How does MS Project determine PERT pessimistic and optimistic?

[rhino]

I suspect Microsoft Project is NOT doing a full PERT to beta
distribution and then applying the central limit theorem.

It does seem to be using the three durations (optimistic + 4x expected
+ pessimistic all over 6) to determine a weighted average for each
task, and then it adds up all those tasks on the critical path to
determine the nominal value for the whole project.

But when it works out the pessimistic (optimistic) value for the whole
project it seems to use ALL the pessimistic (optimistic) values, which
lacks statistical robustness. The correct way to do it would be to
determine the variance of the whole project, and then use the normal
distribution to determine the 90% confidence interval for total project
duration. I can see no evidence that MS Project is actually doing this.
Can anyone else put some light on this topic?

 

[Gérard Ducouret]

Hello Rhino,
Yes, you are right. MS Project "PERT Analysis" is working as you said.
For a more powerfull PERT Analysis, look at an add-on such as @RISK from
Palissade, or some others...

Gérard Ducouret

 

[davegb]

I'm not a statistician, but from reading about this, I understand that
PERT analysis is outdated and basically misleading at best, erroneous
at worst, because it doesn' t take into account the cumulative risk
along a given path, just the individual risk of each task. If you're
going to get real, use Monte Carlo analysis. There are a number of
tools out there that do this.

 

[Steve House [MVP - MS Project]]

Not only that, the pessimistic and optimistic durations are assumed to
represent +/- 3 standard deviations which means you have only about 1%
probability of hitting either one. Odds are better at the casino.

 

[davegb]

Maybe I misunderstood the concept, but isn't it the range in between
the pessimistic and optimistic values that matters, not just the
probabilities of hitting the actual values? Which, being 6 sigma, is
well over 99%, if my memory serves me (which it doesn't always).

 

[Steve House [Project MVP]]

You're right, was just using round ballpark figures when I said 99%. By
"hitting" I meant finding your actual results laying at or outside the +/- 3
sigma points. The PERT tools in Project do give you the 6 sigma range, +/- 3
sigma away from the median on both the optimistic and pessimistic side and
those are the numbers that the optimistic and pessimistic Gantts will show.
The problem is I think way to many managers think the 3 sigma point on the
optimistic side especially is actually doable "if you just put your mind to
it", to which I say "about as much chance as a snowball in ...". There's
about a 99% chance that you'll lie inside the range which means of course
there's only a 1% chance of being outside it. And since that 1% is the sum
of both the top and bottom, you've really only got about 1 chance in
something like five hundred to a thousand of actually hitting or doing
better than the optimistic (or doing worse than the pessimistic, but no one
tries to hit that <grin>).