In yesterday’s blog post, I explained that Statistical PERT’s definition of “near certainty” in a most likely outcome meant that the standard deviation for that still-uncertain uncertainty was 7% of the range between the maximum and minimum point-estimates. (Confused already? Read yesterday’s blog post firstly, then).
But someone might say, “My definition of a ‘nearly certain’ most likely outcome is more stringent than that. I wouldn’t say out of 100 trials, 98 outcomes are equal to my most likely point-estimate means that my most likely outcome is nearly certain to occur.”
That’s fine; then change the meaning of what “Near Certainty” means in the SPERT template!
It’s easy to do. Model your definition of “Near Certainty” by modifying the Ratio Scale Modeler worksheet in the SPERT template, or just create a blank spreadsheet to do the modeling.
Start the same way as before. Enter a minimum point-estimate in cell A1, followed by a maximum point-estimate in cell A2. Then, copy your most likely outcome into cells A3:An, where n = whatever number of hypothetical trials you want to model so you can show what ‘near certainty’ means to you. Copy down cell A3 to cell A200, or cell A300 or A500 or A1000. Whatever you want!
When you’re done copying cells in column A, follow the same process I described in the previous post. Use STDEV.P to find the standard deviation for the distribution you’re modeling. Then, divide that standard deviation by the number of hypothetical trials you used. The result is a SPERT Ratio Scale Multiplier (RSM) to be used with a subjective opinion that says the most likely outcome is nearly certain to occur.
If you’re using a perfectly-shaped bell-curve, where only one trial is less than the most likely outcome and one trial is greater than the most likely outcome, you should find that the following:
- 1-198-1 creates an RSM of about 5%
- 1-298-1 creates an RSM of about 4%
- 1-498-1 creates an RSM of about 3%
- 1-998-1 creates an RSM of about 2%
Yes, the actual values of your three-point estimate will affect the RSM values, but you’ll find that the resulting RSM’s don’t change too much, in spite of the three-point estimate you’re modeling. Knowing that, the RSM’s can be used for a number of other, different, three-point estimates within the SPERT estimation process.
If you model a skewed three-point estimate, you’ll find that the RSM’s will be bigger than what I’m showing above, and that makes sense, since there is greater uncertainty surrounding an asymmetrical bell-shaped curve than a symmetrical bell-shape.
You don’t have to be constrained by the default presumption that a ‘nearly certain’ most likely outcome means a 1-98-1 split of 100 hypothetical trials for an uncertainty. You can define for yourself what ‘near certainty’ means to you….and to others.