davegb
Wed Oct 10 07:08:21 PDT 2007
On Oct 9, 10:12 pm, Jim Aksel <JimAk...@discussions.microsoft.com>
wrote:
> Jack - do you have a reference on this discredit?
> I know the 1-4-1 issue is sometimes challenged, but Project allows you to
> adjust the weighting. If you think about it, the requirement that it total 6
> is not restrictive, you can produce any three values you want and normalize
> it to 6.
>
> Monte Carlo seems to work better than 1-4-1 weighting, although it still
> requires a triple point estimate and assumptions on probability
> distributions which are also open to challenge. When using Monte Carlo I
> give myself the widest range reasonable possible with the Triangular
> distribution becuase of its larger standard deviation.
> --
> If this post was helpful, please consider rating it.
>
> Jim
>
> Visit
http://project.mvps.org/for FAQs and more information
> about Microsoft Project
>
>
>
> "Jack Dahlgren" wrote:
> > Not that I am aware of.
> > PERT has been discredited as being statistically valid so I wouldn't expect
> > that they will update this feature in the future.
> > You could conceivably do a monte carlo simulation of work. The results would
> > be somewhat interesting...
> > For some code to start with you can look at my free monte carlo simulation
> > macro here:
>
> >
http://masamiki.com/project/blackjack.htm
>
> > -Jack Dahlgren
>
> > "Andrew K" <Andr...@discussions.microsoft.com> wrote in message
> >news:96A770FA-9E4E-4096-A952-594BE1A7E2D7@microsoft.com...
> > > Is there a way to do a 3-point PERT analysis on Work estimates?- Hide quoted text -
>
> - Show quoted text -
I have looked for the reference I saw years ago showing that PERT is
statistically unsound. I can't find it. But the reasoning was that it
only accounts for the different probable durations of each individual
task, and ignored the cumulative risk along each path through the
project. The cumulative risk being far more significant than the
individual risk. So what he was saying is that PERT accounts for 10%
of the probable deviation and ignores 90% (percentages are made up,
don't know what they really might be, but you get the idea).
For example, one path through the project has 5 tasks. 4 have little
schedule risk, and when you do a PERT analysis, you find you're adding
a day or so the the 20 day duration of that path. Another path may
have 5 tasks, but all 5 have very high schedule risk, and you might,
by PERT analysis, add 5 days to the 20 day total duration. But since
all 5 tasks on this path have high SDs, the real risk of schedule
overrun on this path is much greater than on the other path, because
one task slipping early in this path can cause all the others to go
way over Expected duration, say double the 20 days. So, much more than
5 days should be added to it's probable duration to account for this.
PERT simply doesn't do this. Monte Carlo does.
The article I read was by a statistician who had studied PERT, had
very credible credentials, and made a lot of sense. I have steered my
clients/students away from it ever since. It's better than nothing,
but not much!
Hope this helps in your world.