« The DOI, Made to Slip? | Main | Personal Agility – Free Webinar »

January 07, 2008

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d834527c1469e200e54fd7b6d98834

Listed below are links to weblogs that reference Software Estimates - Managing Expectations via Ranges:

Comments

You point out that when estimates are needed, it is better to manage expectations with ranges than with point estimates. This is a very valuable point and this is what we at the SEI teach that higher maturity organizations should do! :-)

A couple minor comments (in case someone tries to implement exactly what you show):

a) typo in your formula for the variance: the last factor in the numerator should be "Expected - Pessimistic" rather than "Expected - Optimistic."

b) the 50% chance is associated with the Median value, not the Mean value. (The chance of achieving the Mean Value or less will only be (Expected - Optimistic)/(Pessimistic - Optimistic), which given that Expected is generally closer to Optimistic than Pessimistic, will be less than 1/2.)

Hi Mike,

Thanks for your comments and pointing out my typo. You are correct; I will get that fixed up.

Best regards
Mike

Also note that your calculations for variance and standard deviation are officially only correct for "normal" distributions, whereas we know that the time it takes to deliver a feature is not a normal distribution. This explains why almost all projects end up at least one, usually more, standard deviations away from the "mean" estimate to the bad side, and almost no projects end up to the left of the mean estimate. If the distribution were normal, as many projects would end early as end late.

Hi Nils,

Yes, it appears that way and I used to think this too. However, it was then explained to me that because we use a Mean value based on Optimistic, Expected and Pessimistic values with a triangular distribution our Mean is already skewed to the right. Using another triangular distribution or some other right skewed model would be double counting this shift. Whether this fully addresses the triangular distribution I’m not sure. I don’t pretend to be the originator of this approach, please refer to the paper I reference for a more complete coverage of the method.

Maths aside, I believe Don Reinertson’s explanation of why we get no unders (based on human personalities) is true for many projects.

Thanks for your comment.

Regards
Mike

The link to the background on the maths doesn't seem to work anymore. Any ideas where I can find tghe background info.

Neil

The comments to this entry are closed.

AddThis Social Bookmark Button

Your email address:


Powered by FeedBlitz

Tracker

  • Google Analytics
Blog powered by Typepad
Member since 07/2006

May 2014

Sun Mon Tue Wed Thu Fri Sat
        1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30 31