+ - Copyright 2003 James M. Lyneis Dynamics of Project Performance System Dynamics and Project Management Class Three (9/30/03) - ESD.36J SPM 9/30/03 2 + - Copyright 2003 James M. Lyneis Topics Review Practice for 9/30 The dynamics of project performance -- knock-on effects Modeling the feedback effects Model Calibration Practice for 10/2 - ESD.36J SPM 9/30/03 3 + - Copyright 2003 James M. Lyneis Topics Review Practice for 9/30 The dynamics of project performance -- knock-on effects Modeling the feedback effects Model Calibration Practice for 10/2 - ESD.36J SPM 9/30/03 5 + - Copyright 2003 James M. Lyneis What did you conclude re. the effect of productivity, quality, and rework discovery time on time required to finish the project? - ESD.36J SPM 9/30/03 6 + - Copyright 2003 James M. Lyneis Sensitivity of Completion Date to +/-33% Variation in Parameters Quality - Completion Date 0 20 40 60 80 Low Q Base High Q Productivity - Completion Date 0 20 40 60 80 Low Pdy Base High Pdy Rework Discovery Delay - Completion Date 0 20 40 60 80 Low Rwdd Base High Rwdd - ESD.36J SPM 9/30/03 7 + - Copyright 2003 James M. Lyneis Sensitivity of Total Work Done to +/-33% Variation in Parameters Quality - Total Work Done 0 50 100 150 200 Low Q Base High Q Productivity - Total Work Done 0 50 100 150 200 Low Pdy Base High Pdy Rework Discovery Delay - Total Work Done 0 50 100 150 200 Low Rwdd Base High Rwdd - ESD.36J SPM 9/30/03 8 + - Copyright 2003 James M. Lyneis Sensitivity of Completion Date Variations in Quality have biggest impact -- Nonlinear effect Determines total amount of work Rework discovery time has the least impact (but the simple model does not have a quality on quality feedback). - ESD.36J SPM 9/30/03 9 + - Copyright 2003 James M. Lyneis Graph for Undiscovered Rework 8 6 4 2 0 0 5 10 15 20 25 30 35 40 45 50 Time (Month) Undiscovered Rework : Class2 Step1 Task Undiscovered Rework : Class2 Step1 Q100 Task Undiscovered Rework : Class2 Step1 Q50 Task Quality = .5 Quality = .75 Quality = 1.0 - ESD.36J SPM 9/30/03 10 + - Copyright 2003 James M. Lyneis Graph for Work Done 100 75 50 25 0 0 5 10 15 20 25 30 35 40 45 50 Time (Month) Work Done : Class2 Step1 Task Work Done : Class2 Step1 Q100 Task Work Done : Class2 Step1 Q50 Task Quality = 1.0 Quality = .75 Quality = .5 - ESD.36J SPM 9/30/03 11 + - Copyright 2003 James M. Lyneis Rework Shows Up As Revised Work Products Virtually all complex projects have revision cycles. REV 0 REV 1 REV 3... REV 2 Accomplishment of Revisions to Design Products (e.g., Drawings) TIME - ESD.36J SPM 9/30/03 12 + - Copyright 2003 James M. Lyneis Managing the rework cycle Do common project metrics handle (undiscovered) rework? Do management incentives encourage the discovery of rework? - ESD.36J SPM 9/30/03 13 + - Copyright 2003 James M. Lyneis Additional Metrics (Used in Model) Fraction Perceived to be Complete = (Work Done + Undiscovered Rework) / Initial Work to Do Fraction of Work Really Complete = (Work Done) / Initial Work to Do - ESD.36J SPM 9/30/03 14 + - Copyright 2003 James M. Lyneis Perceived vs. Really Complete in Simple Rework Cycle Model Progress 1 0.75 0.5 0.25 0 0 6 12 18 24 30 36 42 48 54 60 Time (Months) Fraction Perceived to be Complete : Step2 Fraction Fraction Really Complete : Step2 Fraction - ESD.36J SPM 9/30/03 15 + - Copyright 2003 James M. Lyneis On an actual project Perceived vs. Actual Progress: SRS Development Perceived Actual 100. 75. 50. 25. 0. TIME 1/93 1/94 1/95 1/96 1/97 1/98 1/99 1/00 1/01 1/02 c:\mydocu~1\jml\hughes\hisjml.dyn 8/19/97 9:29 Disguised results from actual aerospace project Year 1 Year2 Year 3 Year 4 Year 5 Year 6 Year 7 Year 8 Year 10 Year 9 - ESD.36J SPM 9/30/03 16 + - Copyright 2003 James M. Lyneis Topics Review Practice for 9/30 The dynamics of project performance -- knock-on effects Modeling the feedback effects Model Calibration Practice for 10/2 - ESD.36J SPM 9/30/03 17 + - Copyright 2003 James M. Lyneis Examples of Project Behavior Modes Project Staffing Time Typical Plan - ESD.36J SPM 9/30/03 18 + - Copyright 2003 James M. Lyneis Examples (continued) Productivity (Normalised) Time 1 2 Typical Plan - ESD.36J SPM 9/30/03 19 + - Copyright 2003 James M. Lyneis Dynamics of Project Performance The rework cycle Quality Undiscovered rework Feedback effects on productivity and work quality Positive, re-enforcing, often vicious circles Negative, controlling Knock-on effects between work phases - ESD.36J SPM 9/30/03 20 + - Copyright 2003 James M. Lyneis A system dynamics model usually represents several phases of work, but is more aggregate than a CPM model System Engineering Software Code and Test Hardware Design Hardware Build and Test Integrate and Test Software Design - ESD.36J SPM 9/30/03 21 + - Copyright 2003 James M. Lyneis Software System Engineering Software Development Peace Shield Air Defence System Model Architecture SRS Development HW Installation & ICO Test Op. & Main/ ILS Mgt. & Admin. Top Level & Detailed Design Code & Unit Test Integration & Type II Test Type I Test KOSA SW & Testing Program Management Office Software Support CONUS Customer Hardware Logistics Subcontractors Other HASI Programs Downstream Progress, Availability, & Quality Effects Upstream Rework Discovery Effects Support Effects - ESD.36J SPM 9/30/03 22 + - Copyright 2003 James M. Lyneis Inter-phase Effects Downstream &/or Parallel -- Availability of work products Quality of work products Upstream -- Rework discovery Above apply internally and to suppliers General -- Sharing of staff - ESD.36J SPM 9/30/03 23 + - Copyright 2003 James M. Lyneis LABOR APPLIED REWORK DISCOVERY WORK BEING DONE WORK TO BE DONE WORK REALLY DONE UNDISCOVERED REWORK KNOWN REWORK Productivity Quality Inter-phase Connections: Upstream-Downstream LABOR APPLIED REWORK DISCOVERY WORK BEING DONE WORK TO BE DONE WORK REALLY DONE UNDISCOVERED REWORK KNOWN REWORK Productivity Quality Progress; Quality Progress Design Construction - ESD.36J SPM 9/30/03 24 + - Copyright 2003 James M. Lyneis LABOR APPLIED REWORK DISCOVERY WORK BEING DONE WORK TO BE DONE WORK REALLY DONE UNDISCOVERED REWORK KNOWN REWORK Productivity Quality Inter-phase Connections: Parallel LABOR APPLIED REWORK DISCOVERY WORK BEING DONE WORK TO BE DONE WORK REALLY DONE UNDISCOVERED REWORK KNOWN REWORK Productivity Quality Progress; Quality Progress: Quality Software Design Hardware Design - ESD.36J SPM 9/30/03 25 + - Copyright 2003 James M. Lyneis Dynamics of Project Performance The rework cycle Quality Undiscovered rework Feedback effects on productivity and work quality Positive, re-enforcing, often vicious circles Negative, controlling Knock-on effects between work phases Availability and quality of work products Progress to discover upstream rework - ESD.36J SPM 9/30/03 26 + - Copyright 2003 James M. Lyneis How might we represent suppliers? platforms? ... - ESD.36J SPM 9/30/03 27 + - Copyright 2003 James M. Lyneis Topics Review Practice for 9/30 The dynamics of project performance -- knock-on effects Modeling the feedback effects Model Calibration Practice for 10/2 - ESD.36J SPM 9/30/03 28 + - Copyright 2003 James M. Lyneis If the world consists only of stocks and flows, what are those other variables indicated on the diagrams?? Auxillaries Constants (e.g., factors which may be stocks or flows, but which do not change over the time span of the simulation) External inputs - ESD.36J SPM 9/30/03 29 + - Copyright 2003 James M. Lyneis Auxillaries or Converters Serve Several Purposes ... Break up rates into meaningful components Provide alternative measures for stocks or flows Reduce diagram clutter - ESD.36J SPM 9/30/03 30 + - Copyright 2003 James M. Lyneis Potential Work Rate Potential Work Rate = Staff * Productivity Tasks/Month People * Tasks/Month/Person - ESD.36J SPM 9/30/03 31 + - Copyright 2003 James M. Lyneis Productivity & Quality Productivity = ?? Quality = ?? - ESD.36J SPM 9/30/03 32 + - Copyright 2003 James M. Lyneis Productivity -- PRODUCTIVITY = NORMAL PRODUCTIVITY * EFFECT OF STAFF EXPERIENCE * EFFECT OF QUALITY OF PRIOR WORK * ... Dimensions: Productivity -- Tasks/Month/Person Normal Productivity -- Tasks/Month/Person Effects -- Dimensionless - ESD.36J SPM 9/30/03 33 + - Copyright 2003 James M. Lyneis Quality -- QUALITY = NORMAL QUALITY * EFFECT OF STAFF EXPERIENCE * EFFECT OF QUALITY OF PRIOR WORK * ... Dimensions: Quality -- Fraction Normal Productivity -- Fraction Effects -- Dimensionless - ESD.36J SPM 9/30/03 34 + - Copyright 2003 James M. Lyneis Effect of Staff Experience Effect of Experience = Function ( Years on Project) - ESD.36J SPM 9/30/03 35 + - Copyright 2003 James M. Lyneis Effect of Staff Experience Effect of Experience = Function (Years on Project) ? 0 .2 .4 .6 .8 1.0 2 4 6 8 10 Years on Project - ESD.36J SPM 9/30/03 36 + - Copyright 2003 James M. Lyneis How do we determine these effects? The effects are first estimated based on common sense, later to be verified during model calibration. It helps to specify likely values at extreme points, and draw a smooth curve in between. - ESD.36J SPM 9/30/03 37 + - Copyright 2003 James M. Lyneis Effect of Staff Experience Extreme points -- ? 0 .2 .4 .6 .8 1.0 2 4 6 8 10 X X Years on Project - ESD.36J SPM 9/30/03 38 + - Copyright 2003 James M. Lyneis Connecting the extreme points ... 0 .2 .4 .6 .8 1.0 2 4 6 8 10 X X Years on Project - ESD.36J SPM 9/30/03 39 + - Copyright 2003 James M. Lyneis Some other connections are plausible ... 0 .2 .4 .6 .8 1.0 2 4 6 8 10 Years on Project - ESD.36J SPM 9/30/03 40 + - Copyright 2003 James M. Lyneis Others are not ... 0 .2 .4 .6 .8 1.0 2 4 6 8 10 Years on Project - ESD.36J SPM 9/30/03 41 + - Copyright 2003 James M. Lyneis Relationships Similar to the Experience Effect Would Be Specified For Other Effects on Productivity and Quality Experience Adequacy of supervision Availability of resources Morale Schedule pressure Fatigue Quality of prior work Skills available vs. needs Tools Processes Availability of supplier information &/or materials ... - ESD.36J SPM 9/30/03 42 + - Copyright 2003 James M. Lyneis Note difference with statistical estimation -- Productivity = a + b*experience+c*prior quality - ESD.36J SPM 9/30/03 43 + - Copyright 2003 James M. Lyneis Rules of Thumb and Checks 1. Variables and constants should correspond to real system variables 2. Remember to represent delays -- in flow channels, in perceiving information, and in making or implementing decisions 3. Distinguish desires from realizations 4. Make sure that equations balance dimensionally 5. Check the validity of equations at extreme points, e.g., zero or very high values of the inputs 6. Avoid the use of imbedded functions -- they make the equations harder to understand and can lead to errors - ESD.36J SPM 9/30/03 44 + - Copyright 2003 James M. Lyneis Connecting Morale & Overtime ... from a students email: On a couple of occasions in lecture yesterday you discussed "Overtime" as it relates to "Morale". In doing causal loops, I find that the relationship between the 2 can be different and dependent on time. For example, initially the morale of employees may increase or support the increase of overtime which provides a positive relation between the 2. However, as the overtime continues, I believe the morale will decrease as employees will start to desire a life outside of work which supports a negative relation between the 2. I guess that in the actual SD modeling the developer would create a curve/equation for morale over time, but what about for causal loops? How do you handle the dynamics with causal loops? Maybe the loops are just an initial starting point for the development? - ESD.36J SPM 9/30/03 45 + - Copyright 2003 James M. Lyneis Representing morale - ESD.36J SPM 9/30/03 46 + - Copyright 2003 James M. Lyneis Topics Review Practice for 9/30 The dynamics of project performance -- knock-on effects Modeling the feedback effects Model Calibration Practice for 10/2 - ESD.36J SPM 9/30/03 47 + - Copyright 2003 James M. Lyneis Calibration (Parameter Estimation) - ESD.36J SPM 9/30/03 48 + - Copyright 2003 James M. Lyneis Parameter Estimation in System Dynamics A priori estimates based on direct observation educated guesses by managers similar parameters from other applications Refine initial estimates in the process of calibration - ESD.36J SPM 9/30/03 49 + - Copyright 2003 James M. Lyneis Calibration is part of the scientific method Calibration is not curve-fitting via exogenous variables, but systematic comparison of model output to all data and resolution of any discrepancies changes in structure changes in parameters decision to ignore (or sometimes use exogenous effect) - ESD.36J SPM 9/30/03 50 + - Copyright 2003 James M. Lyneis Calibration Should Be a Part of the Standard Method Data is a real source of information Calibration often uncovers errors/omissions in insight-based models Cost is small relative to the stakes in most large organizations - ESD.36J SPM 9/30/03 51 + - Copyright 2003 James M. Lyneis Why do we calibrate a model? Assure model contains all of the structure necessary to create problem Accurately determine leverage points Accurately price cost-benefits of alternatives Sell results to those not on project team Generate a benchmarking data base for future use and learning - ESD.36J SPM 9/30/03 52 + - Copyright 2003 James M. Lyneis Sources of information for calibration Hard, numerical time series (e.g., for staffing, work accomplished, overtime levels, etc.) Soft, management observations of what happened (e.g., how morale, experience, and other factors might have caused productivity and quality to change) such observations must be tied to what was happening on the project - ESD.36J SPM 9/30/03 53 + - Copyright 2003 James M. Lyneis Work Quality to Date Scheduled Completion Time Expected Completion Time Availability of Prerequisites Perceived Progress Schedule Pressure Out-of-Sequence Work Morale Expected Hours at Completion Hours Expended to Date Skill & Experience Hiring Equivalent Staff on Project Staffing Requested Progress Rework Discovery Turnover Organisational Size Changes Staff Productivity Quality Added Work Obsoleted Work Overtime Time Remaining Work To Be Done Undiscovered Rework Known Rework Work Really Done Project Dynamics Data likely to be available - ESD.36J SPM 9/30/03 54 + - Copyright 2003 James M. Lyneis Once calibration is completed ... We have a model that is useful for: Learning what really happened on prior projects what could have been done better Dispute resolution - ESD.36J SPM 9/30/03 55 + - Copyright 2003 James M. Lyneis What happens if you cant calibrate? Use logical parameter estimates Simulation model still better than mental models - ESD.36J SPM 9/30/03 56 + - Copyright 2003 James M. Lyneis Topics Review Practice for 9/30 The dynamics of project performance -- knock-on effects Modeling the feedback effects Model Calibration Practice for 10/2 - ESD.36J SPM 9/30/03 57 + - Copyright 2003 James M. Lyneis Lessons Rework (Errors) Can Linger for Long Periods of Time as Undiscovered Rework Reducing Rework, and Detecting it Earlier, is the Best Way to Cut Cycle Times, Finish Earlier, and increase Delivered Quality Common metrics & reward systems do not encourage high quality and rework discovery Interactions between phases of work, different projects, suppliers, create opportunities for knock- on effects Calibration will improve the accuracy of our parameter estimates, and ultimately policy decisions - ESD.36J SPM 9/30/03 58 + - Copyright 2003 James M. Lyneis Practice for 10/2 1. Download the Class2 Step2 model. This is the rework cycle model with a variable time to discover rework. Time to discovery rework, as illustrated in the following figure, depends on a maximum time and a minimum time, and an effect of work progress. Time to discover rework falls from the maximum to the minimum time as fraction complete increases. - ESD.36J SPM 9/30/03 59 + - Copyright 2003 James M. Lyneis Rework cycle with variable time to discover rework Work to Do Work Done Undiscovered Rework Work Accomplishment Rework Generation Rework Discovery Quality Staff Productivity Potential Work Rate Project Finished Switch Cumulative Work Done Rate of Doing Work <Rework Generation> <Work Accomplishment> Feasible Work Rate Maximum Work Rate Minimum Time to Perform a Task Time to Discover Rework <Work Done> Work Believed to Be Done Fraction Really Complete Fraction Perceived to be Complete <Initial Work to Do> Effect of Work Progress Maximum Time to Discover Rework Table for Effect of Work Progress Minimum Time to Discover Rework - ESD.36J SPM 9/30/03 60 + - Copyright 2003 James M. Lyneis Practice for 10/2 2. Add the effect of prior work quality on quality as discussed in class and illustrated in the following figure. 3. Simulate the model and compare to the original Class2 Step 2 results. How does the addition of quality on quality feedback affect project performance? 4. If we essentially eliminate undiscovered rework by setting the maximum time to discover rework to 0.25 months, what happens to project performance? - ESD.36J SPM 9/30/03 61 + - Copyright 2003 James M. Lyneis Work to Do Work Done Undiscovered Rework Work Accomplishment Rework Generation Rework Discovery Quality Staff Level Productivity Potential Work Rate Time to Discover Rework Project Finished Switch Cumulative Work Done Rate of Doing Work <Rework Generation> <Work Accomplishment> Effect of Prior Work Quality on Quality Normal Quality Table for Effect of Prior Work Quality on Quality <Work Done> Work Believed to Be Done Fraction Really Complete Fraction Perceived to be Complete <Initial Work to Do> Cumulative Effort Expended Effort Expended <Staff Level> <Project Finished Switch> Effect of Work Progress Maximum Time to Discover Rework Table for Effect of Work Progress Average Work Quality Feasible Work Rate Maximum Work Rate Minimum Time to Perform a Task <Initial Work to Do> Minimum Time to Discover Rework - ESD.36J SPM 9/30/03 62 + - Copyright 2003 James M. Lyneis Hints -- 1 Effect of Prior Work Quality on Quality: kup - Table for Effect of Prior Work Qua 1 0 0 1 Note: The effect of prior quality on current quality is assumed to be proportional -- and error in past work creates an error in current work. Given that in this simple model quality represents several effects of work errors, this strong relationship may be reasonable. - ESD.36J SPM 9/30/03 63 + - Copyright 2003 James M. Lyneis Hints -- 2 Average Work Quality: Max(1e-006,Work Done)/Max(1e- 006,Work Believed to Be Done) [Note: the max in the denominator prevents division by 0, and in the numerator allows the equation to return a quality of 1.0 when no work is yet completed.] - ESD.36J SPM 9/30/03 64 + - Copyright 2003 James M. Lyneis Hints -- 3 Normal Quality: Set this to 0.85 (vs. 0.75 in Class2 Step2 model) to reflect that we are now explicitly representing other factors affecting project quality. - ESD.36J SPM 9/30/03 65 + - Copyright 2003 James M. Lyneis Graphical Functions in Vensim - ESD.36J SPM 9/30/03 66 + - Copyright 2003 James M. Lyneis Step 1 - Set Up Model As Shown Below Quality Normal Quality Effect of Prior Work Quality Average Work Quality Table for Effect of Prior Work Quality - ESD.36J SPM 9/30/03 67 + - Copyright 2003 James M. Lyneis Step 2 -- Equations ... Quality = Normal Quality*Effect of Prior Work Quality Effect of Prior Work Quality = Table for Effect of Prior Work Quality(Average Work Quality) - ESD.36J SPM 9/30/03 68 + - Copyright 2003 James M. Lyneis Step 3 -- Define the Table Click to define equation for Table for Effect of Prior Work Quality Select the Type as Lookup Click on the As Graph button to see the function You can then type in values or create a line on the blank graph You can edit the values directly if desired - ESD.36J SPM 9/30/03 69 + - Copyright 2003 James M. Lyneis Voila! kup - Table for Effect of Prior Work Qua 1 0 0 1 - ESD.36J SPM 9/30/03 70 + - Copyright 2003 James M. Lyneis In the short-term, overtime increases morale and productivity Productivity Morale Overtime Income + + + - ESD.36J SPM 9/30/03 71 + - Copyright 2003 James M. Lyneis But in the longer-term, sustained overtime reduces morale and productivity ... Productivity Morale Overtime Income + + + Average Overtime + - - ESD.36J SPM 9/30/03 72 + - Copyright 2003 James M. Lyneis Connecting the feedbacks through progress ... Productivity Morale Overtime Income + + + Average Overtime + - Progress Staff Needed + - + +