PERT uses a special form of the beta distribution but Statistical PERT (SPERT®) Normal Edition uses the normal distribution because, for many estimation problems, using the normal distribution is good enough to make a good decision.
As long as the bell-shaped curve is only slightly or moderately skewed, Statistical PERT using Excel’s normal distribution functions will still yield results that are comparable to a Monte Carlo simulation of the same uncertainty modeled using the beta distribution. Visit this blog post to learn more.
The key benefit of using the normal distribution is that Excel’s two normal functions, NORM.DIST and NORM.INV, are both useful and very easy to use. PERT provides an estimate for the mean argument, and Statistical PERT creates the required standard deviation needed by those two Excel functions.
The goal of any estimation effort should be to develop an estimate that is good enough to make a good decision, and that is accurate enough to be within the estimator’s tolerance for error. While using the beta distribution may be more accurate than using the normal distribution for modeling asymmetrical uncertainties, that fact alone is not a good reason to discard using the normal distribution. For many estimation problems–even involving asymmetrical uncertainties–using the normal distribution leads to a “good enough” estimate that is within the estimator’s tolerance for error.
That said, on March 1, 2017, a new edition of Statistical PERT was released that uses Excel’s two beta functions, BETA.DIST and BETA.INV. The Statistical PERT Beta Edition will accurately model skewed probabilities using the beta distribution’s two shape arguments, alpha and beta. You can visit this site’s Beta Blog to learn more about this new edition of Statistical PERT, and download Version 1 of a SPERT Beta spreadsheet.
Statistical PERT uses a different formula than PERT does to find a standard deviation. Why?
Statistical PERT replaces the PERT formula with a different formula that lets the estimator use subjective judgment, intuition, emotion, and/or private knowledge to rationally adjust the standard deviation. The SPERT standard deviation formula is: (Max – Min) * RSM, where RSM is the Ratio Scale Multiplier that corresponds to the estimator’s subjective opinion about how likely is the most likely outcome. The SPERT SD formula better conforms the implied distribution curve to a specific uncertainty. You can learn more about this here.
(Versions 1.3 and earlier) The SPERT template shows both Unskewed and Skewed probabilities. Which do I use?
Usually, you should rely on the Skewed probabilities. Statistical PERT uses the normal distribution which implies bell-curve symmetry. When a bell-curve is symmetrical, the median, mean and mode (most likely outcome) all have the same value. When a risk is skewed, though, the mean and mode are different. Excel’s two normal distribution functions, NORM.INV and NORM.DIST, use the mean as a required argument, but you can use the mode as a surrogate for the mean if you have a very skewed uncertainty AND you have very high confidence in how likely the most likely outcome really is. In that particular case only, using the Unskewed probabilities will be better than using the Skewed probabilities. In 2017, Statistical PERT-Beta Edition will be released, and this new edition of Statistical PERT will allow you to better model skewed uncertainties. Want to learn more? Check out this blog post.