I was chasing some links the other day and came upon this one, a presentation given in 2004 to the Society of Cost Estimating and Analysis by Stephen A. Book. His presentation, entitled "How to Make Your Point Estimate Look Like a Cost-Risk Analysis (so It Can be Used for Decisionmaking) is eye-catching to begin with, in spite of misspelling decision making throughout.
His main theme is single point estimates for cost [and presumably all else] are bad practice from the git-go, and almost always lead to bad conclusions by decision makers. Ever since Booz-Allen invented PERT for the Navy in the '50's we've been hearing this message--and it's still as true today as it was then.
Book makes three big points that I have taken from his slides:
1. Sum of WBS-Element Most Likely Costs is NOT Most Likely Total Cost
2. “Point” Estimate Not Useful for Decision making, Because No Success Probability Can be Associated with It
3. Only Percentiles Are Meaningful to Decision Makers for Budgeting and Program-Control Purposes
You can't really argue with point #1 because 'most likely costs' are either the "mode" of the distribution or a calculated "moment" of the distribution. Either way, addition between distributions is not just a matter of arithmetic; addition has to take into account probability and is therefore handled with a mathematical process called 'convolution'.
You shouldn't argue with point #2 because any decider worth [their] pay considers success when making a decision; if not, fire them and get someone who knows how to make decisions.
You might argue with point #3 because 'only' is a pretty strong word. Many decision makers can work with statistical information that is not reduced to percentiles.
No matter how you make the argument, Book's presentation is worth a read!
