Over the past months, there has been several posts about the mathematics of project management. Cost estimating, schedule estimating, programmatic and technical risks.
I was cleaning up my server, looking for materials to aide a client for "getting back to green," on several programs. I came across Earned Value Management for Cost Estimators.
Download this material and pay special attention to the statistical information starting around page 109. The naive approach of Gaussian sums, where the Most Likely values of estimates are summed to produce a "normal" distribution, depends on independence between all probability distributions. Then move on to page 162 to see how correlations are critical to estimating cost and schedule impacts.
It turns out of course there are rarely if ever real projects that have no correlations between work activities. With this knowledge, it is now clear, that the simple approaches to estimating - 3 points, or single points, PERT, and all the other non-statistically developed estimates are simply too simple for use on any non-trivial project.
