The discussion of estimating - cost and schedule - has produced references usable in a variety of domains. As a background here's the guidance in the domain I work.

This "mandates" an Estimate At Completion with a Best Case, Worst Case, Most Likely. As well a Schedule Risk Analysis (SRA) is mandated.

And just for clarification for those favoring "hand built" 3 point estimates, the 3 points estimates stated in DID-81560 above are generated using the "ordinal range" method. The outcome from this processes are the percent ranges - upper and lower - of the probability distribution (Triangle) used for the Monte Carlo tool (Risk+ and @Risk for Project). So we have the three points, but they were not gathered by asking the engineers what they should be. The Systems Engineering, Control Account Managers, and Technical Leads, plus historical data, plus some statistical analysis produced the ordinal ranges for 5 classes of programmatic risk. That is then applied to each class of work, and then loaded into the baselined Integrated Master Schedule.

The point of all this is that producing credible estimates is part of the culture of the work we do. It's still just as hard, just as complex, and just as annoying as any other domain. But doing it, and doing it right is what is expected. It's part of being credible. So for those gathering variance ranges by hand, if the credibility - statistical confidence, auto-correlation, assessment of all the biases and influences discussed in The Psychology of Risk and Sources of Estimating Error can be addressed, then you're on the path to credibility.