2

What is the best way to determine a weighted average for the actual time spent on historical tasks.

Let's say for example each project has a task A. And task A varies in time spent depending on certain project variables. I want to calculate and accurate average and use this to compare the task performance for each project. However, I don't want the average skewed by extreme data points such as one project taking 10 times too long for what ever reason.

  • I think you should take Standard deviation into account as well, Average alone doesn't give you many insights. – mamoo Oct 20 '15 at 13:06
2

I'm not sure in what context you are using this task average or using weighting but here is a starting point:

First you need to determine if you data set is normally distributed.

Here's a good resource:

https://statsthewayilikeit.files.wordpress.com/2014/07/tests-for-normality.pdf

Assuming your data set is normally distributed, the 3 most common "averages" will be your mean, median, or mode.

If you're not sure what these are, here's a resource:

http://www.purplemath.com/modules/meanmode.htm

Anyways, a data set that is normally distributed with very little skew will have a mean, median, and mode that are pretty close.

To identify/remove outliers, you may want to look at a histogram of your data or look at min/max values. the range, and 90 or 95% confidence intervals. Another way is to do random sampling of the entire population.

Lastly choose which "average" is most appropriate for your data set. Median and mode averages are more resistant to outliers.


If you determine your dataset is not normally distributed, its a good time to start searching the web about non-normal stats.

0

You have to understand the shape of the distribution for that task, which means you need a reasonable number of observations, i.e., historical data, so that the shape will develop. Then, graph it.

0

It's been a while since my last statistical classes, but I believe what you're looking for is to find out the Standard Deviation of your data.

It'll be pretty much the same as suggested by David, in a less graphical view.

Knowing your Standard Deviation, you'll be able to know what's the skewed data you have to discard on your calculation.

0

Based on my experience, average is not the best way to go, because of the extremes, you mentioned in your question. I'd rather go with David's response and Troy Magennis' work:

http://www.slideshare.net/FocusedObjective/lkce-cycle-time-analytics-and-forecasting-troy-magennis

Draw a distribution diagram with your historical data (see slide 15) and pick the time with is at the 85% of the graph. This point will give a most likely probably of finishing a task. Here is a short tutorial on how to do this for smaller and less complex project(s):

http://zsoltfabok.com/blog/2013/02/when-will-it-be-done/

0

The way I learned to do this is to use a "three-point" method that takes optimistic, pessimistic, and most likely historic values and provides an estimate and standard deviation. It turns out that Wikipedia has a straightforward write-up on the method.

This is also the method recommended in the 5th Edition PMI PMBOK (Section 6.5.2.4, Pg. 170).

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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