Monthly Archives: September 2016

What’s Different About the Beta Edition of Statistical PERT?

The Beta Edition of Statistical PERT uses the beta probability functions inside of Microsoft Excel.  Specifically, the Beta Edition of SPERT uses BETA.DIST (beta distribution) and BETA.INV (beta inverse).

The key advantage of using the beta distribution over the Normal Edition of Statistical PERT is that, using the beta distribution, you can more accurately model skewed bell-shaped uncertainties.  With the Normal Edition of SPERT, you can still model bell-shaped uncertainties that are skewed, but only to a certain point before the resulting probabilistic estimates are not reliable.  For example, when the range between the minimum point-estimate and the most likely point-estimate (the mode) is half or one-third or even one-fourth the range between the most likely point-estimate and the maximum point-estimate, the resulting probabilities are still pretty accurate within a few percentage points.  (This assumes that the standard of truth is using the PERT beta distribution as found in Palisade’s @Risk RiskPERT function — which is a key assumption that may or may not be true for a given estimate).

With the Beta Edition of Statistical PERT, you can model many skewed, bell-shaped uncertainties even to the point where you are dealing with a triangular distribution on either the left-side or the right-side of the curve.

If you download the development build of the Beta Edition of SPERT, play around with the ‘SPERT Beta Curve’ worksheet.  By changing the values for the Minimum, Most Likely and Maximum point-estimates (cells B4, C4 and D4) and also the Most Likely Confidence column (cell J4), you’ll see different representations of the bell-shaped curve on the chart that’s in that worksheet.

The Beta Edition of SPERT uses 150 combinations of the two key parameters required by the BETA.DIST function:  alpha and beta.  Each combination has been carefully calibrated so the resulting bell-shaped curve still retains the mode as found in the Most Likely point-estimate in cell C4.  The mode may not be exactly, precisely the same in the bell-shaped curve, but it will be very close — close enough that it should be fine for most estimation scenarios.

Try using the Beta Edition of SPERT!  Just be aware that this is still a development build.

New SPERT-Beta Release for September 2016

Here is another new release of Statistical PERT – Beta Edition version 0.2 Build 5.  This is a development build, meaning this is still being developed before its first production release.  I expect the initial production release to be in early 2017.

In Build 5, I’ve added several more subjective terms to describe the most likely outcome.  Now, there are 10 subjective terms from which to choose:

  • Near certainty
  • Extremely high confidence
  • Very high confidence
  • High confidence
  • Medium-high confidence
  • Medium confidence
  • Medium-low confidence
  • Low confidence
  • Very Low confidence
  • Guesstimate

I’ve better aligned the results of this Beta Edition with the Normal Edition of Statistical PERT (which uses the normal distribution), meaning, that Medium or Medium-High confidences in the most likely outcome will be approximately the same for symmetrical bell-curves between the Normal Edition and Beta Edition of Statistical PERT.

Also in Build 5, I’ve added the ability to obtain probabilistic estimates from a triangular (or nearly triangular) distribution.

Instead of continuing to offer download files from this blog, all download will be done from the main download area of the Statistical PERT website.  This will allow me to track how many times different versions of Statistical PERT are downloaded.

Feel free to contact me with any questions.  As I get through the development cycle for this new edition, I’ll begin writing more about how this edition utilizes the beta distribution.