Loading ...
Sorry, an error occurred while loading the content.

112404Re: I'm just _so_ Bad. Estimating.

Expand Messages
  • Keith Braithwaite
    Sep 13, 2005
      --- In extremeprogramming@yahoogroups.com, "Robert C. Martin" <UncleBob@o...>
      wrote:
      > > -----Original Message-----
      > > From: John Carter [mailto:john.carter@t...]
      > >
      > > e).. Suggestions please.
      >
      >
      > 3. Statistical estimate (PERT). Go ahead and esimate in man-hours.
      > However, give every task *three* estimates. The best case, the worse case,
      > and the most likely case. For each task compute the mean and variance.
      > Mean = (B+W+4L)/6. Variance = ((W-B)/6)^2. Now add up all the means and
      > variances so that you can compute the mean and variance for the whole
      > project. (See: www.objectmentor.com/resources/articles/PertCpmAgile)

      Interesting article, Bob. Have you seen http://www.cadmus.ca/bookreviewpolaris.htm ?

      It's a review of a review of the Polaris programme, which includes the comment that "PERT
      was less effective than advertised but more so than rain dancing."

      PERT was clever, complicated and distracting. Polaris succeeded for all kinds of interesting
      reasons (the report suggests) of which PERT is the least significant. And it's greatest
      significance may in fact have been as a big heavy stick for the Navy project managers to
      beat their subcontractors with.

      Meanwhile, we use the (B+W+4L)/6 formula in each round of our team estimating process,
      within a Delphi framework. We look at the variances not so much to get a variance fort he
      overall estimate, but to tell us when we don't need another Delphi round.

      John, I can't suggest strongly enough that (your other issues with time management not
      withstanding :) when it comes to generating numerical estimates you use some variation
      of Delphi, see http://www.rand.org/publications/RM/RM5888/

      This works particularly well for us, since we estimate and commit as as team, but the
      mechanics of Delphi can be usefully applied to an individual providing estimates for their
      own individual work.

      It's worth noting that the PERT formulae generate exactly the substantial, structured
      estimates that you'd need to start applying Theory of Constraints ideas to planning (such
      as Critical Chain), somehting we're just starting to explore.

      Keith
    • Show all 16 messages in this topic