• Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

Comments (1)

1 VascoDrecun commented Permalink

Murray, I agree that some statistics must be used to guide the worker through the task estimation. However, I am not a big fan of task classification and parametrization techniques that base estimates on disecting the work being done. I like the statistics based on the worker providing the estimate. In other words, we need to measure the estimation accuracy of the worker itself, or a combination of the worker accuracy and type of work. Thus, we know what is the likeliness that the worker guesses the range right for certain type of work (simple classification), and then what side of the distribution is he skewed against (min or max, e.g. is he/she optimistic or pessimistic estimator).

That way we avoid overengineering the estimation algorithm (let the worker be the objective/subjective black box) and only measure how many times was the black box right or wrong and by how much in each direction. We can also compare black boxes against each other in terms of their level of accuracy. Sometimes you may use several estimators, each with different statistical profile, and apply some sort of adjusted Delphi. That is what the statistics is best in - outcomes. You remember what troubles we had with regression models of estimations with projects containing novelty and environmental uncertainty. Simply, don't measure the task, measure the estimator.
Hope I am helping here...

Add a Comment Add a Comment