Quantifying the impacts of rework, schedule pressure, and ripple effect loops on project schedule performance

Authors


Introduction

Schedule performance is often critical to construction project success. But many times projects experience large unforeseen delays and fail to meet their schedule targets. The failure of large construction projects has enormous economic consequences. A specific example is the Central Artery Project. This megaproject was developed to relocate a primary Boston, MA highway from an above-ground highway to under downtown Boston. When the construction was completed in 2007, nine years later than initially scheduled, the total cost was about $14.8 billion, over five times the original estimate (LeBlanc, 2007). Project schedule failure was a significant driver of those excess costs.

The challenges of managing construction projects regarding their schedule performance stem partially from their dynamic nature and internal feedback processes. System dynamics has a long history of investigating the role of feedback in project schedule performance. System dynamicists have incorporated feedback loops in project models 1 and recognized their important role (e.g. Repenning, 2001; Taylor and Ford, 2006; Godlewski et al., 2012; Fiddaman, 2013). In contrast, the persistence of large project delays implies that their importance has not been fully recognized and incorporated into practice. Traditional project management methods do not explicitly consider the effects of feedback (Pena-Mora and Park, 2001). Project managers may intuitively include some impacts of feedback loops when managing projects (e.g. including buffers when estimating activity durations), but the accuracy of the estimates is very dependent upon the experience and judgment of the scheduler (Sterman, 1992). Owing to the lack of a widely used systematic approach to incorporating the impacts of feedback loops in project management, the interdependencies and dynamics of projects are often ignored. This may be due to a failure of practicing project managers to understand the role and significance of commonly experienced feedback structures in determining project schedule performance. Practitioners may not be aware of the sizes of delays caused by feedback loops in projects, or even the scale of impacts. 2 The system dynamics community has not adequately communicated the potential of reinforcing feedback loops, in particular in construction projects, to create widespread changes in managing project schedules. One reason for the communication failure may be the lack of quantified descriptions of the size of those impacts. System dynamicists have used reinforcing loop structures to predict the impact of contractual changes on the time and cost performance of specific projects (e.g. Cooper, 1980; Godlewski et al., 2012), and Fiddaman (2013) described the non-uniform distribution of the impacts of feedback on project durations. Recently, Parvan et al. (2015) used data from 30 construction projects to estimate the strength of feedback between design and construction phases on project cost and duration. Their study focused on the interaction between two different phases in the project, with an emphasis on the rework cycle. In the current work, a simple validated project model has been used to quantify the schedule impacts of three common reinforcing feedback loops (rework cycle, “haste makes waste”, and ripple effects) in a single phase of a project. Quantifying the sizes of different reinforcing loop impacts on project durations in a simple but realistic project model can be used to clearly show and explain the magnitude of these impacts to project management practitioners and students, and thereby the importance of using system dynamics in project management.

Project modeling involves a tradeoff between model simplicity that supports generalization and model specificity that increases accuracy. Many system dynamics models exist for specific projects. The current work purposefully uses a relatively simple model (seven stocks) developed by Taylor and Ford (2006, 2008), to reveal the relationships between three common project feedback loops and project schedule performance, and to facilitate the generalization of the results. The system dynamics project model was modified, calibrated and validated again to simulate the impacts of reinforcing feedback loops on project duration in a generic but realistic model. A 2k factorial design (for details see Law, 2015, ch. 12; Montgomery, 2014, ch. 14; and Appendix 1) was used to separate and compare the main effects of the three reinforcing loops and their interaction effects on project duration. The results provide valuable information for explaining and expanding the role of system dynamics in projects.

The project model

The Limerick Unit 2 nuclear power plant construction model (Taylor and Ford, 2006, 2008) was used for the study. The model simulates the flow of work through four backlogs of a construction project (Figure 1) and the dynamic allocation of resources to the three development activities (initial completion, quality assurance, and rework). 3 See Ford and Sterman (1998) and Taylor and Ford (2006) for detailed descriptions of the model structure.

Figure 1.

The conceptual project model (based on Taylor and Ford, 2006, 2008)

The model includes one relevant balancing loop and three reinforcing loops. In the most direct path through the project, work moves from the Initial Completion Backlog, into the Quality Assurance Backlog. The balancing loop (Figure 1, B1) releases work as the quality assurance activity approves work packages and they become Work Released.

The model simulates three of the most common barriers to meeting project schedule goals: performing rework to repair errors, accelerating project progress to finish the project by its deadline, and completing unanticipated work created by errors. The first reinforcing loop is the rework cycle (Figure 1, loop R1). It withdraws work from the direct path by discovering work that requires iteration and returns that work to the direct path at the Quality Assurance Backlog. This reinforcing loop increases project durations by requiring some development activities to be performed multiple times on the same work packages. The “rework fraction” defines the strength of the rework loop. This variable reflects the complexity of a project. If a project is relatively simple and the required tools and methods are well understood (e.g. building a house), there may be little or no rework created and the “rework fraction” would be near zero. In contrast, complex projects (e.g. a nuclear power plant) will have large rework fractions. The maximum value of the rework fraction is one, reflecting conditions in which no work passes inspection and all work requires rework.

The second reinforcing loop (Figure 1, R2) models “haste makes waste” impacts. When the time required to complete a project exceeds the time available, the pressure to complete the project increases. Managers can respond in several ways (see Ford et al., 2007, for details), but most responses result in more errors and the discovery of more errors in the rework cycle (R1). This reinforcing loop can increase project durations by increasing the amount of work that requires rework. The variable “sensitivity to schedule pressure” reflects how sensitive the project is to stress due to schedule pressure. As an example, if the project contract includes large penalties for missing the deadline, the project will be more sensitive to delays and workers will be under more pressure to get the project back on schedule. Higher values of “sensitivity to schedule pressure” reflect projects that perform poorly when faced with time pressure. As a result, more mistakes are made and a higher fraction of work is found to require change in such projects.

The third reinforcing loop (Figure 1, R3) models the ripple effects of discovering required rework. This new, unplanned work must pass through the direct path and possibly through the rework cycle for final approval. This reinforcing loop delays project completion by adding scope to the project. The variable “ripple effects strength” is used to describe the interdependency of project tasks and strength of this loop. When there is no interdependency between tasks, a mistake in one task will not produce additional work. In this case, “ripple effects strength” is zero. However, when tasks are interdependent, a change in one task will generate a series of changes in dependent tasks. For example, if during casting of a girder in a reinforced concrete frame it is discovered that the wrong size of rebar has been used, the adjacent beams must be supported properly while the rebar is removed and replaced. These additional supports are tasks that were not anticipated or included in the original plan and will increase the project scope.

Taylor and Ford (2006, 2008) used the Limerick Unit 2 model to investigate the impacts of these loops on project schedule performance. The combination of these balancing and reinforcing loops creates a tipping point that drives project behavior and schedule performance. In a healthy project that remains under tipping point conditions, the percent of completed work increases in an S-shape behavior until all work packages have been completed at the end of the project. However, the tipping point creates a possible failure mode where, due to the addition of work, the rate of work entering the project backlog (Initial Completion Backlog + Quality Assurance Backlog + Rework Backlog) exceeds the rate of work being approved and released. At this point, the project crosses the tipping point, causing the project backlog to grow and the project to not finish. Taylor and Ford (2006) illustrate how the shifting dominance of the loops described above can help explain project schedule failure (Figure 2).

Figure 2.

Project trajectory of a tipping point-induced project failure (adapted from Taylor and Ford, 2006)

Some minor changes were made to the Limerick model for the study. The equations that represent the special policies used in the Limerick project were removed. An example is that the total staff size was changed from being dynamic in ways that reflected Limerick-specific decisions into a constant. Exogenous variable values (e.g. “scope initial”) were kept equal to or close to those in the Limerick model. The changed model 4 was subjected to standard system dynamics validation tests (e.g. unit consistency, reasonable behavior patterns) and found to be useful for its current purpose.

Research design

The impacts of the reinforcing loops on project duration are of interest to project planners and managers, as these loops can create enormous project delays. To quantify the impacts of individual and sets of reinforcing loops on project delay, a 2k full factorial design was used. In a factorial design, two levels are chosen for each of the k variables. Simulations are run for all combinations of variable levels (2k combinations). The main effect of each variable is calculated by taking the average response over all combinations of other variables’ levels. A factorial design is more efficient than the common approach of changing one variable at a time in regard to the number of simulations (Law, 2015). In addition, it is the only way to measure the interactions between the variables (Montgomery, 2014). For a more comprehensive description of the design of experiments (DOE) using a 2k full factorial design see Law (2015, ch. 12) or Montgomery (2014, ch. 14). System dynamicists have previously used DOE to find the effect of different variables in their models. Kleijnen (1995) used DOE in a case study to investigate the effects of individual variables and three different policies. Wakeland et al. (2004) combined DOE with broad-range sensitivity analyses to investigate a hybrid system dynamics–discrete event model that was based on a classic system dynamics software project model. Recently, Chuang and Oliva (2015) used DOE to find the primary drivers of inventory record inaccuracy in their store inventory management model.

The results of previous research were used to select the variables used in the experimental design to control the strength of the feedback loops. In the Limerick case, Taylor and Ford (2006) used sensitivity analysis to identify the variables with the most impact on project duration: the “rework fraction”, “sensitivity to schedule pressure”, and “ripple effects strength”. These variables control the strength of each reinforcing loop studied here. The low level of these loop-strength variables was set at 0. These values reflect conditions in which the reinforcing loops have little or no impact on project duration. The high-level values of the loop-strength variables should be large enough to reveal the impact. However, when the values of the variables are set to greater than 0.35, the project crosses a tipping point. In this case, the project enters a failure mode in which the backlog of the project increases over time and the project does not complete in a practical time frame. 5 Because the dynamics of project failure are not the focus of this study, the high-level values of all three variables was set at 0.35. This excludes tipping point-induced project failure behaviors from the results. See Taylor and Ford (2006, 2008) for analyses of the model used here under conditions that cause tipping point-induced project failure. A full factorial design with three variables and two levels has 23 = 8 combinations where the variables are systematically set either at their low or high level in an orthogonal array. The design matrix and the variable level values are shown in Table 1.

Table 1. Design matrix for the 23 factorial design
Factor combinationRework fractionSensitivity to schedule pressureRipple effects strengthProject duration (D)
1000D1
20.3500D2
300.350D3
40.350.350D4
5000.35D5
60.3500.35D6
700.350.35D7
80.350.350.35D8

The use of the design of experiments requires that the model also reflect some of the underlying uncertainty in the system. Variables that span across the three development activities in the project (i.e. initial completion, quality assurance, and rework) and reflect uncertainty in both resources and processes can best describe the uncertainty in a wide range of projects. Therefore, six exogenous variables that meet these criteria and also affect project duration were chosen to model project uncertainty. These variables are the productivity of the workforce and the minimum process time required to finish a work package in each of the three development activities. These variables were randomly chosen from a uniform distribution within 20 percent of their base case values. 6 A thousand simulations were run for each factor combination. The design of experiment provides an objective mathematical procedure to compare the mean project durations of different factor combinations and thereby quantify the main and interaction impacts of reinforcing loops. The details of the steps and calculations are described in Appendix 1.

Results

The main impacts of individual reinforcing feedback loops (i.e. when applied alone) and their interactions on project duration were quantified and disaggregated by using design of experiments. Table 2 shows the contribution of main and interaction impacts of reinforcing loops to total project delay (see Appendix 1 for details and calculations). Analysis of variance (ANOVA) and Student's t-test show that all main impacts and the interaction effects are statistically significant (see Appendix 2).

Table 2. Impacts of reinforcing feedback loops on project duration
#Reinforcing loop(s)Average delay (months)Average delay (%)Relative contribution (%)
I1Rework cycle main impact8010722.4
I2“Haste makes waste” main impact537114.9
I3Ripple effects main impact466213.0
I12Rework cycle/“haste makes waste” interaction impact527014.7
I13Rework cycle/ripple effects interaction impact466212.9
I23“Haste makes waste”/ripple effects interaction impact395311.1
I123Rework cycle/“haste makes waste”/and ripple effects interaction impact395311.0
 Total with all three loops active355477100.0

The results show that the three common project reinforcing loops can cause major project delays. In this study, changing only the “rework fraction” from 0 to 0.35 will, on average, add 80 months to the project duration (107 percent of the 75-month planned project duration). For the case in which both the rework cycle and “haste makes waste” loops are present simultaneously, the total delay is the sum of the main impact of the rework cycle (I1 = 80), the main impact of “haste makes waste” (I2 = 53), and the interaction impact between these two loops (I12 = 52): a total of 185 months (248 percent of the planned duration). The table can be used to find the delays caused by other combinations of reinforcing loops. For the ultimate case when all three reinforcing loops are simultaneously active, project delay is the sum of all main and interaction impacts (I1 + I2 + I3 + I12 + I13 + I23 + I123). In this case, the average project delays increase to 355 months (477 percent of the planned duration), which is near the delay experienced by Boston's Central Artery project.

Specific results depend on experimental variable values, specifically the high values used in the design of experiments. However, evaluation of the relative contribution of each reinforcing loop on the total delay reveals that the rework cycle always makes the largest contribution. This is not a feature of variable values. Figure 3 shows the main impacts of these reinforcing loops as the high levels of the variables are changed from 0.05 to 0.35 according to the 23 factorial design matrix, while the low levels are kept at 0. The graph shows that, regardless of the value of the variables, the rework cycle has the most impact on project duration, ranging from 1.2 to 26.5 times more than the next most influential loop. As the high level of the variables increases, the impact of “haste makes waste” and “ripple effects” loops increases. Statistical screening (Ford and Flynn, 2005; Taylor et al., 2010) shows that the rework loop remains dominant throughout the project duration, but the influence of the “haste makes waste” loop generally decreases relative to the influence of ripple effects loop as projects progress (see Appendix 3).

Figure 3.

Relative contribution to total project delay for different high-level values

Importantly, the project completion delays reflected in the results are created solely by the internal feedback structure of projects. In other words, ceteris paribus, the presence of a single internal reinforcing feedback loop can have devastating effects on project schedule performance. Combinations of reinforcing feedback loops can degrade project schedule performance even further. These results support previous studies of projects but also rigorously quantify the sizes of the delays that these common internal reinforcing loops alone can cause in projects.

Discussion and conclusions

Three distinct and common project features create reinforcing loops that can cause project delays: project complexity that generates errors (rework cycle), “haste makes waste” behaviors (Lyneis and Ford, 2007) that increase errors (schedule pressure loop), and the addition of unplanned work due to rework (ripple effects loop). The impacts of these reinforcing feedback loops on project durations (relative to planned duration and each other) have been quantified in the current study to reveal the sizes of their influences. For the realistic project structure and values used here, not adequately managing a few common reinforcing loops can cause enormous delays (between 107 and 473 percent of planned project duration). Simulation results show that there is a statistically significant interaction between the reinforcing feedback loops. These interactions create synergies that accentuate the effects of individual loops, making them more difficult to manage. The synergy created by the interaction of different reinforcing feedback loops implies that to reduce project delays these reinforcing feedback loops should be managed together. Focusing on individual reinforcing loops does not guarantee the success of the project. Finally, the dominance of the rework loop does not depend on the values of the variables. For different values of the three main variables (i.e. rework fraction, sensitivity to schedule pressure, and ripple effects strength), the rework cycle remains dominant. This implies that, without ignoring other feedback loops, project managers should focus on the rework cycle.

This work has taken a first step in filling an important gap in system dynamics: the need to effectively, efficiently, and broadly communicate the sizes of the impacts of reinforcing loops on project duration and thereby project performance. The main impacts and interactions between three common reinforcing loops in construction projects have been quantified and disaggregated. The results show that, even without other structures or exogenous changes, failing to account for even a few common reinforcing loops can cause enormous delays. Quantifying those delays in a small and simple project model provides system dynamicists a valuable exemplar and information set for communicating how system dynamics can improve project planning and management. The validated and simple project model used in the study can be applied by project management trainers and educators to demonstrate and explain the impacts of reinforcing loops and other factors on project success and duration. The model can be expanded to include other feedback loops and project control (e.g. overtime, hiring, and deadline changes) for further studies.

There are some limitations to this work; however, they create research opportunities. These include that the results reflect the basic variable values of the Limerick project and partially depend on the specific experimental values chosen. To improve the description of the impacts of reinforcing loops on project duration, this work can be extended to evaluate the influence of exogenous variables such as initial scope, project deadline, and total staff; and the positive impacts of schedule pressure on project performance. Finally, in this study, the three variables controlling loop strength were assumed to be a constant throughout each project simulation. However, these variables can change over the life of a project. Changing these variables into dynamic variables is another area for further improvement.

Acknowledgements

The authors thank the reviewers of a preliminary version for their valuable comments, and Tom Fiddaman for inspiring this investigation through his blog entry on the distribution of project durations based on a similar model.

Appendix 1: 23 factorial design

The assumption in a two-level factorial design is that the effects of the parameters are linear. For a two-level factorial design with three parameters, the generic design matrix is shown in the Table A1. In the table, “−” represents the low level of the parameter and “+” represents the high level of the parameter (0 and 0.35 for the current study). The column “Response” 7 shows the dependent variable under study (project duration in the current study).

Table A1. Generic 2k factorial design matrix
Factor combinationFactor 1Factor 2Factor 3Response
1R1
2+R2
3+R3
4++R4
5+R5
6++R6
7++R7
8+++R8

The main effect of a parameter is the average change in the response by changing the parameter from its low level to its high level (while other factors are fixed). It can be computed by finding the difference in the response when the parameter is at its high level from when it is at its low level. For example, the main effect of factor 1 can be calculated by subtracting R1 from R2. However, this effect is only valid for the cases that factor 2 and factor 3 are at their low level. The main effect of factor 1 can be calculated in four different ways using the results from the table above: R2 − R1, R4 − R3, R6 − R5 and R8 − R7. Thus the main effect of factor 1 is the average of these combinations:

display math(1)

Similar reasoning is applied to find the main effect of factor 2 and factor 3 as follows:

display math(2)
display math(3)

When the effect of a factor depends on the level of another factor, it is said that they interact. The interaction between factor 1 and factor 2 can be measured by finding the difference in the average effect of factor 1 when factor 2 is at its high level and the average effect of factor 1 when factor 2 is at its low level. One half of this difference is referred to as a two-factor interaction effect. For example, the effect of factor 1 when factor 2 is at its high level can be calculated as R4 − R3 or R8 − R7 (depending on the level of factor 3). Thus the average effect of factor 1 when factor 2 is at its high level is math formula. Similarly, the average effect of factor 1 when factor 2 is at its low level is math formula. Therefore, the two-factor interaction effect is

display math(4)

Using similar steps, the other two-factor interaction effects can be computed:

display math(5)
display math(6)

The higher-factor interaction effects can be calculated using the same reasoning. The three-factor interaction in a 23 factorial design is computed by the following formula:

display math(7)

Law (2015, ch. 12) and Montgomery (2014, ch. 14) give a full description of design of experiments using 2k factorial design, as well as methods for determining the statistical significance and instructions to develop the response surface.

After running the simulations, the average responses (R1–R8) were found. The formulas were used to calculate the main effects and the interaction effects of loop-strength variables as shown in Table A2. The last column demonstrates the contribution of each effect to the total delay.

Table A2. Main and interaction effects of the reinforcing loops
EffectReinforcing loop(s)Delay (months)Relative effect (%)
e1Rework cycle main effect8022.4
e2“Haste makes waste” main effect5314.9
e3Ripple effects main effect4613.0
e12Rework cycle/“haste makes waste” interaction effect5214.7
e13Rework cycle/ripple effects interaction effect4612.9
e23“Haste makes waste”/ripple effects interaction effect3911.1
e123Rework cycle/“haste makes waste”/ripple effects interaction effect3911.0
 Total with all three loops active355100.0

These results were used to calculate the delays in the six cases as described below:

C1

Rework cycle loop = e1 = 80

C2

“Haste makes waste” loop = e2 = 53

C3

Rework cycle and “haste makes waste” loops = e1 + e2 + e12 = 80 + 53 + 52 = 186

C4

Rework cycle and ripple effects loops = e1 + e3 + e13 = 80 + 46 + 46 = 172

C5

“Haste makes waste” and ripple effects loops = e2 + e3 + e23 = 53 + 46 + 39 = 138

C6

Rework cycle, “haste makes waste”, and ripple effects loops = e1 + e2 + e3 + e12 + e13 + e23 + e123 = 80 + 53 + 46 + 52 + 46 + 39 + 39 = 355

Appendix 2: Results of analysis of variance (ANOVA) and Student's t-test

ANOVA and Student's t-test were performed to find the statistical significance of each impact. The F-statistics, t-statistics and their correspondent p-values show that all main impacts and the interaction effects are statistically significant (Table A3). The adjusted R2 of the overall model is 0.94, indicating that the impacts account for 94% of the variability in the project duration.

Table A3. Results of statistical significance testing
RankReinforcing loop(s)F-valueProb. > Ft-valueProb. > |t|
1Rework cycle40982.6<.0001202.44<.0001
2“Haste makes waste”18143.9<.0001134.70<.0001
3Rework cycle and “haste makes waste”17554.9<.0001132.50<.0001
4Ripple effects13722.5<.0001117.14<.0001
5Rework cycle and ripple effects loops13635.3<.0001116.77<.0001
6“Haste makes waste” and ripple effects loops10114.0<.0001100.57<.0001
7Rework, “haste makes waste”, and ripple effects loops10039.1<.0001100.20<.0001

Appendix 3: Statistical screening

Procedures for statistical screening were followed using Ford and Flynn (2005) and Taylor et al. (2010). Figure A1 shows the correlation coefficients between the three main parameters (i.e. “rework fraction”, “ripple effects strength” and “sensitivity to schedule pressure”) and the project duration. As shown in the figure, the rework cycle is highly correlation with the project duration. The correlation between the two other factors and the project duration increases over time.

Figure A1.

Statistical screening results

Biographies

  • Yasaman Jalili is a PhD candidate in Construction Engineering and Management in the Zachary Department of Civil Engineering at Texas A&M University. She holds a BS and MS in Civil Engineering from Tehran Polytechnic. Her research interests include tipping point dynamics, strategic project management, and real options.

  • David N. Ford is the Beavers Charitable Trust / William F. Urban ′41 Professor in Construction Engineering and Management in the Zachary Department of Civil Engineering at Texas A&M University. He teaches and researches project dynamics, managerial real options, innovative financing of sustainability, and community recovery from shock events. As an engineer in practice Dr Ford designed and managed government, commercial, and residential development projects. He received Bachelor's and Master's degrees in Civil Engineering from Tulane University and a PhD from MIT in Dynamic Engineering Systems.

Footnotes

  1. 1See Lyneis and Ford (2007) for a summary and review, and Ford and Lyneis (2013) for a collection of relevant papers.
  2. 2The dynamic complexity of projects may also makes it difficult for most system dynamicists to estimate the sizes of these impacts.
  3. 3The model as used here includes no deliberate project control (e.g. overtime, hiring, and deadline changes) other than distributing resources among the three development activities. See Ford et al. (2007) for a project model with proactive project controls.
  4. 4The changed model with documentation (Martinez-Moyano, 2012) is available as supporting information.
  5. 5Defined as a project that completes within 20 times its original project deadline.
  6. 6The experiment was repeated for different ranges of uncertainties (10–30 percent) and the ranking of the relative impacts of the feedback loops was found to be the same. In addition, the same analysis was performed using a normal distribution instead of a uniform distribution and the results were robust.
  7. 7The abbreviation “D” was used in the text for the Project Duration response variable to avoid confusion with the naming of the reinforcing feedback loops (R1, R2, R3).