Page 1 of 1


Posted: Tue Dec 04, 2007 5:26 pm
by Jason
Do have any recommendations, what type of distribution should I use for duration? I have a number of activities ranged from 3 days to 2 month. I estimate uncertainties as low (around -10%), most likely and high (around + 10%).

Posted: Tue Dec 04, 2007 5:40 pm
by Intaver Support
We recommend using statistical distributions in RiskyProject in three cases:
1. If you have reliable historical data about task duration or cost
2. If you established eliciting judgment procedure where you asked experts questions about the their previous experience related to similar tasks.
3. To model the 'noise' or fluctuation of duration and cost which cannot be attributed to specific events.

In all other cases we recommend defining uncertainties using global and local risks.

Our analysis shows that in most real life project the type of distribution may not very critical. However there are a number of things to consider:

1. Distribution can be bounded and unbounded. All unbounded distributions in RiskyProject (normal and lognormal) have a cut offs.
2. If you have historical data, you may use Stat::Fit to generate the most appropriate distribution.
3. If you elicit judgment about distribution, you may use discrete or custom distribution.

In your case you may try beta distribution. It is easy to use bounded distribution.

Posted: Wed Aug 27, 2008 5:19 pm
by Tommi_
I'd like to define beta distribution for my task start time, but start low and start high fields for this task are grayed out. What should I do?

Posted: Wed Aug 27, 2008 5:21 pm
by Intaver Support
Most likely your task has predecessors. You cannot assign the distribution for start time for the task with predecessors.

recommendations for distributions

Posted: Wed May 19, 2010 9:11 pm
by Rankin
Hi, do you have any recommendations for which type of distributions we should use for duration. We have good data for the high and low, but not sure which distribution is best.

Distribution recommendations

Posted: Wed May 19, 2010 9:14 pm
by Intaver Support
If you are just starting out, triangular distributions will be more than adequate. Uncertainties caused by risks and uncertainties will be much more important in understanding the real course of your current baseline than the distribution.

Intaver Support

Posted: Wed Sep 29, 2010 12:22 pm
by RiskUser
I just would like to mention that I tried different distributions including normal, beta, lognormal. etc. Results are very close. I think that it makes sense to use different distribution if you have a reliable historical data. Then you can fit distribution and use it for cost and duration calculation. I used Stat::Fit to do it, because it is integrated with RiskyProject. Works very well.

Posted: Wed Sep 29, 2010 12:25 pm
by Intaver Support
Here is additional information about distribution fitting in RiskyProject

If you have empirical data for Cost, Income, and Duration, you may use Stat::Fit software to generate most appropriate statistical distribution. You cannot use Stat::Fit for start time. Stat::Fit is the third party software by Geer Mountain Software ( and can be purchased separately.

Fitting a distribution
1. Click on the Fit Distribution button. The Stat::Fit software opens.
2. Enter data to the Data Table. Make sure that you are entering data in the current units for RiskyProject e.g. days
3. Click on the Auto::Fit toolbar button
4. Select a recommended distribution from the list of possible distributions. Make sure that RiskyProject supports distribution you selected. RiskyProject can use only Uniform, Normal, Lognormal, Beta, Gamma, Triangular, and Exponential distributions from Stat::Fit.
5. When you are satisfied with selected distribution, click the Export button. From the list of applications, select RiskyProject. Output should be set to Clipboard (default selection).
6. Click OK.
7. Save empirical data in Stat::Fit if you which and close Stat::Fit. RiskyProject will generate distribution and assign low and high estimates for cost, income, and duration based on empirical data.