understanding and managing iterative error and change cycles in construction

S. Lee and F. Peña-Mora: Iterative Error and Change Cycles 35

Published online in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/sdr

Understanding and managing iterative errorand change cycles in construction

SangHyun Lee*a and Feniosky Peña-Morab

System Dynamics Review Vol. 23, No. 1, (Spring 2007): 35–60Published online in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/sdr.359Copyright © 2007 John Wiley & Sons, Ltd.

35

Abstract

Errors and changes in construction often result in significant schedule and cost overruns

affecting project performance. To understand the nature of these errors and changes and to

ultimately reduce their detrimental impacts on project performance, this paper presents a systemdynamics-based construction model, which focuses on the dynamics of error and change

management in construction, including quality management, scope management, the request

for information process, and the decision-making process for the approval of changes,and their consequent detrimental impacts on project performance. In particular, the developed

model integrates several concepts in traditional network-based tools to enhance the applica-

bility of the model. Describing the dynamic behaviors generated by the developed model andapplying the model to a couple of real-world construction projects, this paper concludes

that: (1) realism should be added to schedule planning; (2) an efficient coordination process

is needed; (3) proactive contingency plans need to be taken into consideration; and (4) integrationof network-based tools and system dynamics-based models can contribute to management of

errors and changes. Copyright © 2007 John Wiley & Sons, Ltd.

Syst. Dyn. Rev. 23, 35–60, (2007)

Keywords: construction management; error and change management

Introduction

One of the major drivers of uncertainty and complexity in construction isiterative cycles caused by errors and changes (Lee et al., 2003). Errors, changes,and consequent conflicts are common and lead to significant schedule and costoverrun. For example, during the execution of a project, the time taken torectify errors is estimated to be 11 percent of the total working hours allocatedfor a project and the cost to correct these errors is approximately 6 percent ofproduction costs (Josephson and Hammarlund, 1999). The cost to implementchanges is estimated to be 5.1–7.6 percent of the total project cost (Cox et al.,1999). This could translate into $50 billion being spent annually on new changeorders by the construction industry in the U.S.A. alone (Ibbs et al., 1998).

However, widely used traditional network-based tools, such as the CPM/PERT/PDM network (Critical Path Method, DuPont Inc. and Remington Rand,1958; Program Evaluation and Review Technique, U.S. Navy, Booz-Allen

SangHyun Lee holds a

PhD and MS from MIT

in ConstructionManagement and

Information

Technology and is anAssistant Professor of

Construction

Engineering andManagement at the

University of Alberta.

Prior to joining theUniversity of Alberta,

he worked as a

consultant at CRA(Charles River

Associates)

International andtaught at MIT. His

research area includesconcurrent design and

construction

management, changemanagement,

proactive buffer

management, hybridsimulation for large-

scale design and

construction, andinformation

visualization. His

current researchprojects are dynamic

project management,

reliability and stabilityschedule buffering

approach, and

visualization forconstruction progress

monitoring.

Feniosky Peña-Mora

holds an ScD and MS

from MIT and is aProfessor of

Construction

a Civil and Environmental Engineering Department, University of Alberta, Edmonton, Alberta T6G 2W2, Canada.E-mail: [email protected] Civil and Environmental Engineering Department, University of Illinois at Urbana-Champaign, IL 61801,U.S.A.

* Correspondence to: SangHyun Lee.

Received March 2005; Accepted December 2006

36 System Dynamics Review Volume 23 Number 1 Spring 2007


Engineering and

Information

Technology at theUniversity of Illinois

at Urbana-Champaign.

His research areaincludes information

technology support for

collaboration, changemanagement, conflict

resolution, and

process integrationduring design and

development of large-

scale civil engineeringsystems. His current

research projects are

collaborativepreparedness,

response and recovery

against disastersinvolving critical

physicalinfrastructures (CP2R),

dynamic project

management (DPM),and dynamic conflict

avoidance, mitigation

and resolutionmethodology (D-

CART). His previous

research projectsinclude support for

the development of

disaster aware/resistant critical

infrastructure (DARE),

mobile interactivecollaboration

environments (MICE),

and meetingmanagement and

facilitation on

collaborativeengineering

environments

(CAIRO).

Hamilton and Lockheed Co., 1958; Precedence Diagramming Method, IBMCo., 1964), do not explicitly take into account errors and changes and do notwork well when construction is heavily constrained by either time or re-sources in a dynamic environment (Hegazy, 1999). Thus, they may have diffi-culty capturing dynamic feedback processes caused by errors and changes,which are very common in construction (Lyneis et al., 2001).

Significant research in system dynamics has been undertaken to addresssome of the issues related to errors and changes in projects, emphasizing theimpact of the rework cycle on project performance (Cooper, 1980; Richardsonand Pugh, 1981; Abdel-Hamid, 1984; Ford and Sterman, 1998; Lyneis et al.,2001). In the construction arena, Park and Peña-Mora (2003) have developed amodel that deals with the rework cycle customized to construction character-istics (e.g., construction projects prefer to undertake “extra work” rather than“rework” because “rework” may accompany physical demolition of alreadybuilt components).

Based on the above contributions to an understanding of the rework cyclefor effective project or construction management, this paper further pursuesin-depth understanding of errors and changes in construction. In particular,the applicability of system dynamics-based models in the construction industryand the identification of the mechanism associated with errors and changesare this paper’s main interests. Specifically, this paper proposes a meansto integrate the fundamentals of the de facto network methods in the A/E/C(architectural/engineering/construction) industry into the system dynamicsmodel. Thus, the current practice in the A/E/C industry can be kept and, at thesame time, new features for better understanding of the dynamics of errorsand changes can be added to traditional network-based tools. In addition,this paper focuses on the role of additional work scope generated by rectifyingerrors and implementing changes as a source to disrupt the whole construc-tion. Considering the fact that construction plans are developed based on anestimated initial work scope, adding work scope during actual constructionmay be very disruptive, corrupting procurement, resource allocation, man-power loading plans, and so forth because it may require the changes of allinitial plans. Further, when the time given to account for consequent changesis insufficient (e.g., fast track projects), the impact can be even more severe.In this sense, the identification of the underlying mechanism of error andchange impacts on construction, highlighting additional work scope as asource to generate complex dynamics, and can greatly contribute to effectivemanagement.

Integration of network-based tools and system dynamics models

The de facto standard of project management modeling in the A/E/C industryis network-based models such as CPM, PERT, and PDM (Senior and Halpin,



1998). CPM allows the logical analysis and manipulation of a network todetermine the best overall program of operation, while PERT was developedto deal with uncertainty in projects by incorporating probabilities into theduration of project activities. Since the development of CPM and PERT, manynetwork-based tools have been developed supplementing CPM and PERT. Forexample, PDM added different types of precedence relationships, such asStart-to-Start, Start-to-Finish, Finish-to-Start, Finish-to-Finish with leads andlags,1 between activities to the traditional CPM. Even though their rigid andstatic features make them difficult to represent the dynamic construction pro-cess, they have been widely used in the construction management area mainlyowing to their easy applicability to diverse projects (Rodrigues and Williams,1998; Park and Peña-Mora, 2003). In order to incorporate their applicability,the fundamentals of network-based models are integrated into the systemdynamics-based Dynamic Design and Construction Project Model (D2CPM).

Modeling concepts in network-based models

In the D2CPM model, concepts from network-based models are representedas follows. First, there are two basic stocks, “WorkToDo” (WTD) and“WorkCompleted” (WC), as seen in Figure 1. The rate that tasks in the WTDstock move to the WC stock is determined by productivity and is labeled“WorkRate”. In network-based models, productivity will be assumed to beconstant, and work constraints among activities (i.e., precedence relation-ships) drive whether and when the succeeding activity can start. Thus,“WorkIntroductionRate”, which assigns the quantity of the WTD stock, isconstrained by the variable “WorkAvailable” determined by the constraintscaused from precedence relationships in network-based models.

In addition, to capture dynamic concurrency that represents work depend-encies within an activity, internal concurrence (Ford and Sterman, 1998) hasbeen adopted. Thus, “WorkAvailable” is also constrained by internal concur-rence from the existing system dynamics model.

Fig. 1. Modeling work constraints



Table 1. Simulation setting

Network Reliability Sensitivity QM and SM RFI CCMand stability Thoroughness response time period

CPM CASE N/A N/A N/A N/A N/A

CASE 1 0.85 7 days

CASE 2 0.5 1 5 days

CASE 3 0.9

Fig. 2. CPM CASE: behavior of WorkToDo and WorkCompleted stock

In order to investigate whether the developed model generates thesame behaviors observed in network-based models, the CPM CASE wassimulated with three concurrent activities: a final design, an excavation, anda superstructure activity.2 Each activity has a 100-day duration and Start-to-Start relationship with a lag of 60, as seen in Table 1. For simplicity, WUis used as a hypothetical work unit, and 1000 WU is assumed to be equivalentto 1 day’s work. This scenario follows the CPM assumptions that everythingwill be performed as expected with a uniform production rate. As seenin Figure 2, the CPM CASE confirms that the WTD and WC stocks are theexpected straight lines and that the total duration is 220 days.

Modeling dynamic error and change cycles

Differing from the CPM assumptions, there can be errors and changes inreality. Errors and changes usually cause non-value-adding iterations that



deteriorate productivity and quality. These iterations are originally set accord-ing to an initial work scope. Thus, when an additional work amount is intro-duced by errors and changes, the designed productivity and quality mayno longer be maintainable. The deterioration of productivity and qualityare one of the main sources of multiple feedback processes and the corres-ponding actions necessary to recoup the deficit often generate unanticipatedside effects.

Suppose an excavation activity can be completed by 10 backhoe loaderswithin 10 working days. If an additional 20 percent of work is added to theinitial work amount to rectify errors and implement changes, a schedule delayof 2 days will occur (i.e., initial productivity will then require 12 days with10 backhoe loaders). Managers will then typically utilize the tool of overtimeto keep the initial 10 days. As expected, prolonging work hours will takecare of the additional work, but the extended work hours simultaneously helpdeteriorate the workforce’s productivity due to increasing fatigue of the work-ers. Therefore, unexpected effects caused by various compensatory mechan-isms need to be anticipated.

In this context, identifying how errors and changes are introduced, how theyproduce additional work scope and, ultimately, how they affect performance,would be the key to the success of managing errors and changes in construc-tion projects. Concentrating on the additional work scope as the source thataffects construction performance, the following sections will illustrate a modelthat has been designed to show the impact of errors and changes in the con-struction process.

ERROR AND CHANGE MANAGEMENT The basic work execution associated with errordiscovery is shown in Figure 3, extending the model structure in Figure 1. For

Fig. 3. Work

execution with errormanagement (adapted

from Pugh-Roberts

Associates, 1980, asreferred to in Sterman,

2000; Ford and

Sterman, 1998; Parkand Peña-Mora, 2003)



example, after tasks in the WTD stock are executed with the available workrate that considers resource availability, tasks done will await Quality Manage-ment (QM). The main idea behind this structure takes into account that justbecause all of the work in a project has been performed, there is no guaranteethat the work would have been done correctly. Some of the work may needto be re-executed (i.e., errors) (A in Figure 3) and this is captured by reliability(B in Figure 3), the degree to which the performed task has been done correctlyduring actual execution (Lee et al., 2005). For example, if a final design activityhas 90 percent reliability, the amount of errors would be 10 percent ofthe total work scope of the final design activity. However, some portion of theerrors may not be identified during the QM process. In other words, the re-maining work often needs to proceed without recognition of the existence ofthe predecessor’s errors. This is because the applied QM techniques are oftennot perfectly executed or tight schedules place pressure on QM, which mayhinder the prompt identification of errors. This situation is denoted as latency,which correspondingly results in hidden errors (the same as undiscoveredrework in the Pugh-Roberts models).

This latency is determined by “Quality Management Thoroughness” (QMTH)(C in Figure 3), the degree to which the existing quality problems of an activ-ity have been identified during the QM process. Thus, a “Re-ExecutionOf-UncoveredErrorRate” (A in Figure 3) is governed by “UncoveredError-GenerationRatio” (D in Figure 3), which can be represented by reliability andQMTH. If the final design activity has 90 percent reliability and 80 percentQMTH, the actual re-execution of uncovered errors by the QM process is asmuch as 8 percent (= 0.1 × 0.8) of the total work scope of the final designactivity, and hidden errors are 2 percent (= 0.1 × 0.2).

On the other hand, in a case where hidden errors are identified at a laterstage, there is a possibility that completed work needs to be worked again (i.e.,reworked). In the model, E in Figure 3 represents this flow, and the situationswill be explained in a later section.

Along with this work execution structure, a parallel co-flow structureis used to model the errors. For example, in terms of uncovered errors, thestock of “UncoveredError” (UE, A in Figure 4) has an inflow, “Uncovered-ErrorDiscoveryRate” (B in Figure 4), which can be obtained from the averageerror rate in the work done (i.e., AvgErrorInWorkAwaitingQM and Total-OutflowFromWorkAwaitingQM,3 C and D in Figure 4, respectively) andQuality Management Thoroughness (E in Figure 4). The rest of the co-flowstructure, including the model about the hidden errors (H in Figure 4), isomitted in this paper, but it, including the complete model structure, can befound at http://web.mit.edu/lsh/Public/.

The model in Figure 3 can be extended to incorporate change (order)management. Change orders in construction are an essential mechanism forsatisfying owners’ construction needs throughout the project delivery processand responding effectively to errors in construction (Moselhi et al., 2005).



Fig. 4. Co-flow

structures for errors

However, at the same time, they cause detrimental impacts on constructionperformance. Suppose the owner requests the change in space usage fromnormal office to document storage in steel frame office construction. Usually,document storage requires greater structural capacity to bear the load. In thiscase, braces could be added to increase structural capacity. However, thesebraces could generate other unexpected issues. For example, adding bracescould require redesign of the original interior design, which could accompanyother changes such as furniture procurement, HVAC (heating, ventilating,and air conditioning) systems, and so forth. Also, schedule, cost, and theresource allocation plan could be consequently changed. Thus, modeling changemanagement is particularly important in understanding its impact on projectperformance.

As seen in Figure 5, change management can be modeled similarly to themodel of error management. For example, instead of reliability, change isassumed to be related to stability (A in Figure 5), which indicates the degreeto which the given work scope would be performed without a request ofa change. Like latency in errors, latency in changes can occur, at this time,in the Scope Management (SM) process, which aims to make sure that thegiven scope of work and the corresponding work settings for execution are the



Fig. 5. Work execution with change management [adapted from Motawa et al., 2007]

same, as specified in drawings and specifications. To represent latency, “ScopeManagement Thoroughness” (SMTH) (B in Figure 5) is defined as the degreeto which the potential changes have been identified during the SM process.In addition, the changes can be modeled using a parallel co-flow structure,similar to the model for the errors in Figure 4.

One of the important processes in managing changes is the Claim and ChangeManagement (CCM) process, the decision-making process that determines theadoption of changes. In the model, if changes are identified, related tasks in theWTD stock are sent to the stock of “WorkAwaitingCCMDecision” (WACCMGD),which needs the decision of, or analysis by, the CCM group.4 One of the majorfactors to affect a change decision is the feasibility of the raised issue. Forexample, a blasting method (i.e., dynamite) was planned for certain excavationwork. However, a closer investigation found that by utilizing the blastingmethod damage could occur to adjacent buildings, due to the existence of asofter ground than expected. Usage of the scrape-and-excavate method (i.e.,excavation with backhoe loaders) was requested as a change, and the request



was sent for review to the CCM process (C in Figure 5). If the CCM groupapproves this change request (D in Figure 5), the scrape-and-excavated methodwill replace the blasting method. However, the CCM group rejected this changeon the grounds of its unfeasibility. It was determined that it would take moretime to complete the job with the scrape-and-excavate method than with theblasting method, and the CCM group felt that this would generate a delay in theproject schedule that was not acceptable at that time (E in Figure 5).5 Thus,based on the CCM group’s decision, the request form change can be approvedor rejected mainly based on its feasibility or unfeasibility.

With this model, CASE 1 is simulated applying 85 percent of reliability andstability into the previous three activities,6,7 as seen in Table 1. Figure 6 showsthat introducing errors and changes generates additional work scope. If weexclude control policies such as overtime and hiring new workers, this addi-tional work scope will delay the completion time. Furthermore, if we assumeCCM Period (i.e., the time that the CCM process takes—F in Figure 5) as 7 days,the corresponding work can be delayed by as much as 7 days until the decisionis made. This could contribute to further delay in the execution of work.

LATENCY SETTLEMENT AND ADDITIONAL WORK SCOPE Errors and changes not iden-tified or completely discarded (e.g., hidden errors and latent changes) mayinvolve another process that helps coordination with other activities. An

Fig. 6. CASE 1: additional work scope caused by errors and changes



example of one of these processes, widely used in design and construction, isthe Request For Information (RFI) process, which coordinates design issuesbetween the design and the construction team. In this paper, the RFI process isextended and used to incorporate coordination with other preceding construc-tion activities, as well as with design activities. This is because the workclarification (e.g., design error or change) process often requires the involve-ment of the preceding construction activities. Such latency and its settlementthrough RFI would be particularly important in projects following concurrentdesign and construction,8 which has been widely adopted in the constructionindustry recently to reduce the project completion time. In concurrent designand construction, there is a higher possibility that errors and changes becomelatent than with a sequential development. This is because: (1) succeedingactivities often have to proceed without finalized information from precedingactivities so that many assumptions need to be made for succeeding activities(Tighe, 1991); (2) overloaded workers sometimes fail to respond to communica-tions, thereby compounding the information supply problem and compromis-ing others’ performance (Chachere et al., 2004); and (3) the decision-makingprocess may be accelerated due to a shortened project completion time (Leeet al., 2005). Thus, the understanding and modeling of latency and its settle-ment process is significant not only for error and change management in generalbut also the consequent dynamic problems it causes in concurrent design andconstruction.

In the model, if a hidden error or a latent change in the predecessor activityis found at the activity under study, it is passed through the RFI process, backto the predecessor activity. This is illustrated in Figure 7. The tasks identifiedas hidden errors and latent changes are accumulated in the stock of “Work-AwaitingRFIReply” (WARFIR) (A in Figure 7).

At this stage, three options are available for these pending tasks. The firstoption is that the activity under study requests the predecessor activity tocorrect the hidden error or to issue a latent change (B in Figure 7). This casewould arise because the origin of the hidden error and the latent change is thepredecessor activity and the activity under study asks the predecessor to dealwith them. The second option is that hidden errors can be sent back to WTD(C in Figure 7). A hidden error may be determined to be absorbed by the activ-ity under study. Thus, tasks would proceed to the activity under study, but noaction is taken to the predecessor activity. The third option deals with latentchanges, which directs the flow into the stock of WACCMGD (D in Figure 7).If the manager judges that a latent change needs urgent attention since it issignificant to the whole project performance, the manager may request adecision from the CCM group directly, rather than by way of the predecessoractivity. In the model, two variables are used to deal with these differentoptions: “RFIAdoption”9 and “CCMFromRFI”.10 While the former determineswhich hidden errors or latent changes need to be further considered byrequesting the correction to either the predecessor (B in Figure 7) or the CCM



Fig. 7. Work execution with the RFI process (adapted from Pugh-Roberts Associates, 1980, as referred to in Sterman, 2000;Ford and Sterman, 1998; Park and Peña-Mora, 2003)

group (D in Figure 7) after RFI, the latter shows the fraction of latent changeswhich need the CCM group’s attention (D in Figure 7).

Another important point in understanding the impact of errors and changesis that they can be amplified and propagated to related activities due toimposed, technical, and procedural relationships (Badiru and Pulat, 1995). Tounderstand and quantify this propagation impact within or among activities,sensitivity (Eppinger, 1997) is adopted in this framework as a work multiplier.The scope that results from errors and changes would be multiplied by sensi-tivity. In the model, an additional work amount caused by errors and changescan be represented by multiplying corresponding work scope with sensitivity(E in Figure 7).

CASE 2 simulates the application of sensitivity in order to capture a propa-gation impact keeping the previous CASE 1 simulation setting (e.g., 85 percentof realiability and sensitivity—Table 1) except sensitivity. In this experiment,sensitivity of 50 percent was applied, indicating that errors and changes couldgenerate as much as 50 percent of additional work scope to the other relatedactivity. For example, if an error generates 10 WU (Work Units) in the pre-ceding excavation activity, it can cause as much as 5 WU of additional work to



Fig. 8. CASE 2:

propagation impact

the succeeding superstructure activity. Figure 8 shows the total additionalwork scope that incorporates the impact of sensitivity compared with that inCASE 1. Sensitivity implies that we need to consider a propagation impact inorder to deal with unexpected additional work scope and, moreover, anti-cipate the capacity necessary to react to such additional work scope.

In addition, CASE 3 is simulated assuming that QM and SM thoroughnessare both 90 percent (i.e., 10 percent of errors and changes could be hidden orlatent) keeping the previous CASE 2 simulation setting (Table 1). The left sideof Figure 9 shows that work can be re-executed even though it is perceivedthat work is completed. In other words, the late discovery of hidden errors orthe late approval of latent changes could demand a request for the additionalwork which was already completed. The right-hand side of Figure 9 shows

Fig. 9. CASE 3: impact of latency



Fig. 10. CASE 3:

late discovery of

additional work scopedue to latency

that a significant amount of work is waiting for an RFI response, which triesto identify and implement hidden errors and latent changes. Together with theCCM period (e.g., 7 days in this simulation), RFI response time (i.e., the timeto be taken for RFI response—F in Figure 7) can also delay work executionbecause the RFI response time is assumed to be 5 days in this simulation.

Figure 10 shows each activity’s additional work scope generated from errorsand changes. One notable point is that when we apply latency it generates lessadditional work in the preceding final design activity (A in Figure 10) andmore additional work in the succeeding excavation activity (B in Figure 10).Due to latency, some portion of additional work scope may not be identifiedearly and may be identified at later stages. Thus, latency reinforces the idea of“90 percent syndrome”, the sudden work overflow at a later stage of a project(Ford and Sterman, 1998).

Full model and validation

Based on the aforesaid work execution structure with errors and changes,the full D2CPM model is developed. To represent the multiple activities ina project, the developed model structure (Figure 7), which represents eachactivity, was replicated using Subscript Control in Vensim™. In addition, thefull model includes other supporting model structures, such as the life cycle oferrors and changes (e.g., co-flow structure for errors in Figure 4), resources



(e.g., workforce allocation), and productivity (e.g., adequate schedule pressurecan increase productivity up to a certain threshold11). Furthermore, severalcourses of action that may be taken in response to productivity loss are alsomodeled, such as the adoption of overtime and hiring new workers.

This model has been validated in terms of its usefulness in identifying theimpacts of errors and changes on construction performance. Table 2 outlinestests that have been applied to the developed model.

As an example of diverse test techniques, the simulation result shouldproduce CPM calculations if the simulation setting follows CPM assumptions.As already seen in the CPM CASE, if we assign 100 percent reliability andstability (i.e., no error and change), a straight-line production rate (i.e., workexecution with a uniform rate), and no delay (i.e., instant time for the qualitymanagement process), the project duration generated by the model shouldbe identical to the one calculated by the CPM method. The execution of thesimulation confirms that these results are identical. In addition, the writersconducted a couple of case studies applying the developed model to real-world construction projects; one is a laboratory building project in Malaysiawith a contract sum of roughly U.S. $8 million and a 15-month span onconstruction (hereinafter, Malaysia project) and the other is a highway bridgeinfrastructure project in Massachusetts, itself one of 27 underpass and over-pass bridges that are part of a U.S. $400 million Design/Build/Operate/Transfer (DBOT) project with a 42-month span from design to constructioncompletion (hereinafter, Mass. project). Through these case projects, behaviorreproduction tests were conducted to determine whether the simulation effec-tively predicts behavior observed in a real-world project.

In order to determine the simulation input (i.e., model parameters),extensive individual interviews, surveys, and focus groups are conducted.Through these channels, construction managers, superintendents, schedulers,and workers are asked to estimate values, such as forecast stability, forecastedreliability,12 and sensitivity, based on their experience and historical data.At the initial data-gathering stage, it is not easy to convey to participantsthe precise meaning of parameters and how they can be measured becausethe participants have never used such parameters in their project planning.In addition, the use of unfamilar parameters may cause subjective assessment.To overcome this situation, we divided the input elicitation process into thethree steps shown in Figure 11. First, we conducted a semi-structured inter-view to collect the preliminary data in an attempt to understand project andactivity characteristics (A in Figure 11). Not only does the semi-structuredformat generate consistent answers to a predetermined questionnaire, but italso has the flexibility to incorporate interviewees’ opinions that may be unan-ticipated by our prepared questionnaire. Figure 12 shows an example of thepredetermined questionnaire used in one of the case projects. Here, thequestionnaire was specifically intended to address different activity charac-teristics. In addition, for some parameters’ behaviors, interviewees were asked



Table 2. Applied test techniques [adapted from Lee et al., 2005]

Test

Boundary adequacy

Structure assessment

Dimensionalconsistency

Parameter assessment

Extreme conditions

Integration error

Behavior reproduction

Sensitivity analysis

Tools and procedures used at model testing

Data collected from project documents and records,interviews from industrial partners (InteCap Inc.,Modern Continental, and Barletta Heavy Division)and literature reviews are used to confirm boundaryadequacy—to see whether there is a potentially importantfeedback omitted from the model

Subsystem diagrams, causal loop diagrams, stockand flow maps, and direct inspection of modelequations—to test the boundary used in the model

Subsystem diagrams and stock and flow maps—toreveal the level of aggregation

Casual diagrams—to confirm the information cues usedin each decision

Direct inspection of the equations—to check theheuristic assumed at each decision point

The developed model is evolved from severalproven model structures (Cooper, 1980; Richardsonand Pugh, 1981; Abdel-Hamid, 1984; Ford and Sterman,1998; Rodrigues and Williams, 1998; Lyneis et al., 2001;Park and Peña-Mora, 2003)

Automated dimensional analysis function,provided by Vensim™ (system dynamics simulationpackage used in this research)—to checkdimensional errors

Direct inspection of the equations—to ensure thatevery equation must be dimensionally consistentwith real-world scenarios

Formal statistical estimation—to get available numericaldata (e.g., ‘RFI response time’ is set to 7 days as a defaultduration estimated from the statistical mean of the datafrom different design and construction projects)

Judgmental estimation—to supplement unavailablenumerical data (e.g., the degree of project completionwill be used to update the different values based on thedata collected on the project. It can be set to take the meanas quartile of the actual data, such as 25%, 50%, 75%,and 100%) (e.g., “RFI response time” can be modified byusers since project type or communication technologymay change its duration)

Putting the extreme value—to test the robust behaviorof the model (e.g., if reliability and stability are 100%,i.e., no error and change, the simulated duration is thesame as CPM estimation because CPM does not considerthe error and change generation during actual execution)

Try different time steps and run the model again—to showthat there is no major difference among them (e.g., no majordifference in the model: 0.1. 0.25, 0.5)

Compare the variables’ behavior with the known—to seehow well it reproduces historical behavior (e.g., comparisonwith the actual progress curve from the case projects)

Numerical, behavior mode, and policy sensitivity analysis—to know if a change in assumptions changes thenumerical values of the results, patterns of behavior, andimpacts of a proposed policy, respectively (e.g., differentprogress curves by different options)

Purpose of test

To assess the appropriatenessof the model boundary in termsof the model purpose

To determine whether the modelis consistent with knowledgeof the real system relevant tothe model purpose

To ensure that all variables aremathematically consistentand further, to have variablesthat have realistic meanings

To estimate the values of eachparameter in terms of itsreasonableness

To determine whether themodel behaves in a realisticfashion no matter how extremethe inputs or policies imposedon the system are

To make sure that the model isnot sensitive to the choices oftime step or integration method

To assess a model’s ability toreproduce the behavior of interestin the system

To test the robustness of the modelbehavior in uncertain conditions



Fig. 11. Input elicitation process

to draw the shape of behavior over time if necessary. All answers were thentranscribed for preliminary data analysis. Preliminary data analysis, denotedby B in Figure 11, was then done using statistical methods. For example, theaverage RFI response time for this case project and its approximate probabilitydistribution were obtained. Finally, with this preliminary data analysis andthe input guide explaining how to choose numerical input values, agreementwith interviewees was reached in order to determine reasonable numericalinput values (C in Figure 11).

Fig. 12. Exampleof the predetermined

questionnaire for the

input elicitationprocess



This input elicitation process has another benefit. It helps participants betterunderstand their activities and projects. For example, one of the intervieweesin the Malaysia project confirms:

I implicitly know that more additional work due to errors and changes will beintroduced. However, there are no tools to address and, further, incorporate thisidea in planning. Actually, answering the questionnaire helps me better under-stand how I need to prepare a resource profile such as workforce allocation incase that unexpectedly more work is introduced. (Project Manager A in the Malaysiaproject)

The explicit use of such dynamic activity characteristics, which are in theirmental database, provides a different perspective which enables participantsto better understand their activities and projects.

Figure 13 shows the resultant simulation from the Malaysia project. Specifi-cally, Figure 13 illustrates both the initially planned Percentage of Work Com-plete (PWC—A in Figure 13) and the actual PWC (B in Figure 13), which hadbeen gathered weekly, along with the simulated PWC (C in Figure 13). Thesimulated PWC curve is obtained through two-time simulation: one conductedprior to 13 March and the other done after 13 March because a project policy

Fig. 13. Model testing

example: projectprogress of the

Malaysia project

[adapted from Lee andPeña-Mora, 2005]



was changed at that time (i.e., to accelerate behind schedule, the new policy,hiring new workers, was added on 13 March). The actual PWC, as of 17 July2005, is far behind the planned PWC due to delay caused by errors andchanges. However, the simulated PWC from our model closely follows theactual PWC, with 3.38 percent (until 13 March) and 3.42 percent (until 17 July)root mean square error.

Analysis

Based on diverse simulation experiments and several applications to real-world case projects, this section discusses how we can understand the behaviorgenerated from errors and changes and, further, how we can manage them.

Importance of considering errors and changes

Taking into account errors and changes has a significant meaning in managinga project because additional work scope generated by errors and changes coulddisrupt the entire prepared plan. This also implies that realistic scheduleplanning that considers the possible errors and changes is needed, and bydoing this a consequent realistic contingency plan can be achieved.

For instance, the project manager and the scheduler in the Malaysia projectdid not have a systematic mechanism or sufficient knowledge base to judgehow the project was going to progress. Thus, they simply speculated thePercentage of Work Complete based on constant and perfect production with-out any consideration of the possibility of generating errors and changes (i.e.,reliability and stability), their propagation impacts caused by interdepend-ency (i.e., sensitivity), and their possible latency (i.e., hidden errors and latentchanges). As for safety, the manager or scheduler merely put in place a largefraction of contingency buffers based on experience (e.g., contingency bufferfor 20–40 percent activity duration). These buffers created the expectationof quite an unrealistic plan (e.g., unnecessarily long duration was assignedto activities without confidence). In fact, the plan was so unrealistic that theyhad to maintain two schedules: one for schedule management purposes, andthe other for reporting purposes. The latter schedule contained a large fractionof buffers in order to avoid potential liability. Clearly, much time and effortwas wasted because of the lack of a mechanism or sufficient knowledge tounderstand such dynamic behaviors in the construcion process associatedwith errors and changes.

On the other hand, this unrealistic plan causes an unrealistic contingencyplan, which in turn could disrupt the whole construction. For example, addi-tional work scope would require a different workforce utilization ratio fromthe one in the original plan (e.g., CPM CASE), as seen in Figure 14.13 Thisimplies that there is a need for realistic workforce loading planning that takes



Fig. 14. Disrupted

workforce utilization

ratio

care of the possible increase of work scope during actual execution. Broadlyspeaking, a contingency plan against additional work scope is needed to avoidsevere problems such as workforce shortage on a project.

Continuing with the Malaysia project, its contingency plan against theincrease of scope was the adoption of overtime. However, due to a lack ofability to estimate the actual need for overtime in advance, the scope issue wasnot handled well. Thus, the management team decided to hire new workersfrom an adjacent foreign country—a decision not in their contingency plan.Although the hiring process proved time consuming, when the new workersfinally did start, the results were positive owing to the increase in the workdone. However, some time later the newly hired workers called a strike whenthey realized that they had received considerably less payment than thedomestic workers. The strike had the potential to cause severe schedule andcost overrun. Fortunately, the strike was resolved early, but it caused signifi-cant costs. This case example shows the importance of the early considerationof additional work scope that can occur during actual execution. This ulti-mately contributes to the development of a realistic contingency plan based onan accurate estimate of project performance.

Need for efficient coordination

From the previous simulation experiments (Figure 9), it can be surmised thatthe time taken for the RFI and CCM processes could delay work execution.This implies that implementing an efficient coordination process can reducethe detrimental impact of errors and changes. If a project cannot avoid the



generation of errors and changes, the second-best policy would be to improvethe coordination process that deals with errors and changes. The Malaysiaproject and the Mass. project indicate that the RFI response time and the CCMperiod could take, on average, over 20 days. In such cases, the impact could bemore detrimental than simply delaying the project completion. Consideringthe involvement of many multiple temporary organizations (e.g., subcontrac-tors) in a construction project, implementing and maintaining an efficientcoordination process also contributes to streamlining information flow.Particularly, since the existence of errors and changes in uncertain and com-plex projects is inevitable, providing an efficient mechanism to coordinateerrors and changes would be an appropriate means to rectify them.

For example, Figure 15 shows “WorkAwaitingRFIReply” and “Work-AwaitingCCMDecision” simulated with 14 days RFI response time and CCMperiod and 7 days,14 respectively. Simulation results clearly inform the re-duction of tasks waiting for RFI reply and CCM decision when a shorter RFIresponse time and CCM period are applied. Eventually, a reduction of projectcompletion time could be achieved, streamlining information flow.

On the other hand, the Mass. project involved a design–build contract(a single contract between owner and design–builder and thus, one in whichdesign and construction are performed by one team). The coordination processbetween design and construction was supposed to work well in this contractbecause a design team could work continuously to correct their design errorsand to implement changes without causing any additional cost, contract,or liability issues. However, it turned out that this project was the first design–build contract in which these two groups worked together. Due to the lack ofexperience working with a design–build contract, there was a significant delayin coordinating the settlement of errors and changes through the RFI and CCMprocesses. This example highlights the importance of accounting for the

Fig. 15. Impact of shorter RFI response time and CCM period



increase in the level of coordination among project functions (i.e., reducingRFI response time and CCM period) in managing errors and changes, eventhough the contractual solution (e.g., a design–build contract) might in theoryprovide for easier coordination.

Need for a proactive contingency plan

As seen in Figure 10, latency can explain the sudden work overflow at a laterstage of a project. This detrimental impact of latency implies that currentcontingency plans need to be approached differently. In other words, mostcontingency plans in construction are based on a reactive approach: acting inresponse to something. Contingency plans can help absorb the detrimentalimpact of latency by accelerating the progress, but they do not provide for anyprevention of a possible disruption; rather, they just give more resourcesto recover from a disruption. One possible way to overcome such situations isthe adoption of a proactive mechanism that aims to look ahead at possibleproblems and issues. Such proactive measures could be implemented in theform of a collaborative meeting with the design group and involve subcontrac-tors before an activity starts. The meeting would be designed to share anddiscuss potential problems, such as clarifying ill-defined tasks (e.g., designs),ensuring the accuracy and constructability of drawings, and securing resourceprocurement. Through collaborative efforts directed at reducing potential prob-lems in advance, opportunities for identifying hidden errors and latent changeswould be increased, ultimately reducing the number of time-consuming RFIs.When the authors applied this collaborative meeting into the simulation of theMalaysia project and the Mass. project, the results were about 5–11 percentsavings in project duration. For example, Figure 16 shows additional workscope caused by errors and changes in a final design activity and a shop

Fig. 16. Early identification of errors and changes by proactive collaboration



drawing submittal activity, which has a Finish-to-Start relationship in theMass. project. When the collaborative meeting was applied, additional workscope was introduced (A in Figure 16). However, this enabled the early incor-poration of errors and changes, preventing them from reappearing at a laterstage of the project.15 In the end, it contributed to the significant reduction ofactivity duration (B in Figure 16).

Network-based tools vs. SD

Network-based tools, such as CPM, PERT, and PDM, have been widely usedin the construction industry, particularly because of their ease of applicability.However, they assume that the attributes of project activities such as durationare known at the beginning of a project and do not change during projectexecution. Thus, they are not adequate for representing the actual projectprogress, which results in frequent manual updates to the schedule to reflectthe actual performance in the schedule (Martinez and Ioannou, 1997).

In contrast, SD’s focus on the system structure to understand dynamicbehavior can greatly contribute to managing a project. Taking into accountfrequent errors and changes, and the resultant deviation between expected andactual performance, understanding the system structure can provide a greatopporutunity to make an effective policy to manage them. This is clearlydifferent from network-based tools’ rigid and static sturcture. Further, theintroduction of the system structure (e.g., how errors and changes affectconstruction performance) to the project manager helped them to understandtheir project in-depth and to incorporate this understanding into managingtheir project.

Nonetheless, the applicability of SD to real-world projects is still in questionbecause of the lack of detailed operational explanation (Rodrigues and Bowers,1996). To overcome this situation, this research integrated traditionalnetwork-based tools and an SD-based project model. Not only does the SDmodel incorporate concepts in network-based tools, but also the output ofthe SD model is translated to the network-based tools. For example, Figure 17is a screenshot of the web-based system showing the impact of errors andchanges on the schedule network16 (A in Figure 17) as well as dynamic beha-viors generated by the SD model (B in Figure 17). As Rodrigues and Bowers(1996) pointed out, incorporating quantitative data from SD simulation intothe network-based tools provided industry practitioners with more familiaritywith SD simulation.

Conclusions

This paper discussed the detrimental impact of errors and changes on con-struction performance, focusing on their resultant additional work scope.



Fig. 17. Screenshot of web-based system [adapted from Lee et al., 2005]

To understand this, the system dynamics (SD) model is developed, integratingseveral concepts in traditional network-based tools. Based on diverse simula-tion experiments and a couple of real-world case projects, the writers con-cluded that: (1) realism should be added to schedule planning, taking intoaccount the impact of errors and changes; (2) an efficient coordination processis needed to manage errors and changes; (3) proactive contingency plans needto be taken into consideration, particularly against latency; (4) integration ofnetwork-based tools and a System Dynamics-based model can contribute tomanagement of errors and changes.

Although the current model has established the potential for overall errorand change management, further research efforts are necessary to improveits detailed representation of different construction situations. For example,comprehensive change management can be achieved by addressing the differ-ent levels of changes that can occur in construction. These include the changesapproved by managers at an organizational level and/or by an owner. Allchanges made by different players in a project can have different characteris-tics and impacts on design and construction performance. In addition, strength-ening operational details could enhance the model’s ability to represent changemanagement. For instance, stability is used as a general concept to filter out



possible changes in this research. This idea could be further specified (e.g.,material stability, technology stability, and workforce stability) and formu-lated in detail to obtain the operational aspects of change management. In thisway, the model could improve its practical simulation capability and providemore comprehensive information.

Notes

1. If Activity A and B have a Start-to-Start relationship with a lag of 10 days,this means that Activity B can start 10 days later after Activity A starts.

2. These three activities will be continuously used as a base for the otherupcoming simulation runs, which will add more variables about thedynamic characteristics of errors and changes, as seen in Table 1. Thus,how adding these variables will change simulation results will be ob-served and discussed in the following sections.

3. It is the sum of Re-ExecutionOfUncoveredErrorRate and WorkCompleted-Rate in Figure 3.

4. Some potential changes may not be identified, similar to hidden errors inerror management. This is called latent changes.

5. Otherwise, it will be approved and back to the WTD stock to be performed,as denoted by E in Figure 4.

6. In this simulation run, a change approval rate, which represents the degreeto which the identified changes would be approved, is assumed as 50percent for simplicity. Thus, the 50 percent of the identified change wouldbe approved while the rest would be rejected.

7. CASE 1 assumes 100 percent QM and SM thoroughness, thereby generat-ing no hidden errors and latent changes.

8. It includes not only overlapping design and construction, which is calleda fast-tracked project, but also overlapping activities during design andconstruction.

9. In this paper, they are assumed as 50 percent through all simulation runsfor simplicity.

10. In this paper, they are assumed as 50 percent through all simulation runsfor simplicity.

11. Yerkes–Dodson law (1908), as referred to in Sterman (2000).12. After inputting forecast reliability, actual reliability will be generated

through model simulation. This is because many factors can affect actualreliability during execution. For instance, learning would increase initialreliability because workers get used to their work as time goes by.

13. This simulation is continued from the previous hypothetical experiments.14. This simulation is also continued from the previous hypothetical experi-

ments. Except for the RFI response time and CCM period, the rest of thesimulation setting is the same as CASE 3 in Table 1.



15. The succeeding construction activities enjoyed less additional work scopedue to this early identification.

16. SD’s analysis of the impact of errors and changes was translated into theschedule network.

Acknowledgements

The authors would like to acknowledge contributions to this paper by Joe Peck, formerCorporate Planning and Scheduling Manager, currently with Charles River Associates;Bill Lemoine, Vice President; and John Foster, Senior Project Manager at the ModernContinental Company. We also would like to thank Philip Helmes, Vice President; andMargaret Fulenwider, Senior Consultant formerly at InteCap Inc., currently at CharlesRiver Associates. Also, we appreciate the assistance from Dr Indra Gunawan, AssociateProfessor, and Mr Tai Soon Chee, Graduate Student at the Malaysia University ofScience and Technology. We also would like to thank Dr Moonseo Park, AssistantProfessor at Seoul National University, and Dr Mikio Shoji, Senior Managing Director atKajima Corporation. The writers would also like to acknowledge the financial supportfor this research received from the National Science Foundation, award CMS-0324501.Any opinions, findings, and conclusions or recommendations expressed in this publi-cation are those of the writers and do not necessarily reflect the views of the NationalScience Foundation.

References

Abdel-Hamid T. 1984. The dynamics of software development project management: anintegrative system dynamics perspective. PhD thesis, Sloan School of Management,MIT, Cambridge, MA.

Badiru A, Pulat S. 1995. Comprehensive Project Management: Integrating OptimizationModels, Management Principles, and Computers. Prentice-Hall: Upper Saddle River,NJ; 107–115.

Chachere J, Kunz J, Levitt R. 2004. Observation, theory, and simulation of inte-grated concurrent engineering: risk analysis using formal models of radical projectacceleration. CIFE Working Paper #WP088, August, 2004, Stanford University,Palo Alto, CA.

Cooper K. 1980. Naval ship production: a claim settled and a framework built. Inter-faces 10(6): 20–36.

Cox ID, Morris JP, Rogerson JH, Jared GE. 1999. A quantitative study of post contractaward design changes in construction. Construction Management and Economics17(4): 427–439.

Eppinger S. 1997. Three Concurrent Engineering Problems in Product DevelopmentSeminar, Sloan School of Management, MIT, Cambridge, MA.

Ford D, Sterman J. 1998. Dynamic modeling of product development processes. SystemDynamics Review 14(1): 31–68.

Hegazy T. 1999. Optimization of resource allocation and leveling using geneticalgorithms. Journal of Construction Engineering and Management 125(3): 167–175.



Ibbs W, Lee S, Li M. 1998. Fast-tracking’s impact on project change. Project Manage-ment Journal 29(4): 35–42.

Josephson P, Hammarlund Y. 1999. The causes and costs of defects in construction:a study of seven building projects. Automation in Construction 8(6): 681–687.

Lee S, Peña-Mora F. 2005. System Dynamics Approach for Error and Change Manage-ment in Concurrent Design and Construction. 2005 Winter Simulation Conference,IEEE, Piscataway, NJ.

Lee S, Peña-Mora F, Park M. 2003. Reliability and stability buffering approach inconcurrent design and construction projects. In International Group of Lean Con-struction 11th Annual Conference, Blacksburg, VA, 22–24 July 2003.

——, ——, ——. 2005. Quality and change management model for large scale con-current design and construction projects. Journal of Construction Engineering andManagement, ASCE, 131(8): 890–902.

Lyneis J, Cooper K, Els S. 2001. Strategic management of complex projects: a case studyusing system dynamics. System Dynamics Review 17(3): 237–260.

Martinez JC, Ioannou PG. 1997. State-based Probabilistic Scheduling Using Strobo-scope’s CPM Add-On. Proceedings of Construction Congress V, Minneapolis, MN,Oct. 4–8, 1997, pp. 438–445.

Moselhi O, Assem I, El-Rayes K. 2005. Change orders impact on labor productivity.Journal of Construction Engineering and Management, ASCE 131(3): 354–359.

Motawa I, Anumba C, Lee S, Peña-Mora F. 2007. An integrated system for changemanagement in construction. Automation in Construction, Elsevier: Oxford, UK.Vol. 16, pp. 368–377.

Park M, Peña-Mora F. 2003. Dynamic change management for construction: introduc-ing change cycle into model-based project management. System Dynamics Review19(3): 213–242.

Richardson G, Pugh AL III. 1981. Introduction to System Dynamics Modeling withDynamo. MIT Press: Cambridge, MA (available from Pegasus Communications:Waltham, MA).

Rodrigues A, Bowers J. 1996. System dynamics in project management: a comparativeanalysis with traditional methods. System Dynamics Review 12(2): 121–139.

Rodrigues A, Williams T. 1998. System dynamics in project management: assessingthe impacts of client behavior on project performance. Journal of the OperationalResearch Society 49(1): 2–15.

Senior B, Halpin D. 1998. Simplified simulation system for construction projects.Journal of Construction Engineering and Management, ASCE 124(1): 72–81.

Sterman J. 2000. Business Dynamics: System Thinking and Modeling for a ComplexWorld. McGraw-Hill: New York; 587–595.

Tighe J. 1991. Benefits of fast tracking are a myth. International Journal of ProjectManagement 9(1): 49–51.

understanding and managing iterative error and change cycles in construction

Documents