note 2: All the correlations are statistically significant at the 0.05 level (P-value < 0.05).

Note 3: All the variables are normally distributed.

illustration not visible in this excerpt

Figure 4.12. Trends of DFPN and DFNN over time - Case Study 2

4.2.2 Results of Second Case Study (Case Study 3)

The second case was a single floor, $50 million, 150,000 square data center. The main subcontractor crews (Mechanical (Mech.), Electrical (Elec.), Concrete (Conc.), Drywall (Dryw.), Painting (Paint.), Steel Fabrication (Stee.) and Fire Protection (Fire P.)) of the project were studied from early to the completion of the project. The project was construction of a data center and multiple white-space computer labs on an existing warehouse building. The data were collected through several site visits and access to reports of the project weekly meeting. In the weekly meeting project engineer, superintendents, and crew foremen reviewed the previous week’s work plan and reported whether or not each task was completed. Additionally, each subcontractor crew reviewed the three week look-ahead plan they had developed. Since the methodology application was reviewed in details for the first case study, it is not repeated step-by-step for the second case. The results are limited to the developed social networks (Step 3) and the correlation analysis for local and global conformity (Step 7).

1. Measure WPR Indices (WPRI) of Subcontractor Crews: WPRIs were measured based on Equations 3.7 to 3.10 for each of the subcontractor crews at each month (calculation method is similar to what was done in Case Study 2).

2. Identify WPRI Norm of Project: Using Equation 3.11 and 3.12, WPRI norm of project was calculated for each of the indices at each month (calculation method is similar to what was done in Case Study 2).

3. Construct Social Network of Subcontractor Crews: The project management team divided the overall project into 5 main working areas and for each of the tasks in the project work plan, a working area was allocated. Figure 4.13 shows the social networks of subcontractor crews over seven months of the project. Appendix C shows all the adjacency matrices of Case Study 3.

illustration not visible in this excerpt

Figure 4.13. Social networks of the subcontractor crews over the course of the project - Case Study 3

4. Identify WPRI Norm of Neighborhood: NeighborhoodNorm was calculated for each of the indices and for each crew separately at each month (totally four NeighborhoodNorms corresponding to the four WPRIs for each subcontractor crew).

5. Measure deviation from project norm (DFPN): Similar to what was shown for the first case study, similarity/dissimilarity of the crews’ WPRI to the norm of project was measured by DFPN (based on Equation 3.15).

6. Measure deviation from neighborhood norm (DFNN): Using Equation 3.16, DFNN was calculated for each subcontractor crew in each month to quantify the similarity/dissimilarity of the subcontractor crews to the neighborhood norm.

7. Conform to local or global?: The results of Pearson, Kendall Tau, and Spearman correlation analysis are tabulated in Table 4.8. Figure 4.14 demonstrates the trend of DFPNs and DFNNs over the project time. Results show that there is a significant negative correlation between each of the DFNNs and time. There is no meaningful monotone trend found for the DFPN1 and DFPN4. Results also reveal that the DFNN has a stronger negative correlation coefficient with the project time than the DFPNs (slops is steeper for DFNNs in Figure 4.14). This finding is as the same as what was found in the first case, subcontractors tends to follow the norms of their neighborhood (local) more than the project norm (global). This confirms the role of social network in changing of subcontractors’ WPR over the course of the project.

Table 4.8. Correlation analysis results for the relationship between DFNN/DFPN and project time - Case Study 3

illustration not visible in this excerpt

Note 1: Number of data points was 44. And all the variables are normally distributed

Note 2: All the correlations are statistically significant at the 0.05 level (P-value < 0.05).

illustration not visible in this excerpt

Figure 4.14. Trends of DFPN and DFNN over time - Case Study 3

4.2.3 Summary and Conclusion (Objective 2)

Most of the research conducted on identifying the impact of WPR on the project performance focused on its task-sequence related impact, which can be considered as the direct and immediate impact of WPR. Construction industry lacked a good understanding of the social aspect of WPR variability. Beyond the immediate impact on the tasks’ sequence of other subcontractor crews, it would appear that the reliability of subcontractor crews’ work plan in a project can gradually develop a kind of norm reflecting the level of WPR expectation from the subcontractors. Through the second research objective, a social network based approach was proposed to quantify how a subcontractor WPR varies when the WPR norm changes over the course of the project. Four indices were developed to measure reliability of crews’ work plan regarding to task start and completion time. Then, examining relationship spreads based on conformity concepts in sociology the consistency of subcontractor WPR with their neighborhood or project norm was quantified.

Two commercial construction projects with several subcontractor crews were studied. Statistical analysis results of case studies show that deviation of WPRI from norms (both project (DFPN) and neighborhood (DFNN)), reduced over the course of the project. This generally shows that the crews of the given cases had tendency of conforming to the norm of the WPR. The comparison between the correlation coefficient of deviation from project norm (DFPN) and neighborhood (DFNN) and project time shows that the DFNN has a stronger negative correlation in both cases. This points out that the tendency of subcontractors to follow the norms of their neighborhood (local) is higher than their willing to follow the project norm (global). Thus, correlation analysis results in both cases uncover the role of social network (refers to neighborhood) of subcontractor crews in their WPR. The conclusion was based on the results of the two case studies; however, the approach can be implemented in any other construction project.

The research findings have various benefits for the project/construction managers and superintendents. First of all, it helps the construction managers to understand that the WPR fluctuation of subcontractors can have an important indirect impact on the performance of other subcontractor crews other than the task-sequence related impact. It can be argued that the consequence of the WPR variability will be felt more than what the project management team expects due to the social impact discussed through Objective 2. The proposed approach also introduced new indices to construction managers to quantify the conformity level of subcontractor crews to the WPR norms. It can help them to identify the conformist/non- conformist crews in the project, so that they can include this information in efficient development the subcontractors’ schedule and sequence and WPR initiatives. One way is to put the unreliable crews in reliable neighborhood, or controlling the norms by setting some incentives (rewards/punishment) for those have higher influence on the norm creation. Additionally, the proposed approach helps management team to have a better prediction of the WPR of the subcontractors over the course of the project, so that they know how much effort is necessary to spend on WPR improvement initiatives. Knowing the behavior of subcontractors regarding the WPR can also be valuable information for general contractors when they need to establish a long-term working relationship with a particular subcontractor.

Case studies, as expected, showed that the conformity level of subcontractors varies from a crew to another. The findings showed that there was a level of conformity for most of the subcontractors in the studied projects. However, it should be noted that sometimes there are some predominant incentives for a subcontractor, other than the WPR expectation of the project or general contractor, influences his/her WPR level. Maintaining good reputation, pursuing long-term relationship with general contractor, and willing to excel are the examples of subcontractors’ incentives to keep a good WPR in an unreliable project/neighborhood.

4.3 Analysis Results and Discussion pertaining to Research Objective 3, To Identify the Improvement Direction (Benchmarks) for Inefficient Construction Crews

In this section, the analysis results of the methodology proposed to achieve the third objective are presented and discussed through a case study (Case Study 4).

1. Determine Data of Inputs/Outputs: For the inputs, the levels of seven precondition readiness variables (as mentioned in Research Methodology) were tracked before execution of each task. The data were collected through interviews with GC site managers who tracked the project progress, updated the schedule, and controlled the crews’ performance. GC site managers include a project manager, a superintendent and a project engineer. The interview and the data collection were done at the Week 10 of the project; therefore the data are based on the records from the beginning of the project to the Week 10. Via a questionnaire, the three site managers were asked to rate the average level of precondition readiness for activities (tasks) done by any of the crews. The seven input variables are quantified for each crew out of 100. For example, if the required materials for the crew are always available and provided, a scoring of 100 was given by the site managers to the variable “material availability” for that crew. Otherwise, based on the available records and their estimation, they gave a score between 0 and 100 depending on the situation. The data were originally collected based on the 10-point scale (10, 20,…, 100), but in order to have more accurate data, in the case a respondent was not sure between two numbers, a number in between were given based on the situation. At the end, the average of the three respondents’ evaluation was used. For example, a value of 0.75 for data input “Material” for the Metal crew (ML) means that although the management group has always tried to provide enough material for Metal crew to complete the work smoothly, averagely, in 25 percent of time Metal crew has suffered because of the lack of material (Note that in this project, providing the material on time is the GC’s responsibility and not that of the Metal crew). Table 4.9 shows the values obtained for the seven input variables for all the STs.

Table 4.9. Summary of the Raw Data

illustration not visible in this excerpt

For the outputs, the performance of each crew is evaluated by measuring his/her work plan reliability and cost/schedule performance. The following four types of outputs were measured as indicators of performance level: 1) Percent of tasks starting on time (the number of planned tasks is started at the planned time / number of total planned tasks), 2) Percent of tasks finishing on time within the planned duration (the number of planned tasks is finished within the planned duration / the number of total planned tasks.), 3) Cost performance ratio (total planned cost / total actual cost), and 4) Schedule performance ratio (total planned work hours / total actual work hours). The summary of raw input and outputs data for the case study are represented in Table 4.9.

2. Principal Component Analysis on Inputs and Outputs: Based on the eigendecomposition of the covariance matrix of the data (PCA), it can be concluded that three first PCs for the inputs and two first PCs for the outputs can account for most of the variance of the data (87% of the inputs’ variance and 93% of the outputs’ variance). Eigenvalues and corresponding eigenvectors of the selected PCs are shown in Table 4.10.

Table 4.10. Eigenvalues and corresponding eigenvectors of the selected PCs

illustration not visible in this excerpt

Therefore, the final PCs of inputs (I) and PC of outputs (O) are:

illustration not visible in this excerpt

PCs have been weighted based on their eigenvalues, i.e. their importance, so that they represent the original data well. For example, 0.55 in the PC1 of I is the result of (0.0161)/(0.0161+0.0072+0.0062). Now, by using these PCs instead of the real data, the number of variables will be reduced from 7 to 3 for the inputs and 4 to 2 for the outputs, without losing much information. By replacing the true value of the variables in the PCs, the new set of data, which will be used for DEA, was generated and tabulated in Table 4.11.

Table 4.11.The New Set of Data based on PCA

illustration not visible in this excerpt

4.3.1 Results of Conventional DEA benchmarking

3. Run Basic DEA: To conduct DEA, Equation (3.17) was solved (inputs: PCs of I; outputs: PCs of O) for each of the twelve crews. In addition to obtaining the efficiency score (θ*), the suggested benchmarks were assigned to each of the inefficient crews. Due to space limitations and the repetitive nature of the calculation, the steps are shown for one of the crews, “Finishing” Crew FI:

illustration not visible in this excerpt

Using DEAfrontier© Software, the model analysis was performed, and the set of unique optimal solutions of θ*=1.15, λ5 = 0.71, λ12 = 0.33, and λ1, λ2, λ3, λ4, λ6, λ7, λ8, λ9, λ10, λ11 = 0 was obtained. The efficiency score of 1.15 by the FI indicates that the Finishing crew is inefficient (Efficiency Score > 1.00) compared to the best-performing crews/benchmarks (CO, WP, ME, EL, and LA) in the studied project. The difference between the FI and the benchmarks is 15% (1.15 - 1.00). Furthermore, the conventional DEA benchmarking perspective, based on the values of λi, suggests that the DMU5, “Mechanical” crew (ME) and the DMU12, “Landscaping” crew (LA), would be the best benchmarks for the FI. Since the Output-oriented DEA model was used in the calculation, this means that FI will be able to learn from the ME and LA’s practices to improve its output levels (such as cost and schedule performance ratio) with the current level of its inputs (current level of precondition variables such as weather conditions, equipment, labor, and material availability etc.).

4. Identify DEA benchmark(s): By solving the Equation (3.17) for all the crews, the efficiency scores and the suggested benchmarks (if any) were obtained. The minimum efficiency score of 1 was achieved by five crews: CO, WP, ME, EL and LA. These five crews are efficient (best-performing), since they can obtain the optimal efficiency score of 1.00. The remaining seven crews, with a value greater than 1.00 (the greater the efficiency score, the less efficient the ST) are considered to be inefficient (underperforming). Table 4.12, in addition to showing the ranking of crews (based on their efficiency), indicates which ST(s) from those five efficient crews are suitable benchmark(s) for the inefficient crews from the conventional DEA benchmarking perspective.

Table 4.12. Conventional benchmarks for the inefficient crews

illustration not visible in this excerpt

4.3.2 Results of Social Network based DEA benchmarking

In the previous section, the conventional benchmarking was done by performing the basic DEA. Next, our developed procedure, SDBP, was applied to the case to demonstrate its ability to overcome the aforementioned difficulties of the conventional DEA benchmarking, i.e. “availability” and “achievability” of the benchmarks. The final results of the SDBP implementation in the case study will be the determination of the practical benchmark(s), which is (are) “achievable” and “available”, for the inefficient/underperforming crews. This section explains the four steps of the proposed procedure (Steps 5 to 8).

5. Construct Jobsite Social Network of Crews/Trades/Subs based on Communication: The first step in establishing a social network is constructing the ASN. Table 4.13 shows the ASN of the case study. The aij shows the tightness of the relationships between the Crews “i” and “j”, which is also usually indicated by “Line Value” on the top of each tie (connection) in the social network depiction.

Table 4.13. ASN of the specialty crews

illustration not visible in this excerpt

Figure 4.15 shows the social network of the crews, which was built according to the ASN (Table 4.13). The line values are not shown in this depiction to prevent showing the messy and confusing network. Instead, the thickness of each connecting line in the social network was used to represent the tightness of each connection. The tighter relationships (aij) are shown by the thicker lines. As a common rule, the crews which have more connection lines are also located close the center of the SN.

illustration not visible in this excerpt

Figure 4.15. The specialty crews organized by social network 6. Measure the Relationship Score (RS) based on sub-SNs. A sub-SN for each inefficient crew is developed in order to see its interactions/interdependencies in a separate network. The sub-SN of the FI includes TM, SP, ME and MA, since it connects with only these four crews out of 11 possible connections. Figure 4.16.a depicts the sub-SN for the FI.

RSs were calculated (Equations (4.7) to (4.10)) based on the Equation (3.18) and are shown in Figure 4.16.b (Line Values (aij) are available from ASN in Table 4.13). The values obtained for the RSs indicate that the FI has the most frequent relationships with MA.

illustration not visible in this excerpt

7. Measure the Relative Efficiency Distance (RED) on sub-SNs: DEA was conducted for each sub-SN to measure the efficiency level of the crews, relative to their neighbors. The calculation is shown here for the FI sub-SN which includes the FI, TM, SP, ME and MA.

illustration not visible in this excerpt

The efficiency scores of the FI, ME, TM, MA, and SP were calculated to be 1.12, 1.00, 1.00, 1.00 and 1.09 respectively. The scores indicate that the ME, TM, and MA are the most efficient crews in the sub-SN of the FI. Note that two of these efficient crews (TM and MA) were not efficient in the scale of the entire project based on the conventional benchmarking results, done in the previous section. Figure 4.16.c shows the efficiency scores of crews in the FI sub-SN.

Using the efficiency score assigned by the DEA, the REDs of the crews in the FI subSN were calculated as follows based on the Equation (3.19) (4.16.d):

illustration not visible in this excerpt

Figure 4.16. Steps of SN-based DEA benchmarking procedure for the “Finishing” crew (FI): (a) FI sub-SN; (b) FI sub-SN showing Relationship Score (RS); (c) FI sub-SN showing efficiency level; and (d) FI sub-SN showing Relative Efficiency Distance (RED)

8. Identify practical benchmark(s): Practical benchmarks are those best-performing crews, which are “available” and “achievable” enough for the inefficient crew to learn from their practices. A two-dimensional scatter plot was developed for the FI, in which both “availability” (the greater RS value the better) and “achievability” (the closer the value of RED is to +15% the better) of its neighbors have been considered to find the best practical benchmark(s). As can be seen in the Figure 4.17, MA with the RS of 45% and RED of +10.7% is the best practical benchmark for the FI among the other possible benchmarks. This means that getting to MA is the most reasonable target for the first improvement move for the FI, and it is effective for the FI to learn from the MA’s practices. The same process was done for all of the inefficient crews in the jobsite and the results were summarized in the Table 4.14.

illustration not visible in this excerpt

Table 4.14 also compares the results of conventional DEA and SDBP. As predicted, the benchmarks assigned by the conventional DEA benchmarking are different from those obtained from SDBP. For instance, MA and SP are those crews determined as benchmarks by the SDBP more often than the others, while they are not efficient at all when the DEA was run among all the 12 crews (conventional DEA benchmarking). The reason is that the SDBP performs the analysis inside the crews’ sub-SN and takes the benchmarks’ availability (via RS ratio) and achievability (via RED ratio) into account, but the conventional DEA benchmarking does not.

Table 4.14. Results of Conventional DEA versus SN-based DEA Benchmarking

illustration not visible in this excerpt

It should also be noted that there may be an inefficient crew without any interactions/interdependencies with the better performers in its sub-network. For instance, MA and PL are efficient in their own sub-SN, while they are not efficient among all the 12 crews (based on the conventional DEA benchmarking results). In these cases, the benchmarks are assumed to be assigned from the crews outside of the sub-SN in such a way that it will be possible to make connections between them and the inefficient ST. This may vary depending on the conditions and features of the projects.

4.3.3 Summary and Conclusion (Objective 3)

A DEA is based on the ratio between a linear combination of outputs over a linear combination of inputs. The ratio and the directions of those linear combinations could change dramatically if samples with different scales and operational nature are chosen. In other words, the statistical properties of the samples can influence the optimization process, and in return the optimization process could also have impacts on the statistics. This study provides a feedback mechanism between the sampling and optimization processes. As a result, this study helps identify the benchmarks that are both DEA achievable and available through SN.

The research contributes to the body of knowledge by developing an innovative procedure, called SDBP, which combines the DEA (assessing the relative efficiency of the crews) and SNA (concentrating on the interactions/interdependencies among the crews) to untangle one of the underlying influential behavioral features, i.e. “social learning” in the construction jobsites and take advantage of that in identifying practical benchmarks for the inefficient crews over the course of the project. It overcomes the two difficulties associated with the conventional DEA benchmarking method, i.e., the “availability” (via measuring the availability level of the neighbors in the social network (RS ratio)) and “achievability” (via measuring the achievability level of the neighbors in the SM (RED ratio)).

A $2.5M residential project involving twelve crews was selected to show the application of the SDBP. The developed procedure identified a crew for each of the inefficient/underperforming crews as the practical benchmark. For instance, SDBP suggested the MA (with RS = 45% and RED = +10.7%) as the best practical benchmarks of the FI crew, so that to learn from the other good performers’ practices, the MA crew could be the best choice for the FI.

Feedback from various experienced project site managers and construction crew’s supervisors indicate that the SDBP technique will be helpful for project/site managers as it outlines a way to identify the improvement direction for the inefficient crews in the construction projects, not through investigating and following the supreme objectives, but by “learning from the others” or “social learning”. It provides project managers with the capacity to evaluate the crews’ potential for acting as practical benchmarks. For example, from the case study results, it can be concluded that the MA and SP crews could play a key role in assisting the inefficient crews in performance improvement because they were determined as the benchmarks by the SDBP more than the other crews. The SDBP is also beneficial to the managers by assessing the existing relationships and dividing them into two categories: one is those relationships contributing to improving the crews’ performance and second is those do not help in efficiency improvement. Therefore, one possibly effective task for the project manager would be taking some actions to meet and intensify the effective relationships. This procedure not only can be used in an on-going project, it also will be useful in arranging and organizing the foremen/supervisors’ connections/relationships for the future projects of the company.

This research is significant in three primary regards. First, the research is unique as it looks at interdependencies and relationships of construction crews from an exclusively analytical perspective. Second, the analytical process is presented step-by-step, thus, it is repeatable for others to adopt and apply to their own research case within the construction industry. And third, this discusses the combination of the DEA and SNA for the first time as a quick way to reach the optimal solution in construction projects.

4.4 Analysis Results and Discussion pertaining to Research Objective 4, To Investigate the Cost/Benefit of Low or High Reliable Construction Crews and Develop a New Educational Version of Parade Game

In this section, the analysis results of Objective 4 is presented and discussed. A new version of Parade Game is developed to demonstrate the cost/benefit of having the crews with low and high reliable production capacity in construction projects. The simulation model is designed to be usable for construction management educators. The results (outputs) are divided into two categories: Production and Cost.

4.4.1 Production Output

Figure 4.18 provides the results of the Parade Game simulation when the first station (Station 1) is replaced by a crew with a low/high reliable capacity (high/low variability). The goal is to identify the consequence of having higher or lower variability at first station. Figure 4.18.a shows how the completion time is affected by this change. As expected, the crews completed the project (processing 100 units) earlier when there is a crew with high reliable capacity in the upstream side (and vice versa). The other notable point is that the impact of the crew with a low reliable capacity (at the first station) is felt immediately in the system by the crew at second station, while the positive influence of the crew with high reliable capacity becomes tangible from Station 4. The project completion time will be also reduced when the first station is replaced by crew with a high reliable capacity. The completion time reduction (either for the crews or the project) is small. Part of that is because the difference among the SD (as indicator of variability) of the PDFs of the various capacity reliability levels is not large. The other reason is that the most of the impact of the crew’s variability at Station 1 will be faded away in the next moderately reliable stations.

Figures 4.18.b and 4.18.c show Crew Number in Buffer and Lost capacity respectively. The assumption is that there is no buffer or lost capacity for the first station, so the charts exclude that. As can be seen, when the first station is replaced by a crew with a low reliable capacity, both Number in Buffer and Lost Capacity increase for the subsequent crews. This will be opposite when the first station is replaced by a crew with a high reliable capacity. The impact is instant (short-term) on Number in Buffer since mainly felt by Station 2. This is because the first station’s variability contributes to an unstable work flow right at the beginning of the game (first few weeks of the project), and therefore, the unprocessed units are accumulated between Stations 1 and 2 very quickly. As a result, after a few weeks from start of the game, the crew at Station 2 usually has enough unprocessed units in his/her backlog to process. This means that his/her production is not affected by the variability of the first station after a certain point, so the variability of the first station passes to the downstream only in the first few weeks of the game.

illustration not visible in this excerpt

Figure 4.18. Parade Game simulation model results - Replacing first station by a crew with a low/high reliable capacity Note: Results are based on 100 simulation model replications.

Lost Capacity increases by about 2 in most of the stations when the first station is replaced by a crew with a low reliable capacity and decreases by between 1 and 2 when the first station is replaced by a crew with a high reliable capacity. This impact is resulted from the high variability flow in the first few weeks of the project (game).

Figure 4.18.d compares Project Completion Time, Total Number in Buffer and Total Lost Capacities among the scenarios of the first station replacement. Project finishes earlier, the total capacities lost and the units remained unprocessed during the project decreases when the variability is lower upfront (by putting a crew with a high reliable capacity at the first station). The result will be opposite when the variability is high at the first station.

Figure 4.19 presents the results of the Parade Game simulation when the middle station (Station 4) is replaced by a crew with a lees/high reliable capacity. The goal is to identify the consequence of having higher or lower variability at the middle of the project task sequence by comparing the Arrangement Scenarios “M M M M M M M”, “M M M L M M M”, and “M M M H M M M”. Overall, the project completion delays, the total number of units remained in the buffer and the capacities lost increase when there is a crew with a low reliable capacity at middle station (see Figure 4.19.d). This would be opposite when the middle station is replaced by a crew with a high reliable capacity.

illustration not visible in this excerpt

Figure 4.19. Parade Game simulation model results - Replacing middle station by a crew with a low/high reliable capacity Note: Results are based on 100 simulation model replications. Figure 4.19.a, 4.19.b, and 4.19.c show the impact of the middle station replacement on Crew Completion Time, Number in Buffer, and Lost Capacity. As shown by the graphs, the first three stations are not affected since the variability changes from Station 4 in the sequence. Stations 4, 5, 6 and 7 complete the project earlier due to the lower variability of the middle station. What a crew with a high reliable capacity at middle station does is absorbing some of the coming variation from the upstream and creates a stable work flow for the downstream. This not only increases the productivity of the downstream stations (Station 4, 5, 6, and 7), but also decreases their lost capacities and unprocessed units in their buffer. The other remarkable point is that the most impact of the middle station replacement regarding the Number in Buffer is on the Station 4 itself. The accumulated variation of the first three stations results in a certain number of units in the buffer of Station 4 (in between Station 3 and 4). If the Station 4 is replaced by a crew with a high reliable capacity, this variation will be absorbed to a certain level by Station 4, so the number of units remained unprocessed in the following buffers will decrease. If it is replaced by a crew with a low reliable capacity, the variation will be intensified and the units remained in the buffer will increase. The impact starts fading away from Station 5 to the last station due to the medium variability at those stations.

Figure 4.20 shows the results of the Parade Game simulation when the last station (Station 7) is replaced by a by a crew with a low/high reliable capacity. The aim is to examine the consequence of having higher or lower variability at the end point of the project task sequence by comparing the Arrangement Scenarios “M M M M M M M”, “M M M M M M L”, and “M M M M M M H”. Since the last station is replaced, the first six stations will not be affected by this change and therefore the impact on the project outcome will be little. The only impact will be on the last station itself. When the last station is replaced by a crew with a high reliable capacity, he/she can complete his/her own part a little bit earlier with lower number of lost capacities and unprocessed units in the buffer. This will also result in lower project completion time.

illustration not visible in this excerpt

Figure 4.20. Parade Game simulation model results - Replacing last station by a crew with a low/high reliable capacity

Note: Results are based on 100 simulation model replications.

Figure 4.21 demonstrates the results of the Parade Game simulation when all the seven stations are replaced by crews with a low/high reliable capacity. This refers to the original version of Parade Game where all the players use the same die (same variability in all the stations). Comparison was done among the three scenarios “M M M M M M M”, “L L L L L L L”, and “H H H H H H H”. The completion time of the crews and the project, the lost capacities, and the number of units remained in the buffer will increase dramatically when all the stations have a low reliable capacity. The outcome will be opposite when all the crews have a low variability. Tommelein et al. (1999) concluded the same.

There is another notable fact can be inferred from the results of the original version of Parade Game, in addition to its commonly known conclusion (i.e. productivity decreases when variability increases): Variability has an accumulative impact on the crews. From upstream towards the downstream, the effect of the variability change (in our study change from “M M M M M M M” to “L L L L L L L” or “H H H H H H H”) on the completion time and the lost capacity (see Figures 4.21.a and 4.21.c) becomes more significant. This increasing effect is a result of accumulative impact of variability. The existing variability in any station is added to the receiving variability (which is in the form of variation) from the previous station and then passes to the next station. For example, the crews at Station 3 receives the variability of the crew at Station 2 which also includes a portion (if not all) of the variability of the crew at Station 1. Thus, it can be concluded that the crews in downstream stations suffer more from variability since they receive the accumulated variability of all their previous stations. While the impact of variability change on the completion time and lost capacity of the crews in the downstream stations are more significant, its effect on their Number in Buffer (inventory) is lower the upstream stations (see Figure 4.21.b). This also occurs due to an indirect consequence of the accumulative impact of variability. As discussed above, by moving towards the downstream, the crew’s productivity is affected by an stronger variability (accumulative impact), so the crews in the downstream stations, most of the times, receive lower number of units than the upstream ones from their previous station. Therefore, their capacity is more than their inventory most of the times (in any of the scenarios) and this generally results in lower number of units in their buffer.

illustration not visible in this excerpt

Figure 4.21. Parade Game simulation model results - Replacing all the stations by crews with a low/high reliable capacity

Note: Results are based on 100 simulation model replications.

A summary of the results have been tabulated in Tables 4.15, 4.16 and 4.17. They show the quantitative comparison between the basis (which is Scenario 1 where all the crews have medium variability) and all other Arrangement Scenarios. Table 4.15 represents how the completion time varies for the crews and project when the first, middle, last, or all the stations are replaced by different capacity reliability levels. The second row shows the completion time of the first scenario where all the crews have medium variability, the rest shows how the values change in other scenarios relative to the first scenario. For instance, it can be understood that the project completion time will increase by 1.1% if there is a crew with a low reliable capacity at the first station. Tables 4.16 and 4.17 show a similar comparison for the number in buffer and lost capacity. Generally (as shown by the charts), the outcomes are always improved when any of the stations is replaced by a crew with a high reliable capacity and worsened if any of the stations is replaced by a crew with a low reliabile capacity. In addition to the generic conclusion, couple of noteworthy points can be inferred from this quantitative comparison:

1. Results show that if all the crews have medium variability and there is an opportunity to make an investment on only one of them to become a crew with a high reliable capacity (low variability), the best choice regarding the project completion time would be the crew at the middle of the sequence (Station 4), although the common perception prefers the earlier stations (See Table 4.15, Arrangement Scenario 5). Comparing Arrangement Scenarios 3, 5, and 7 in Tables 4.15 and 4.16, it can be inferred that the project will complete earlier and the total number of units remained in the buffer will be lower if the middle station is replaced by a crew with a high reliable capacity and not the first or last stations. That is because a crew with a high reliable capacity at the middle station improves the work flow in two ways: he/she 1) absorbs some of the coming variation from the upstream, and 2) benefits the downstream stations by sending them a stable work flow. A crew with a high reliable capacity would not have the opportunity to absorb any variation if he/she was at the first station (because there is no upstream for his/her) and would not be able to benefit the downstream if he/she was at the last station (because there is no downstream for his/her).

2. Results show that if all the crews in work sequence have medium variability except one, which has a high variability level, the ideal alternative to have the minimum loss is to relocate him/her to the last station. Contrary to what was concluded for relocating a crew with a high reliable capacity (refer to previous paragraph), a crew with a low reliable capacity has strongest negative impact on the project completion time and lost capacity if he/she stands at first station of the sequence. However, Total Number in Buffer will be affected more when he/she stands at the middle of the sequence.

Table 4.15. Comparison between the nine Parade Game Arrangement Scenarios - Completion Time (week)

illustration not visible in this excerpt

Table 4.16. Comparison between the nine Parade Game Arrangement Scenarios - Number in Buffer (unit)

illustration not visible in this excerpt

Table 4.17. Comparison between the nine Parade Game Arrangement Scenarios - Lost Capacity (capacity)

illustration not visible in this excerpt

4.4.2 Cost Output

Modeling and analyzing the various scenarios of the Parade Game in the previous section showed that enhancing the production capacity reliability of any of the crews (or all of them) can lead to improvement of the production-related outputs (such as completion time, units remained in buffer and number of lost capacity). However, maintaining a high reliable capacity (i.e. keeping the variability low) requires additional efforts (costs) by the crews on planning and coordination, while being a crew with a low reliable capacity does not. The cost input data were added to the model to explore if the investment on the crews’ capacity reliability (capacity reliability enhancement or variability reduction) is also beneficial from economic standpoint. As explained in Methodology, adding Cost Scenarios to Arrangement Scenarios resulted in 81 sub-scenarios. Tables 4.18 and 4.19 show how the Total Crews’ Cost and GC’s Total Cost changes in 81 sub-scenarios, where the first, middle, last, or all the stations are replaced by crews with a low or high reliable capacity (Arrangement Scenarios), the ratio of the indirect to total cost varies among 20, 30 and 40%, and the ratio of the planning to total cost can be 5, 10 and 15% (Cost Scenarios). The third row shows the cost of the first Arrangement scenario (Arrangement Scenario 1) where all the crews have medium variability, the rest shows how the costs change in other sub-scenarios relatively.

The outcome of the investment of the crews on the production capacity reliability are shown by Arrangement Scenarios 3, 5, 7, and 9, where the first, middle, last and all the stations’ capacity reliability are assumed to improve (from moderately to high reliability level) by additional efforts (cost) on planning. Results reveal that:

1. The crews’ investment will be beneficial for GC in any of the studied sub-scenarios (see Table 4.19). Production-related model outputs in the previous section showed that replacing any of the stations with a high reliable crew will result in lower project completion time. As a result, GC pays lower indirect cost.

2. Although the crews’ investment on capacity reliability is always beneficial for GC, its profitability for crews depends on whether what they get back is higher than what they invest or not. There are two ways for their investment to return: 1) saving in lost capacity loss due to the lower number of lost capacity when they receive lower variation from others, and 2) saving in indirect cost due to the completion time decrease. Table 4.18 represents the total loss/gain of the crews (summation of the total cost of the seven crews) in each of the 81 sub- scenarios. The investment is profitable for the crews when it is made on the first or middle stations. As expected it is not beneficial to invest on only the last station because there will be no impact on the upstream. The investment on all the stations is profitable in most of the sub-scenarios except when the largest ratio (15%) is assumed for the planning to total cost. Since, the investment on capacity reliability is taken into account as an additional crew’s planning cost by the model, assuming a higher ratio of planning to total cost means that the model assumes a higher portion of the total cost as the planning cost and subsequently, it considers a proportionally higher cost as the amount of investment. Therefore, the investment amount increases while the return does not change. This will result the lower net gain by the crews.

3. Having a crew with a low reliable capacity at the first or middle stations will increase the total cost of the crews and GC in all of the studied sub-scenarios. It can be concluded that having a crew with a low reliable capacity at the early or middle stations of a work sequence will not only worsen the productivity (as shown in the previous section), but also be unbeneficial from an economic standpoint.

Table 4.18. Summation of the Crews’ Total Cost for the 81 Parade Game sub-scenarios Cost Scenarios

illustration not visible in this excerpt

Note: Results are based on 100 simulation model replications.

Table 4.19. GC Total Cost for the 81 Parade Game sub-scenarios Cost Scenarios

illustration not visible in this excerpt

Note: Results are based on 100 simulation model replications.

4. Profitability of the investment is more sensitive to what is assumed as ratio of the planning to total cost than the other cost assumption of the model, i.e., the ratio of the indirect to the total cost. Comparing Cost Scenarios 1, 4, and 7 (or 2, 5, and 8), it can be realized that the return on the investment will not change a lot when the ratio of the indirect to total cost changes and the ratio of the planning to total cost remains the same.

Cost improvement for the crews were shown as the summation of the total cost of all the seven stations in the result (refer to Table 4.18), however, the profitability may vary crew by crew depending on their location in the work sequence. The crews at the downstream stations suffer from the work flow instability more than the upstream ones, so what they gain from the capacity reliability improvement of the other crews will be higher. On the other hand, the investment of the crews at the upstream stations may not return to them completely because their variability affects only the downstream stations. In this paper, our focus was on the entire project, so the total cost of all the stations is presented together to demonstrate how the investment to improve the capacity reliability (or decrease the variability) is profitable for the entire project. Authors will focus on profitability of each of the crews in their upcoming research paper.

As presented, crews can gain profit from their investment in two ways: reduction of their lost capacity and decrease of their indirect cost due to their completion time reduction. To simplify the model, the inventory cost (cost of keeping the unprocessed units in the buffers) was not considered in the cost calculation. Since the results showed that the number of units in the buffer decreases considerably by the capacity reliability improvement, it can be inferred that the return on the investment will be more than what has been shown if the inventory cost is taken into account.

In many construction projects, there are some internal and external incentives for GC to complete each part of the project ahead of the planned schedule (for examples he/she earns reputation for his/her on-time delivery, saves some extra time to absorb the future uncertainty). As shown in the results, GC always benefits when the crews make investment to increase their reliability, either due to the production enhancement or cost reduction (look at the results of Arrangement Scenarios 3, 5, 7, and 9). However, the investment will not return to the crews completely in some sub-scenarios. In addition, the investment does not seem reasonable to the upstream stations. Therefore, a smart action by the GC could be offering reward to those subcontractors maintaining high reliable capacity as an encouragement. The rewards can be paid from the GC’s cost saving.

4.4.3 Summary and Conclusion (Objective 4)

This study, through Objective 4, extended the original version of the Parade Game (also called Parade of Crew or Dice Game) by developing a simulation model that uses PDFs instead of the die to determine the variability level, is able to get different variability levels at different stations, and analyze the outcome from cost standpoint (in addition to the production standpoint). This paper first identified the consequence of having crews with a high or low reliable capacity at early, middle or late stations and then, explored how the capacity reliability improvement (variability reduction) can benefit the production and cost of the crews and project. Analysis of nine arrangement-related and nine cost-related scenarios (totally 81 sub-scenarios) showed that the production will enhance (completion time, number of units remained in the buffer, and number of capacities lost will reduce) if any of the stations is replaced by a crew with a high reliable capacity and the investment made to improve the capacity reliability will return in most of the scenarios (or sub-scenarios). Results also revealed that reducing the variability of the middle station has the highest impact on the productivity (compared to first or last stations), thus, if there is only one crew with a high reliable capacity in the sequence, the best decision toward the productivity improvement will be relocating him/her to the middle station.

The simulation model can be used by construction managers to help them firstly to understand the consequence of having different variability levels in different locations of the work sequence and secondly, as a decision support system, to realize how they can gain the maximum benefits out of the investment on production capacity reliability.

This study also enhances the educational value of the Parade Game. The proposed simulation model, as a user-friendly educational tool, helps construction educators/students to comprehensively understand all the aspects of variability in a construction production system. The simulation model inputs (like cost and variability inputs) are adjustable, so the model provides the opportunity to the construction educators/students to create and examine their own scenario(s) and explore the outputs.

CHAPTER 5 SUMMARY

In recent years, the project management research turns to the social sciences to solve the issues related to construction. However, research pertaining to the interpretation and investigation of the construction crews’ interdependencies during construction project is still rather limited. The main focus of the past research was on the project parties’ relationship based on the information exchange and formal communication, while the construction crews’ interdependencies and the way they can impact the project outcomes have not been well documented in industry handbooks or academic research work.

The crews during a construction project make relationship with each other on the jobsite mainly when they work in a task sequence (task-dependency) or when they work in the same working area at the same time (working area-dependency). These interdependencies can have impact on their performance, the decisions their supervisor make and their action in various ways. The main aim of this research was to better understand the existing interdependencies among the construction crews in construction phase and the impact of that on their performance. The study first focused on development of the jobsite social network and its immediate area-sharing impact on the construction crews. The research then investigated the social aspects of existing jobsite interdependencies where it explores the impact of social conformity and social learning on the construction crews. At the end, an educational simulation game was developed based on the Parade Game to investigate the dependencies of the crews/trades with different variability levels and its impact on project production and cost.

Results demonstrate that the performance of construction crews/trades is under the influence of the social aspect of the interdependencies as well as the engineering aspect. The research completed for this dissertation is summarized below to include the research objectives, associated methodology, and conclusions. The table (Table 5.1) is intended to highlight some of the key findings discussed in detail throughout the dissertation.

Table 5.1. Summary of Research Objectives, Methods, and Conclusions

illustration not visible in this excerpt

5.1 Intellectual Merit and Broader Impact

Most of the existing interdependencies in construction jobsite are inevitable and changing/eliminating them is probably infeasible, however, understanding and addressing how they play role in the project will help us to control them through better planning and leadership, consequently increasing the jobsite productivity. How are the relationships among the crews created during construction? What are the impacts of these interdependencies from production standpoint? What is the social aspect of the jobsite relationships for the construction crews? How knowing jobsite social network can be beneficial for managers/superintendents? Answering these questions requires an understanding and analyzing of the interdependencies of crews during a project.

This research contributes to the body of knowledge by taking initial steps into the topic of social network analysis at the construction site management level. No other research has specifically studied the social aspects of existing interdependencies among various actors of the construction jobsites. The research is unique as it looks at jobsite interdependencies and relationships from an exclusively analytical perspective. It uses systematic and applicable approaches/procedures to quantify the jobsite social network impact and demonstrate how the interdependencies play a role in the performance of the construction crews over the course of a project.

The research analysis methods and results will have a broader impact. First of all, one of the author’s primary goals is that the participating companies benefit from the case studies research. The results of the case studies as well as other information revealed during the research will help them focus their improvement efforts. Secondly, the analytical methods are presented step-by-step, thus, they repeatable for others to adopt and apply to their own research case within the construction industry.

In addition, the proposed approaches in this research, like SDBP, can not only be applied to construction, but also to a wide variety of decision, risk, and management problems, such as supply chains, disaster planning, transportation, security, etc. There are a wide variety of sectors and industries including public education, highways, water supply, real estate, manufacturing, and any sector with diverse and large collections of production assets that could be potential beneficiaries of this research. The approaches taken for performance modeling are transcendent, and the social network models are generally applicable as well. The findings can be applied where there is a social system - a group of people systematically sharing control of a common process - in which connections are based on interactions/relationships.

CHAPTER 6 FUTURE RESEARCH

This study takes initial steps toward the investigation of the construction crews’ interdependencies by covering the main social and engineering aspects of them. The impact can be beyond the influences covered in this study, however, covering all the aspects of the crews’ interdependencies is not in the scope of this research. In the following paragraphs, some recommendations for future research are discussed.

There are three recommendations for future research of jobsite social network (refer to Objective 1):

1) Since the type and amount of labor, equipment and materials vary from one task to another, in addition to number (frequency) and duration (severity) of the tasks, it is recommended to include the more details about the tasks in development of the networks (for example the number of workers, the required space for the equipment, or being on the critical path) to enhance the accuracy of the jobsite social network analysis.

2) The research is limited to a specific project and is based on spatial relationships. The work of construction trades in a project happens to as a function of precedence that cannot be broken. This research did not directly take into account the precedence of a schedule. It considered the working area sharing relationship which is an outcome of working schedule. It is recommended that future jobsite social network includes the schedule precedence in addition to the spatial relationships to have more accurate results.

3) Common scheduling software (such as Primavera or Microsoft Project) does not include the interdependencies among the crews. They mainly focus on the task dependencies. Jobsite social network analysis can be added to those software as a new dimension of scheduling. This new features of the software can provide additional information to the managers such as criticality of the tasks’ performer.

There are three recommendations for future research of social conformity in construction projects (refer to Objectives 2 and 3):

1) The cases studied in this research were selected commercial construction projects (reasons were discussed in Methodology), however, future research can explore the applicability of the proposed approach in the heavy/civil construction projects or other types.

2) The method examined the impact of WPR norm on the subcontractors’ WPR. This impact is positive when it helps a subcontractor with an unreliable work plan improves his/her WPR by being among other subcontractors with reliable work plan, and the impact can be negative when a subcontractor with a reliable work plan decreases his/her WPR by being among subcontractors with unreliable work plan. Positivity and negativity aspect of conformity was not in the scope of this research and can be examined in future research.

3) This study was a preliminary step in developing the relationship between the conformity and WPR in construction projects. Deeper social network analysis can provide further information with regard to the role of social network in WPR variability. For instance, a model can be developed to combine the social network centrality analysis and WPR norms to determine the role of each subcontractor in the WPR norm creation.

There are five recommendations for future research of Parade Game (refer to Objective 4):

1) Due to simplicity, the current model does not include the inventory cost and reward or punishment fee for early or late completion; a future version of Parade Game model can include those.

2) Variability is considered constant over the course of the project. If a crew is chosen to be low reliable in the model, it will remain low reliable through the completion of the project. However, the variability degree of a crew can be affected by internal and external factors (such as other crews’ performance or GC behavior, or weather) and change during the project. The next version of the game could model dynamic behavior of the crews regarding variability. This also refers to different characteristics of the players. For example, someone takes risk and spends to reduce his/her crew’s variability and someone is conservative and chooses to be low reliable when the production is unstable. Agent-based simulation modeling can be included to allocate different characteristics to the work stations and play the game in different scenarios.

3) In the current version of Parade Game, the assumption is that all the work is done by subcontractors and GC and subcontractors are on a traditional contract agreement (bid- build). In future research, contract format and agreements between GC and subcontractors can be taken into the simulation model. The game can be played with different scenarios where in each scenario the agreement type (between GC and subcontractors) is different.

4) In the current version of Parade Game, there is no formal agreement among the subcontractors, as it is based on the traditional delivery method. Relational contracting, where is based upon a relationship of trust between the parties of project, has received attention by the construction industry. A new version of Parade Game can be designed to explore the cost/benefits of relational contracting in construction projects.

5) Due to simplicity, the current version is based on a single line production system. In future, a model can be developed to explore the impact of joining flows that are more or less reliable. Turn on/off buttons can be designed for each of the joining path, so users will be able to play the game in their own scenarios.

REFERENCES

Abbasian-Hosseini, S., Hsiang, S., Leming, M., and Liu, M. (2014) From Social Network to Data Envelopment Analysis: Identifying Benchmarks at the Site Management Level. Journal of Construction Engineering and Management, 140(8), 04014028.

Abbasian-Hosseini, S. A., Nikakhtar, A., & Ghoddousi, P. (2014). Verification of lean construction benefits through simulation modeling: A case study of bricklaying process. KSCE Journal of Civil Engineering, 18(5), 1248-1260.

Abbasian-Hosseini, S.A. , Liu, M. & Hsiang, S.M. 2015, 'Social Network Analysis for Construction Specialty Trade Interference and Work Plan Reliability' In:, Seppänen, O., González, V.A. & Arroyo, P., 23rd Annual Conference of the International Group for Lean Construction. Perth, Australia, 29-31 Jul 2015. pp 143-152

AbouRizk, S. M., and Halpin, D. W. (1992). “Statistical properties of construction duration data,” Journal of Construction Engineering and Management, 118(3), 525-544.

Ahuja, H. N., and Campbell,W. J. (1988). Estimating from concept to completion, PrenticeHall, Englewood Cliffs, NJ.

Alarcón, L.F. and Ashley, D.B. (1999). “Playing games: Evaluating the impact of lean production strategies on project cost and schedule.” Proc. 7th annual conference of international group for lean construction, Lean Construction Institute, Berkeley, CA, 263- 274

Altiok, T. and Melamed, B. (2007). Simulation modeling and analysis with ARENA. Elsevier, p 66.

Argote, L., Ingram, P., Levine, J., and Moreland, R. (2000). “Introduction: Knowledge Transfer in Organizations: Learning from the Experience of Others,” Organizational Behavior and Human Decision Processes, 82 (1), 1-8.

Asch, S. E. (1951). “Effects of group pressure upon the modification and distortion of judgment,” In H. Guetzkow (Ed.), Groups, leadership and men. Pittsburgh, PA: Carnegie Press, 117-190.

Assaf, S., Bubshait, A., Atiyah, S. and Al-Shahri, M. (2001). "The management of construction company overhead costs," International Journal of Project Management, 19(5), 295-303.

Bala, V. and Goyal, S. (1998). “Learning from Neighbors,” The Review of Economic Studies, 65 (3), pp. 595-621.

Ballard, G. (1993). “Lean construction and EPC performance improvement,” Proc. 1st Annual Meeting of the International Group for Lean Construction, Lean Construction Institute, La Jolla, CA, 79-92.

Ballard, G. (1994). “The last planner,” Proc. Conf. of the Northern California Construction Institute, Lean Construction Institute, La Jolla, CA, 1-8.

Ballard, G., and Howell, G. (1998). “Shielding production: essential step in production control,” Journal of Construction Engineering and Management, 124 (1), 11-17.

Ballard, H. G. (2000). The last planner system of production control, Doctoral dissertation. Univ. of Birmingham, Birmingham, UK.

Ballard, G., Koskela, L., Howell, G., and Tommelein, I. (2005). ”Discussion of “Improving Labor Flow Reliability for Better Productivity as Lean Construction Principle” by H. Randolph Thomas, Michael J. Horman, R. Edward Minchin Jr., and Dong Chen,” Journal of Construction Engineering and Management, 131(5), 615-616.

Becker, T. C. (2012). Improving the predictability of construction project outcomes through project level indirect construction cost practices, Doctoral dissertation, North Carolina State Univ., Raleigh, NC.

Becker, T., Jaselskis, E., and El-Gafy, M. (2014). ”Improving Predictability of Construction Project Outcomes through Intentional Management of Indirect Construction Costs,” Journal of Construction Engineering and Management, 140(6), 04014014.

Bertelsen, S., Henrich.G, Koskela, L., and Rooke, J. (2007). “Construction physics,” Proceedings IGLC-15, Michigan, U.S.

Bikhchandani, S., Hirshleifer, D. and Welch, I. (1998). “Learning from the Behavior of Others: Conformity, Fads, and Informational Cascades,” The Journal of Economic Perspectives, 12 (3), 151-170.

Bresnen, M., Swan, J. and Goussevskaia, A. (2005). “Managing projects as complex social settings,” Building Research &Information, 33, 487-493.

Buechel, B., Hellmann, T., & Klößner, S. (2015). “Opinion dynamics and wisdom under conformity,” Journal of Economic Dynamics and Control, 52, 240-257.

Charnes, A. A., Cooper, W. W., and Rhodes, E. (1978). “Measuring the efficiency of decision making units,” European Journal of Operation Research, 2(6), 429-444.

Chinowsky, P., Diekmann, J., and Galotti, V. (2008). “Social Network Model of Construction,” Journal of Construction Engineering and Management, 134(10), 804-812.

Chinowsky, P., Diekmann, J., and O’Brien, J. (2010). “Project Organizations as Social Networks,” Journal of Construction Engineering and Management, 136, SPECIAL ISSUE: Governance and Leadership Challenges of Global Construction, 452-458.

Chinowsky, P., Taylor, J., and Di Marco, M. (2011). “Project Network Interdependency Alignment: New Approach to Assessing Project Effectiveness,” Journal of Management in Engineering, 27(3), 170-178.

Choo, H. J., Tommelein, I. D., Ballard, G., and Zabelle, T. R. (1999). “WorkPlan: Constraint-based database for work package scheduling,” Journal of Construction Engineering and Management, 125(3), 151-160.

Cialdini, R. B., & Trost, M. R. (1998). Social influence: Social norms, conformity and compliance, The Handbook of Social Psychology, 4th edition, McGraw-Hill, New York, NY, Chapter 21,151-192.

Costantino, N., Pietroforte, R., & Hamill, P. (2001). “Subcontracting in commercial and residential construction: an empirical investigation,” Construction Management and Economics, 19(4), 439-447.

De Nooy, W., Mrvar, A., and Batagelj, V. (2005). Exploratory Network Analysis with Pajek®, Cambridge University Press, New York.

Dvir, D., Raz, T., & Shenhar, A. J. (2003). “An empirical analysis of the relationship between project planning and project success.” Int. J. Project Manage., 21(2), 89-95.

Dvir, D., & Lechler, T. (2004). “Plans are nothing, changing plans is everything: the impact of changes on project success.” Research policy, 33(1), 1-15.

Dogan, S., Arditi, D., Gunhan, S., and Erbasaranoglu, B. (2013). “Assessing Coordination Performance Based on Centrality in an E-mail Communication Network,” Journal of Engineering in Management, 31(3), 04014047.

El-Mashaleh, M. (2003). Firm performance and information technology utilization in the construction industry: an empirical study, Doctoral dissertation, University of Florida, Gainesville, FL.

El-Mashaleh, M., O’Brien, W., and Minchin, R. (2006). “Firm performance and information technology utilization in the construction industry,” Journal of Construction Engineering and Management, 132(5), 499-507.

El-Mashaleh, M. S., Minchin, R. E. Jr., and O’Brien, W. J. (2007). “Management of construction firm performance using benchmarking,” Journal of Engineering in Management, 23(1), 10-17.

El-Mashaleh, M.S., Rababeh, S. M., and Hyari, K. H. (2010). “Utilizing data envelopment analysis to benchmark safety performance of construction contractors,” Journal of Project Management, 28(1), 61-67.

Ellison, G. and Fudenberg, D. (1993). “Rules of Thumb for Social Learning,” Journal of Political Economy, 101 (4), pp. 612-643.

Enshassi, A., Rashid Abdul Aziz, A., & El Karriri, A. A. (2008). “Investigating the overhead costs in construction projects in Palestine,” Journal of Financial Management of Property and Construction, 13(1), 35-47.

Fehr, E., & Fischbacher, U. (2004). “Social norms and human cooperation,” Trends in cognitive sciences, 8(4), 185-190.

Fente, J., Schexnayder, C., and Knutson, K. (2000). ”Defining a Probability Distribution Function for Construction Simulation,” Journal of Construction Engineering and Management, 126(3), 234-241.

Freeman, C., and Seppänen, O. (2014). “Social Aspects Related to LBMS Implementation - Case Study,” In proceedings of 22nd Annual Conference of the International Group for Lean Construction, June, Oslo, Norway.

González, V., Alarcón, L., and Mundaca, F. (2007). “Investigating the Relationship Between Planning Reliability and Project Performance: A Case Study,” Proc. 15th Annual Conference of the International Group for Lean Construction, Lean Construction Institute, Michigan, USA, 98-108.

González, V., Alarcón, L., Maturana, S., Mundaca, F., and Bustamante, J. (2010). ”Improving Planning Reliability and Project Performance Using the Reliable Commitment Model,” Journal of Construction Engineering and Management, 136(10), 1129-1139.

Guo, S. J. (2002). “Identification and resolution of work space conflicts in building construction,” Journal of Construction Engineering and Management, 128(4), 287-295.

Hajifathalian, K., Wambeke, B., Liu, M., and Hsiang, S. (2012). “Effects of Production Control Strategy and Duration Variance on Productivity and Work in Process: Simulation- Based Investigation,” Journal of Construction Engineering and Management, 138(9), 1035- 1043. Han, S., Park, M. (2011). “Interactive parade game: impact of managerial reactions to workflow variability,” Journal of Information Technology in Construction (ITcon), 16, 105- 118.

Henrich, J., and Boyd, R. (2001). “Why People Punish Defectors: Weak Conformist Transmission can Stabilize Costly Enforcement of Norms in Cooperative Dilemmas,” Journal of Theoretical Biology, 208, 79-89.

Hossain, L. (2009). “Communications and coordination in construction projects,” Construction Management and Economics, 27(1), 25-39.

Hinze, J. and Tracey, A. (1994). ”The ContractorǦSubcontractor Relationship: The Subcontractor's View,” Journal of Construction Engineering and Management, 120(2), 274- 287.

Hosseini, S. A., Nikakhtar, A., Wong, K. Y., & Zavichi, A. (2012). Implementing Lean Construction Theory into Construction Processes' Waste Management. In Reston, VA: ASCEProceedings of the 2011 International Conference on Sustainable Design and Construction| d 20120000. American Society of Civil Engineers.

Hosseini, A., Nikakhtar, A., & Ghoddousi, P. (2012). Flow production of construction processes through implementing lean construction principles and simulation. IACSIT International Journal of Engineering and Technology, 4(4), 475-479.

Howell, G., Laufer, A., and Ballard, G. (1993). “Uncertainty and Project Objectives,” Project Appraisal Journal, 8(1), 37-43.

Howell, G. A., Ballard, G., and Hall, J. (2001). “Capacity utilization and wait time: A primer for construction,” Proc. 9th Annual Conference of the International Group for Lean Construction, Lean Construction Institute, Singapore, Singapore.

Howell, G. and Liu, M. (2012). “The Oops Game: How Much Planning Is Enough?,” In:, Tommelein, I.D. & Pasquire, C.L., Proc. 20th Annual Conference of the International Group for Lean Construction, Lean Construction Institute, San Diego, USA. Hsieh, T. (1998). ”Impact of Subcontracting on Site Productivity: Lessons Learned in Taiwan,” Journal of Construction Engineering and Management, 124(2), 91-100.

Javernick-Will, A. (2012). ”Motivating Knowledge Sharing in Engineering and Construction Organizations: Power of Social Motivations,” Journal of Management in Engineering, 28(2), 193-202.

Jolliffe, I. T. (2004). Principal Component Analysis, 2nd edition, Springer, New York. Kendal, J., Giraldeau, L., and Laland, K. (2009). “The evolution of social learning rules: Payoff biased and frequency-dependent biased transmission,” Journal of Theoretical Biology, 260, 210-219.

Kim, Y., and Jang, J. (2005). “Case Study: An application of Last Planner to heavy civil construction in Korea,” Proc. 13th Annual Conference of the International Group for Lean Construction, Lean Construction Institute, Sydney, Australia, 405-411.

Knutson, K., Shexnayder, C., Fiori, C., and Mayo, R. (2009). Construction Management Fundamentals. McGraw-Hill Higher Education, 2nd edition, NY. Koskela, L. J. (2000). An exploration towards a production theory and its application to construction. Doctoral dissertation, Teknillinen Korkeakoulu, Helsinki, Finland.

Koskela, L., Howell, G., Ballard, G. and Tommelein, I. (2002). “The foundations of lean construction,” Chapter-14, Design and Construction: Building in Value, ButterworthHeinemann, Oxford, U.K., 211-226.

Lamberson, P. (2010). “Social Learning in Social Networks,” The B.E. Journal of Theoretical Economics, 10 (1).

Lai, M., Huang, H., and Wang, W. (2011). “Designing a knowledge-based system for benchmarking: A DEA approach,” Knowledge-Based Systems, 24, 662-671

Lapinski, M. K., & Rimal, R. N. (2005). “An explication of social norms,” Communication Theory, 15(2), 127-147.

Lee, J. S., A. Norouzi, and Y. R. Kim. Determining Specimen Geometry of Cylindrical Specimens for Direct Tension Fatigue Testing of Asphalt Concrete. Journal of Testing and Evaluation, in press, 2015.

Lin, S. (2014). “An Analysis for Construction Engineering Networks,” Journal of Construction Engineering and Management, 141(5), 04014096. Lindhard, S. (2014). ”Understanding the Effect of Variation in a Production System,” Journal of Construction Engineering and Management, 140(11), 04014051.

Liu, G., Wong, L., & Chua, H. N. (2009). “Complex discovery from weighted PPI networks,” Bioinformatics, 25(15), 1891-1897. Liu, M., and Ballard, G. (2008). “Improving labor productivity through production control,” Proc. 16th Annual Conference of the International Group for Lean Construction, Lean Construction Institute, Manchester, UK, 657-666.

Liu, M., Ballard, G., and Ibbs, W. (2011). ”Work Flow Variation and Labor Productivity: Case Study,” Journal of Management in Engineering, 27(4), 236-242.

Luu, V. T., Kim, S., and Huynh, T. (2008). “Improving project management performance of large contractors using benchmarking approach”, International Journal of Project Management, 26 (7), 758-769. McCabe, B., Tran, V., and Ramani, J. (2005). “Construction prequalification using data envelopment analysis,” Canadian Journal of Civil Engineering, 32(1), 183-193. McFadden D. and Train K. (1996), “Consumers' Evaluation of New Products: Learning from Self and Others,” Journal of Political Economy, 104 (4), 683-703.

McCord, P. J. (2010). Subcontractor perspectives: Factors that most affect their relationships with general contractors-A pacific Northwest study, Doctoral dissertation, Washington State University, Pullman, WA. Moghaddas Tafreshi, S. N., and A. H. Norouzi. Application of Waste Rubber to Reduce the Settlement of Pavement Systems. Journal of Geomechanics and Engineering, Vol. 9, No. 1, 2015. Moghaddas Tafreshi, S. N., and A. H. Norouzi. Bearing Capacity of a Square Model Footing on Sand Reinforced with Shredded Tire-An Experimental Investigation, Journal of Construction and Building Materials, Vol. 35, Pages 547-556, 2012.

Moreno, J. L. (1960). The sociometry reader, The Free Press, Glencoe, IL.

Nikakhtar, A., Hosseini, A. A., & Wong, K. Y. (2012). Sensitivity analysis of construction processes using computer simulation: a case study. Advanced Science Letters, 13(1), 680-684.

Nikakhtar, A., Hosseini, A. A., Wong, K. Y., & Zavichi, A. (2015). Application of lean construction principles to reduce construction process waste using computer simulation: a case study. International Journal of Services and Operations Management, 20(4), 461-480

Norouzi A., M. Sabouri, and Y. R. Kim. Evaluation of the Fatigue Performance of High RAP Asphalt Mixtures. Proceedings of 12th International Society for Asphalt Pavements, Raleigh, NC, 2014. Norouzi, Amirhossein, and Y. Richard Kim. "Mechanistic evaluation of fatigue cracking in asphalt pavements." International Journal of Pavement Engineering (2015): 1-17.

Norouzi, A. H., and Y. R. Kim. Ruggedness Study on the Dynamic Modulus Testing of Asphalt Concrete in Indirect Tension Mode. Journal of Testing and Evaluation, 2015, in press.

Opsahl, T., Agneessens, F., & Skvoretz, J. (2010). “Node centrality in weighted networks: Generalizing degree and shortest paths,” Social Networks, 32(3), 245-251.

Park, H., Han, S., Rojas, E., Son, J., and Jung, W. (2011). “Social Network Analysis of Collaborative Ventures for Overseas Construction Projects,” Journal of Construction Engineering and Management, 137(5), 344-355.

Pilateris, P., and McCabe, B. (2003). “Contractor financial evaluation model,” Canadian Journal of Civil Engineering, 30(3), 487-499. Priven, V. and Sacks, R. (2015). “Effects of the Last Planner System on Social Networks among Construction Trade Crews,” Journal of Construction Engineering and Management, 141(6), 04015006. Proctor Jr., J. (1996). ”Golden Rule of Contractor-Subcontractor Relations,” Practice Periodical on Structural Design and Construction, 1(1), 12-14. Pryke, S. D. (2004). “Analysing construction project coalitions: exploring the application of social network analysis,” Construction Management and Economics, 22(8), 787-797. Pryke, S. D., Zagkli, G., & Kougia, I. (2011).

“Resource provision ego-networks in small Greek construction firms,” Building Research and Information, 39(6), 616-636. Pryke, S. (2012). Social network analysis in construction. John Wiley & Sons, West Sussex, UK. Sacks, R. (2004). “Towards a lean understanding of resource allocation in a multi-project sub-contracting environment,” In proceedings of 12st Annual Conference of the International Group for Lean Construction, 3-5 Aug., Helsingør, Denmark. Senior, B. A. (2011). “Enhancing the Construction Parade of Trades Game,” 47th ASC Annual International Conference Proceedings, Nebraska, Omaha.

Seppänen, O. (2009). Empirical Research on the Success of Production Control in Building Construction Projects, Doctoral Dissertation, Helsinki University of Technology, Finland, pp. 175.

Shapira, A., & Laufer, A. (1993). “Evolution of involvement and effort in construction planning throughout project life.” Int. J. Project Manage., 11(3), 155-164. Takus, D. A. and Profozich, D. M. (1997). “ARENA software tutorial,” Proc. 1997 Winter Simulation Conference, Piscataway, N.J., 541-544. Tavistock Institute (1966). Interdependence and Uncertainty: A Study of the Building Industry. Tavistock, London, UK. Taylor, J.E., Levitt, R., and Villarroel, J.A. (2009).

“Simulating learning dynamics in project networks,” Journal of Construction Engineering and Management, 135(10), 1009-1015. Tekla. (2015), Porin Puuvilla, the Tekla Global BIM Awards winner 2013, Retrieved on Oct. 10, 2015, from http://www.tekla.com/us/references/project-puuvilla-factory-shopping-center- bim-collaboration Thomas, H., Horman, M., de Souza, U., and Zavřski, I. (2002). ”Reducing Variability to Improve Performance as a Lean Construction Principle,” Journal of Construction Engineering and Management, 128(2), 144-154.

Thomas, H., Horman, M., Minchin, R., Jr., and Chen, D. (2003). ”Improving Labor Flow Reliability for Better Productivity as Lean Construction Principle,” Journal of Construction Engineering and Management, 129(3), 251-261. Thomas, H. R., Riley, D. R., & Sinha, S. K. (2006). “Fundamental principles for avoiding congested work areas-A case study,” Practice Periodical on Structural Design and Construction, 11(4), 197-205.

Thomas, H. and Flynn, C. (2011). ”Fundamental Principles of Subcontractor Management,” Practice Periodical on Structural Design and Construction, 16(3), 106-111.

Thornton, R. and Thompson, P. (2001). “Learning from Experience and Learning from Others: An Exploration of Learning and Spillovers in Wartime Shipbuilding,” The American Economic Review, 91(5), 1350-1368.

Thorpe, T. and Mead, S. (2001). “Project-Specific Web Sites: Friend or Foe?,” Journal of Construction Engineering and Management, 127(5), 406-413.

Tommelein, I., Riley, D., and Howell, G. (1999). ”Parade Game: Impact of Work Flow Variability on Trade Performance,” Journal of Construction Engineering and Management, 125(5), 304v310.

Vitner, G., Rozenes, S., and Spraggett, S. (2006). “Using data envelope analysis to compare project efficiency in a multi-project environment,” International Journal of Project Management, 24(4), 323-329. Wambeke, B., Hsiang, S., and Liu, M. (2011). ”Causes of Variation in Construction Project Task Starting Times and Duration,” Journal of Construction Engineering and Management, 137(9), 663-677. Wambeke, B., Liu, M., and Hsiang, S. (2012). “Using Pajek and Centrality Analysis to Identify a Social Network of Construction Trades,” Journal of Construction Engineering and Management, 138(10), 1192-1201.

Wambeke, B., Liu, M., and Hsiang, S. (2014). ”Task Variation and the Social Network of Construction Trades,” Journal of Management in Engineering, 30(4), 05014008.

Wang, Y., Gibson, Jr., G., and Huang, J. (2008) A Study of Preproject Planning and Project Success Using ANN and Regression Models. AEI 2008: pp. 1-10. Wang, Y. R., & Gibson, G. E. (2010). “A study of preproject planning and project success using ANNs and regression models.” Auto.in Constr., 19(3), 341-346.

Wang, C., Liu, M., Hsiang, S., and Leming, M. (2012). ”Causes and Penalties of Variation: Case Study of a Precast Concrete Slab Production Facility,” Journal of Construction Engineering and Management, 138(6), 775-785.

Watkins, M., Mukherjee, A., Onder, N., & Mattila, K. (2009). “Using agent-based modeling to study construction labor productivity as an emergent property of individual and crew interactions,” Journal of Construction Engineering and Management, 135(7), 657-667.

West, J. (2014). “Collaborative Patterns and Power Imbalance in Strategic Alliance Networks,” Journal of Construction Engineering and Management, 140(6), 04014010.

Wong, K., Unsal, H., Taylor, J. E., and Levitt, R. (2010). “Global dimension of robust project network design,” Journal of Construction Engineering and Management, 136(4), 442-451.

Zhu, J. (2009), Quantitative Models for Performance Evaluation and Benchmarking: Data Envelopment Analysis with Spreadsheets, 2nd Edition, Springer, Worcester Polytechnic Institute, Worcester, MA, USA.

APPENDICES

APPENDIX A - ADJACENCY MATRICES OF CASE STUDY 1

Adjacency matrix based on frequency; Week 1

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 2

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 3

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 4

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 5

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 6

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 7

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 8

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 9

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 10

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 11

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 12

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 13

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 14

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 15

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 16

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 17

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 18

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 19

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 20

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 21

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 22

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 23

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 24

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 25

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 26

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 27

illustration not visible in this excerpt

Adjacency matrix based on frequency; Week 28

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 1

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 2

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 3

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 4

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 5

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 6

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 7

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 8

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 9

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 10

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 11

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 12

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 13

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 14

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 15

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 16

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 17

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 18

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 19

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 20

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 21

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 22

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 23

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 24

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 25

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 26

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 27

illustration not visible in this excerpt

Adjacency matrix based on severity; Week 28

illustration not visible in this excerpt

APPENDIX B - ADJACENCY MATRICES OF CASE STUDY 2

Adjacency matrix of subcontractor crews, Month 2

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 3

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 4

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 5

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 6

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 7

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 8

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 9

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 10

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 11

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 12

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 13

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 14

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 15

illustration not visible in this excerpt

APPENDIX C - ADJACENCY MATRICES OF CASE STUDY 3

Adjacency matrix of subcontractor crews, Month 1

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 2

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 3

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 4

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 5

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 6

illustration not visible in this excerpt

Adjacency matrix of subcontractor crews, Month 7

illustration not visible in this excerpt

APPENDIX D - INTRODUCTION TO DATA ENVELOPMENT ANALYSIS

DEA is one of the common techniques of benchmarking introduced by Charnes et al. (1978), which measure the productivity by the ratio between a weighted sum of inputs and a weighted sum of outputs. It uses operations research techniques to automatically calculate the weights assigned to the inputs and outputs of the production units being assessed. DEA generates a surface, called “empirical frontier”, which connects the relatively best DMUs (the most efficient ones). The empirical frontier can be used as a reference for efficiency improvement.

The original method of DEA is known as CCR (Charnes, Cooper, and Rhodes (1978)), which maximize the ratio of weighted outputs to weighted inputs in such a way that the similar ratio in every DMU is less than or equal to one. Two main alternative approaches are available for the basic DEA model: input-oriented and output-oriented. The following DEA model will be used in the research is an output-oriented Constant-Return to-Scale (CRS) DEA model, where the outputs are maximized and the inputs are kept at their current levels:

illustration not visible in this excerpt

Where DMUo represents one of the n DMUs under evaluation, and yr0 and xi0 are the rth output and ith input for DMUo, respectively. θ is the uniform proportional increase in the DMUo’s outputs and its maximum amount (θ*) is known as the DMUo’s efficiency score. Technically speaking, the presented model states that if there is a weighting vector λ that solves the equation (1), and θ* ≥ 1, then it is concluded that the DMUo is inefficient. The optimal value to (1) is θ*=1, which shows that the current output levels cannot be increased proportionally and the DMUo is efficient and the input-output combination lies on the frontier.

Consider a simple numerical example shown in the Table A-1, where we have five DMUs (for example five construction companies) doing the same activity with the same amount of daily work (100 man-hours), but producing different amount of products and profits (different outputs).

Table A-1: Outputs of the DMUs considered in the numerical example

illustration not visible in this excerpt

Figure 8-1, drawn based on the outputs (Y1 and Y2), illustrates that the DMU3 and DMU5 define the empirical frontier that envelopes the other DMUs. This means that other companies cannot obtain a better score than DMU3 and DMU5 in both dimensions (Y1 and Y2), and that these two companies are efficient. DMU3 produces the highest amount of products, while DMU5 achieves the most profit. The remaining DMUs are inefficient. Note that to draw the empirical frontier in more complicated problems, when we have multiple outputs and multiple inputs, we can use summation of inputs as the X and summation of outputs as the Y (summation can be linear or non-linear) of the two dimensional coordination system.

illustration not visible in this excerpt

Figure A-1 : DEA empirical frontier for the numerical example Now, the model analysis is performed for DMU1 as an example,

illustration not visible in this excerpt

A set of unique optimal solutions of θ* = 1.42, λ3 = 0.571, λ5 = 0.429 and λ1, λ2, λ4 = 0 is obtained, which indicates that DMU1 is not efficient and DMU3 and DMU5 are the benchmarks for DMU1. We can infer from the values obtained for the θ* and λs that DMU1 is able to enhance its outputs to 1.42 times of its current level and reach the frontier (1' in the Figure A-1) by imitating the DMU3 and DMU5 by 57.1% and 42.9% respectively.

Now, if the model analysis is done for one of our two best-performing DMUs, DMU3, we obtain θ* = 1.00, λ3 = 1 and λj = 0 (j ≠ 3), which indicates that DMU3 is efficient and does not need to imitate anyone, and is located on the efficient frontier (see Figure A-1).