What is Wideband Delphi Estimation?

I just wrapped up a three day training class on Software Testing taught by Dale Emery of Quality Tree Software. Dale was the IT Process Improvement Manager for Sun Microsystems from 1999 to 2002. His insights as a developer made the class an awesome opportunity to pick up new tools for the tool box.

One of the more interesting concepts he discussed was the Wideband Delphi Estimation method. WDE is a slight variant of the original Delphi Estimation method developed by the RAND Corporation at the beginning of the cold war to forecast the impact of technology on warfare.

The original method works like this. A set of domain experts are sent a request for an estimate on a specific subject. For example, I send a request asking developers for estimates on how long it would take them to develop a web based message board from scratch.

Each response is then anonymized and added to a chart showing the variability of the estimations. The chart can then be sent out to the estimators and they are offered a chance to revise.

The median value of the estimates is then selected as the final estimate. The median value has a unique nature because it represents a better estimate than 50% of the other estimates regardless of the actual outcome.

An added benefit is that people with strong personalities are not allowed to dominate the estimation process (or weaker personalities for that matter) and skew the results. Each person gets a balanced representation in the estimating process.

Wideband Delphi Estimation differs in that it expands the amount of information being exchanged by the estimators. Estimators provided their considerations as well, which are then presented back to the group. The group can then review the considerations of their peers and adjust their estimations accordingly.

An interesting concept that I stored away in my toolbox. I hope I can drag that one out in the future and find a use for it. Thanks, Dale.