Schedule Characteristics in QSRA
Santosh B.
Independent Consultant for Project Planning and Scheduling, Schedule Risk Analyses and also Co-Founder of Turbo-Chart
Many discussions on Project Quantitative Risk Analysis focus debate on the "quantification" of risks, particularly the parameters of risk impact values, distribution profiles for those impact values etc. While this can be an important factor when all risks are aggregated into a single value, when considering schedule risk, there are far more significant factors - the characteristics and behaviour of the schedule being used for the analysis.
In a recent post by Alexey Belkov, PMP, CRMP I made the below comment and in this article I want to expand on this further
Schedule characteristics are more determinant on the result than each distribution (for anything other than simple schedules)
If you are dealing with a simple schedule, that is, a straightforward sequence of activities linked in series with finish-to-start (FS) relationships, where there is a single dominant path driving criticality, then certainly, shape selection may make some difference to the results, however, most project schedules are not that simple.
To demonstrate lets consider firstly a very basic schedule, five activities of 20days duration each, linked sequentially FS totalling a duration of 100days.
Schedule A
For the purpose of this demonstration, lets also keep the schedule risk being modelled simple, applying a -25%/+25% duration uncertainty (uncorrelated), using a triangle distribution, to all these tasks:
This means each activity in Schedule A will be modelled using a random duration between 15 to 25days using the above profile. The results of the Monte Carlo on this simple analysis produces the below cumulative distribution profile for the total duration:
Now Lets run further analyses, using the similar impact values, but using some of the more commonly debated risk distribution profiles:
The results using all these different distributions for Schedule A can then be compared below:
The above image shows that changing the selected shape, had very little impact on this simple schedule, Even at the extreme ends of the curves at P10 or P90, the difference between profiles represents about 2days.
Schedule C
Lets model a less-simple schedule. This schedule consists of three sets of Schedule A, that is, three separate paths of 100days duration. The completion of the entire project will be affected by merge bias of these separate paths:
Before comparing the different Risk distribution profiles. Lets just run the original Triangle Risk profile onto Schedule C and compare this result with Schedule A:
What this indicates is that, at P50 the results are later in Schedule C than in Schedule A, despite both being 100days duration - due to the effect of Merge Bias.
If we compare just Schedule C using the four different risk profiles, the below is the result:
The above result shows very little variance in result by using different profiles.
Now lets get a little more sophisticated, rather than a basic schedule of FS relationships, lets use a series of 40day tasks, overlapped using Start-to-Start (SS) and Finish-to-Finish(FF) relationships, that again total 100days duration:
Schedule B
Running the original Triangle Risk profile onto Schedule B and compare this result with Schedule A and Schedule C:
The above shows that using the overlapped sequence of tasks generates a n even later result using Triangular distribution profile - this is again due to merge bias within the sequence of activities - here's a link to a 2015 article I wrote explaining this further: How to deal with Schedule Risk in an overlapped sequence of activities | Australasian Project Planning (austprojplan.com.au)
领英推荐
So using Schedule B alone, lets compare the different risk distribution profiles:
The above result again, shows very little different in results using different risk profiles.
Finally lets use three separate paths of overlapped activities still totalling 100days for the project.
Schedule D
Running the original Triangle Risk profile onto Schedule D and comparing this result with Schedule A, C & B:
Introducing the overlapped sequence of work, with the separate paths creates further merge points, resulting in later results. So using Schedule D alone, lets compare the different risk distribution profiles:
Again, very little difference in results using different risk profiles, compared to the change in results using different schedules.
Now consider that most project schedules are not as simple as those used in this example, but consist of hundreds or thousands of activities using many and varied overlapped/parallel sequences of work (ie. significant merge points), date constraints and potentially even multiple available work periods (calendars).
It is clear that these characteristics of a schedule are far more influential in QSRA results than which Risk distribution profile is used.
Risk Drivers
But, the above uses the method of activity ranging, ie, directly assigning risk profiles (impacts and distribution curves) to activities. This has two fundamental issues:
Thankfully, the Risk Driver method, as described by AACE
Recommended Practice No. 57R-09, Integrated Cost and Schedule Risk Analysis Using Monte Carlo Simulation of a CPM Model
Solves the dilemma of which profile to assign to individual activities, by considering the aggregation of individual risks that can result in activities having profiles that look such:
Which also allows us to use the Risk By Exclusion method as developed by David Hulett, Ph.D. FAACE to understand the impact of any one risk (by simply excluding it from the analysis)
Additional Outputs
For completeness, here are the results of each Risk Distribution profile across each schedule
Project Engineering Manager
3 个月Thank for sharing! Interesting article!
LNG & Gas Processing Expert | Process & Flow Assurance Engineer | Chartered Chemical Engineer (CEng, MIChemE) | Operations Readiness | Machine Learning in Process Optimization
5 个月Thanks for sharing, insightful. Schedule risk analysis is really important for all team working on a project, though it might be of a focus of project controls team or schedulers. MonteCarlo distribution is helpful if you’ve got a precise project schedule data from previous projects, and the analysis is dependent on number of iterations you’ve used. I’ve two questions: 1) does it make differ if we choose the MC distribution of current risk as “triangle” and the MC distribution of target risk as “betapert”?. 2) if a task fall within the critical path, can it be not a critical path any more after the analysis, or critical path remains as it is since no changes in dates?
Construction Scheduling Trainer; Author, “CPM Mechanics”
1 年(2 of 2) ICS-Global has long advocated that a Project Schedule, once signed off by the Project Team, is a contract, agreement, a commitment. I have a hard time understanding how or why regression analysis is a positive tool once the project has started (unless, for instance, there are specific Known Unknown risk factors -- not activity durations -- that are looming on the horizon). Can we all agree that such analyses as your article opines about are meant to support predictions of future temporal outcomes? Isn't that the point, or goal, of such exercises? If so, then let us use them in the earliest phases of a project, when long-term estimates of completion deadlines can be beneficial to long-term planning. But once the project begins, there are far better -- and far more accurate -- ways to speculate on the timeliness of the project's forward progress, than treating Activity Durations as static values devoid of human involvement.
Construction Scheduling Trainer; Author, “CPM Mechanics”
1 年Santosh B. Excellent article that well presents a complicated topic in fairly simple terms. But I am in agreement with Raphael M Düa - DBA,FAICD,GPCF,FAPE,MACS(Snr),CP, IP3, GDISC in that our arcane fascination, some what say obsession, with probability analysis has us floating in a theoretical world of statistics, while taking us that much farther from where the rubber meets the road. I have long argued that Risk Analysis has its value during the planning phase of a project, with increasing value correlated to the uncertainty of durations. But where durations are fairly well known (as in most construction, versus, say, software development), the calculation effort seems to generate a distinction without a meaningful difference. Add to that, as my friend Raphael does, that once the project is underway, the very reason for performing such calculations becomes central. As an example, if I tell my daughter that she can count on me to pick up my grand-daughter at her day care, my daughter would shiver if I started running Monte Carlo on the probability that I will be there on time. Instead, the more comforting attitude, by both of us, is "I said I will be there, and I will." (1 of 2)
Director dealing with claims and construction M&A at M&I Doradztwo Sp z o.o
1 年Hello Santosh, I find what you have done here to be extremely valuable and important. The significance relates to the overlap, and this is where in my view a lot of problems arise in collapsed as built due to the commonality of the paths, and the associated changes in the critical path. The difference in the risk profile is correct, and what you have pointed out is very significant.