Monday

CPM Anomalies Invalidate Monte Carlo

A couple of weeks ago I posted on some of the anomalies in CPM logic that will cause unexpected results: CPM Scheduling – the logical way to error #1. A comment on the post by Santosh Bhat started me thinking about the effect of these logical constructs on risk analysis.

The various arrangement of activities and links shown in CPM Scheduling – the logical way to error #1 (with the addition of a few more non-controlling links) follow all of the scheduling rules tested by DCMA and other assessments. The problem is when you change the duration of a critical activity, there is either no effect or the reverse effect on the overall schedule duration.

In this example, the change in the overall project duration is the exact opposite of the change in the duration of Activity B (read the previous post for a more detailed explanation).  For this discussion, it is sufficient to know that an increase of 2 weeks in the duration of ‘B’ results in a reduction of the overall project duration of 2 weeks (and vice-versa).

The effect these anomalies on the voracity of a Monte Carlo analysis is significant. The essence of Monte Carlo is to analyze a schedule 100s of times using different activity durations selected from a pre-determined range that represents the uncertainty associated with each of the identified risks in a schedule. If the risk event occurs, or is more serious, the affected activity duration in increased appropriately (see more on Monte Carlo). 

In addition to calculating the probability of completing by any particular date, most Monte Carlo tools also generate tornado charts showing the comparative significance of each risk included in the analysis and its effect on the overall calculation.  For example, listing the risks that have the strongest correlation between the event occurring and the project being delayed.  

Tornado charts help the project’s management to focus on mitigating the most significant risks.

When a risk is associated with an activity that causes on of the anomalies outlined in CPM Scheduling – the logical way to error #1 the consequence is a reduction in the accuracy of the overall probability assessments, and more importantly to reduce the significance of the risk in tornado charts. The outcome of the anomalous modelling is to challenge the fundamental basis of Monte Carlo. There are more examples of similar logical inconsistencies, that will devalue Monte Carlo analysis, included in Section 3.5 of Easy CPM.

Easy CPM is designed for schedulers that know how to operate the tools efficiently, and are looking to lift their skills to the next level. The book is available for preview, purchase (price $35), and immediate download, from: https://mosaicprojects.com.au/shop-easy-cpm.php

3 responses to “CPM Anomalies Invalidate Monte Carlo

  1. Pingback: Critical Path Scheduling – 4 Things People Don’t Get | Mosaicproject's Blog

  2. I can’t remember if I made this point on LinkedIn but it won’t hurt to say it again

    The fatal flaw in the example network above is that Activity-B is supposed to start when someone knows when Activity-A will end and how long Activity-B will take – someone with the ability to foresee the future

    In reality, the optimum time for Activity-B to start will be when preconditions are in place. In this simple example, the preconditions probably relate to Activity-A having reached a certain stage of completion, or arriving at an internal milestone withing Activity-A, completion of one of its subsidiary activities

    If it will be started when it is expected that Activity-A is going to complete in the time it is expected Activity-B will take to complete, it is usually a programmed date or linked to progress in its predecessor as explained in the following paragraph

    I would decline to use logic like that in a network and ask that, either the client identifies the point in Activity-A where Activity-B can start, breaks Activity-A into two parts in the model and links the end of the first part to Activity-B, or make a SS link from Activity-A to Activity-B with a lag equal to the difference between the two activities’ durations and the simulated duration being made to vary in proportion to the variation in the duration of Activity-A, a measure of the rate of progress in the predecessor. In that arrangement, Activity-B would complete either when Activity-A completed (FF link) or when it’s own intrinsic duration dictates, whichever is the later.

    Any activity that has no criterion for when it starts, starts on a FS link with a negative lag or is set ALAP, in both cases requiring an ability to foretell the future, is unrealistic

    That’s not a flaw in MC simulation

    It’s a modelling flaw

  3. The construct is not a recommended practice (at least not by me) but you will be surprised how many schedules have this built in, often hidden by other links to the activities, but as soon as this pattern becomes the controlling links the results go haywire.

Leave a comment