Project Flight Deck Website

The Earned Schedule Exchange

February 29, 2016
LP Improves ES Reliability

Concept: Some project topologies degrade the reliability of ES metrics early in a project’s timeline. ES Longest Path (ES-LP) alleviates the difficulty.

Table 1

Practice: Research by Vanhoucke and Vandevoorde (2009) and Walt Lipke (2014b) shows that project topology affects the reliability of ES forecasts (see previous post). The research also substantiates the view that serial activities engender better ES forecasts than do parallel activities. As Lipke says,

… the best ES forecast is for a serial schedule. (Ibid, p 4)

Taking this insight to its logical conclusion, Lipke suggests that

... For a completely serial schedule, the SPIt [Schedule Performance Index for time ] must describe performance on the critical path [CP]. (Ibid)

In other words, completely serial schedules do not have non-critical activities. The have only critical activities. So, assessment of schedule performance must apply solely to the CP. As non-critical activities are absent, their performance cannot affect ES metrics. Forecast reliability ceases to be problematic.

The reality is, however, that few schedules are completely serial. Is there a way to capture the benefit of serialization within the reality of parallel and hybrid schedules?

The first choice might be to revert to the traditional CP. It is, after all, the quintessential serial path, and it can be identified in schedules with all sorts of topologies. But, there are limitations in focusing solely on the CP. The limitations have been detailed in previous posts (see January 2016). An alternative is required.

Walt Lipke has identified another way to serialize schedules. He calls it the “Longest Path” (LP). As defined by Lipke (2012), the LP is

… the longest duration from among the paths remaining to be executed from the present status point [to the end of the project]. (Ibid, p 4)

How is this different from the Critical Path? The CP serializes the longest planned duration for each period from the present status point to the project’s end. The LP serializes the longest forecasted duration.

Think of it this way. There is a serial path through the schedule whose stepping stones represent the longest forecasted duration in each period. The green cells in Table 1 illustrate the point. (Ignore the red cell for now.)

Each step in the serial path for LP is a duration forecast. It is the longest forecast among those for the same time period. That reduces the possibility of misleading or false forecasts because the selection is indifferent to whether tasks are critical or non-critical. It works like this.

If the performance of critical activities is good or expected in a time period, it will not override poor performance of non-critical activities. The poor non-critical performance will be reflected in the LP, indicating that there is a longer path to completion than that suggested by the critical activity. So, there is no disconnection between SPIt and the actual duration of the project, and the opportunity for reliability problems to arise is reduced.

Similarly, it is impossible for a period to have poor or expected critical activity performance and off-setting good non-critical performance. For any period, LP selects the worst case. The critical path performance will override the good non-critical path performance. Again, there is no disconnection between SPIt and the actual duration of the project, and the opportunity for reliability problems to arise is reduced.

Given that LP uses the worst-case performance expectation for each period, it sidesteps another pitfall. Standard ES forecasts include shorter paths through the schedule. They therefore represent the lower bound (i.e., optimistic) assessment of total schedule performance. LP includes only the longest durations. Thus, the LP can be longer than the total project forecast, and that means it is not the lower bound.

As Lipke sums it up,

… the LP forecast at each status point resolves the …limitations of the ES forecasts, thereby providing better and more direct information for project control. (Ibid)

The way LP is calculated throws further light on the technique and that, in turn, suggests how LP can be practically applied. LP calculations will be addressed in the next post.

References

Lipke, W. (2012). Speculations on Project Duration Forecasting. The Measurable News, III, 1, 4-7.

Lipke, W. (2014a). Examining Project Duration Forecasting Reliability. PM World Journal, III (III).

Lipke, W. (2014b). Testing Earned Schedule Forecasting Reliability. PM World Journal, III (VII).

Lipke, W. (2015). Applying Statistical Forecasting of Project Duration to Earned Schedule-Longest Path. The Measurable News, II, 31-38.

Vanhoucke, M., & Vandevoorde, S. (2008). Earned Value Forecast Accuracy and Activity Criticality. The Measurable News, Summer, 13-16.

Vanhoucke, M., & Vandevoorde, S. (2009). Forecasting a Project’s Duration under Various Topological Structures. The Measurable News, Spring, 26-30.

Add new comment

No Comments

February 25, 2016
Project Topology and ES Reliability

Concept: Some project topologies degrade the reliability of ES metrics early in a project’s timeline. ES Longest Path (ES-LP) alleviates the difficulty.

Practice: Pioneering research by Mario Vanhoucke and Stephan Vandevoorde (2007) demonstrated that Earned Schedule metrics outperform, on average, more traditional Earned Value-based metrics. The researchers used innovative simulation techniques to establish the superiority of Earned Schedule for predictions of final duration.

The researchers then extended their simulation scenarios to cover specific schedule configurations. Although the initial study was widely accepted, the follow-up studies have been challenged. A summary of the controversy follows. For details, see my posts in January 2016.

In one follow-up study, Vanhoucke and Vandevoorde (2008) purported to test ES reliability for critical and non-critical activities. They concluded that discrepancies between performance of critical and non-critical activities undermine the reliability of ES duration forecasts.

Walt Lipke (2014a) disputed the claim. In Lipke’s view, Vanhoucke and Vandevoorde’s simulations were unrealistic. They failed to reflect the inter-dependency between critical and non-critical tasks. Furthermore, they ignored the fact that, at the end of a project, the SPIt converges to a value that ensures the EACt equals the real duration of the project.

In a second follow-up study, Vanhoucke and Vandevoorde (2009) examined the impact of project topology on the reliability of EVM metrics. The researchers used the term “project topology” as follows:

The topological structure of a network is defined by the distribution of the activities in the network and
the precedence relations between these activities. (Ibid, p 26)

A single topological indicator, serial vs. parallel activities (SP), was the focus of the research. The SP indicator measures where the project’s network lies on a scale between 100% serial activities and 100% parallel activities. Crucially, Vanhoucke and Vandevoorde tied SP to the critical/non-critical distinction of the first follow-up study:

[SP is] a measure for the amount of critical and non-critical activities in a network. (Ibid, p 27)

The researchers concluded that EVM metrics, including ES, do better in serial networks.

A more serial network contains, on average, more critical activities, which results in a better forecast accuracy compared to more parallel networks. (Ibid, p 29)

As shown in Diagrams 1a and 1b, ES duration forecasts suffered less than other EVM forecasts. But, according to their study, ES forecasts still suffered.

Because Vanhoucke and Vandevoorde tied SP to critical/non-critical scenarios, the second follow-up study is subject to the same objections raised to the first one. Again, the objections have been detailed in previous posts.

Still, for several reasons, the study was beneficial. First, it prompted further research on how project topology affects ES metrics. Walt Lipke (2014b) constructed scenarios that avoided the problems faced by Vanhoucke and Vandevoorde. Lipke tested his scenarios on 16 actual projects and found that ES metrics perform better as projects progress.

Diagram 2 shows the results from Lipke’s tests. The number of problems shrinks from a maximum of 40% at the start to 0% at the end of the projects in his study.

Diagram 2

Second, the simulations remain instructive about early parts of projects. Why? Because it is primarily later stages of the simulations that fail to reflect the convergence characteristic of ES. That’s where convergence should happen. Thus, evidence of forecasting problems early in projects remains relatively unimpeached.

For the same reason, Vanhoucke and Vandevoorde’s conclusion about serial projects (or serial parts of projects) remains intact: serial activities engender more accurate ES forecasts than do parallel ones. Lipke’s research bolsters the point. Diagram 2 shows a decline in discrepancies as the projects progress. Assuming that later stages of projects are more serial than early ones, it follows that ES forecasts become more reliable as schedules becomes more serial.

In conclusion, research shows that project topology affects the reliability of ES forecasts. They are more reliable later in projects when activities are more likely to be serial. An extension to ES leverages these insights and virtually eliminates ES reliability problems.