I am reasonably certain I would also have included the abs(), as to make certain overshoot estimates count as much as undercut ones.
I would divide by time spent, not time estimated - but that would make such a calculation only possible when task is finished, whereas given formula allows it to be used for work in progress.

If I had more than a few estimate/result tuples for each worker, I'd try to build a better forecasting equation by calculating arithmetic mean and variance of those couples for each worker, perhaps try to fit a linear correlation on est and act. To lazy at the moment to think through if the squaring step of estimating variances for correlation is enough to eliminate the abs() mentioned before. Pretty certain that degrees of freedom will not approximate Gaussian Bell, but Student T distribution should be good enough.



>A follow-up....and because I respect you as the "just the facts, Sgt Joe Friday, no nonsense", then maybe you can make sense of this.
>
>I found how someone is calculating this....again, for a bad sprint of 60 hours on an original guess of 33.5 hours, I figured the accuracy was 55.83%. But I've got a project manager saying it's about 21%. Here is how the PM is calcing it...this seems more like a factor than an accuracy percentage.
>
>1 - abs ( amount estimated - remaining work left - time spent)
> -----------------------------------------------------------------------------
> amount estimated
>
>
>so that....
>

>
>1 -   abs (  33.5  - 0 - 60)
>       ----------------------------
>              33.5
>
>1 -  ( 26.5)
>      ----------  
>        33.5
>
>1 -  .791
>
>
>accuracy:   21%
>
>