Earned Schedule results versus Critical Path and Finish Variance results

Question

I am monitoring a schedule right now for a project that is about 57% complete (duration). Looking at it at two different perspectives--Earned Schedule and Critical Path--I am finding results that differ quite substantially. I know no predictive model is 100% valid and no two models will produce the exact same results but I was surprised by the degree of difference.

The schedule is a simple water fall, duration-based schedule that was constructed reasonably well, in that it is a constraint free schedule with no orphaned work packages and most of those work packages are decomposed where no duration is greater than two reporting periods. The way progress has developed, the remaining work packages are all on the critical path and the schedule is showing a five day unfavorable finish variance.

Using Earned Schedule, the project is showing over 10 business days late, close to 12, and using the estimate at completion formula the project is likely to be over a month late.

Work is probabilistic so I know that the five days late (CP) or the over a month late (ES) are two points within a range of likely results but I am surprised at the size of the variance between two predictions.

Has anyone seen this type of a discrepancy between two models before? Could this simply be due to the sensitivity of the models and, if so, which model do you believe is more valid / accurate?

Thanks for the help!

Answer 1

Generally speaking, I've found more complex models to be more difficult to "get right". Unless you are willing to spend the time to customize a complex model, the fact that more variables are being used or manipulated the greater the opportunity for one or more of your underlying assumptions to be inaccurate. For this reason I tend to work with critical path estimates.

Another approach could be to pool the different models using a weighted average like PERT, assuming that the mid-point estimate is the most likely case. This is imperfect but at least it helps to mitigate the effects of the model(s) that are "wrong". Based on the data given that would give you a final variance a bit shy of 20 days.

A more rigorous approach is to evaluate your models based on what was known for each reporting period starting at project start. Figure out what each model would predict and compare to actual variance to date, and go with the one that seems to have the greatest predictive power. You may also be able to identify particular project events that have caused particular models to be less predictive.