Nifty. For a while I was working on a commercial drag-and-drop tool for PERT-style estimation[0].
It foundered on the rocks of Angular. But I did enjoy doing research on PERT and estimation generally.
My suspicion is that PERT works more through the "unpacking effect" than through the statistical stuff. PERT sorta-kinda resembles a simplistic normal distribution, which is nonsensical in a lot of ways.
Later operations research got way, way more sophisticated. But it is my suspicion that stuff like stochastic project networks have produced more PhDs than accessible estimates.
The unpacking effect is more compelling, because it's a way to offset the planning fallacy that humans are prone to. Make people break down a task before they estimate, and they will give a higher estimate. This is because humans systematically forget to include lots of tasks.
In software engineering, for example, off-the-cuff estimates typically don't include stuff like time for documentation, waiting for CI and so on. We subconsciously over-weight the tasks that we understand best: the coding.
It foundered on the rocks of Angular. But I did enjoy doing research on PERT and estimation generally.
My suspicion is that PERT works more through the "unpacking effect" than through the statistical stuff. PERT sorta-kinda resembles a simplistic normal distribution, which is nonsensical in a lot of ways.
Later operations research got way, way more sophisticated. But it is my suspicion that stuff like stochastic project networks have produced more PhDs than accessible estimates.
The unpacking effect is more compelling, because it's a way to offset the planning fallacy that humans are prone to. Make people break down a task before they estimate, and they will give a higher estimate. This is because humans systematically forget to include lots of tasks.
In software engineering, for example, off-the-cuff estimates typically don't include stuff like time for documentation, waiting for CI and so on. We subconsciously over-weight the tasks that we understand best: the coding.
[0] http://confidest.com/