On the first worksheet of the SPERT-Beta download (for version 0.2, Build 2), I’ve added a worksheet that does a comparison of nine different, right-skewed uncertainties. The skewing ranges from mild to severe. What I wanted to learn is how this version of SPERT-Beta compares with a Monte Carlo simulation of Palisade’s @Risk7 using the RiskPERT function in that add-in program.
I compared both SPERT-Beta and the SPERT-Normal edition (which uses the normal distribution), which is the original edition of Statistical PERT, already released and available for download and use.
I used two different subjective opinions within the SPERT worksheets. For SPERT-Beta, I chose High Confidence and Medium-High Confidence in the most likely outcome. For SPERT-Normal, I chose Medium and Medium-High Confidence in the most likely outcome. These subjective choices approximate what a Monte Carlo simulation would give using the RiskPERT function in @Risk7.
You wouldn’t expect SPERT-Normal to give very similar results to RiskPERT since SPERT-Normal uses a symmetrical curve to model a asymmetrical uncertainties. And yet, the results, even for severely skewed uncertainties, were not too far from the RiskPERT results.
I don’t recommend using SPERT-Normal for estimating uncertainties that are more than moderately skewed (to the left or right). Using SPERT-Beta will give more accurate results in those cases because SPERT-Beta can build an implicit, skewed curve, whereas SPERT-Normal can’t.