In a manufacturing program I am collecting the estimated time required and the actual time taken for a series of tasks.

Suppose I find that I often take twice as long, say, as I originally estimated for each task, I want to apply a typical difference between estimated and actual time taken to future planned tasks to improve the accuracy of my planning.

Starting with no knowledge of stats, I'm not sure what to google or how to phrase the question - but so far I have thought a) determine a mean variance, positive and negative, for all past tasks between estimated and actual times and apply this value to future tasks to adjust the expected max and min times.

b) determine the standard deviation of past task timing, and apply forward.

Would appreciate any pointers- thanks.

  • $\begingroup$ One thing to consider before you start is whether your model of the estimation error might be relative (they tend to underestimate by 30% say), absolute (they tend to underestimate by a certain number of hours), linear (an offset and a percentage), or some nonlinear function, and how the variability in the estimates may tend to change with actual task time. $\endgroup$
    – Glen_b
    Jan 26, 2014 at 4:59

1 Answer 1


I think it depends on how the deviations are distributed from a statistical POV.

For instance, if deviations with respect to your estimated value follow a normal distribution, you can easily estimate the behaviour following the three sigma rule.

In the long run, you should obviously also think about switching your estimated value with the true mean of the process to center distribution around 0 (obviously if your distribution is simmetric) and testing against these hypotheses to be sure that the process follows the distribution you though, about this you can read about the simple mean analysis.

Hope this helps


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.