Zero probabilities in Proc FMM

[jslejko]

Hi,

I am using Proc FMM to estimate a mixture a two normal distributions.  The parameter estimates match what I have found previously using other software, however, I've run into a problem when examining the mixing and posterior probabilities.  I notice that a vast number of observations have zero values for both component mixing and posterior probabilities.  Not 'missings', but actually zero.  These results seem strange because if one probability were zero, the other should be 1, and in this case they are both zero.  Does anyone have any insight about why this is happening and what it means?  I noticed that a similar problem was documented with a hot fix, but my software appears to be completely up to date so the hot fix was not installed.

Thanks,

Julia

[SteveDenham]

Hi Julia,

Install the hotfix, just in case, and see if the problem persists.  If it does, open a ticket with Tech Support, as this is NOT expected behavior.

Steve Denham

[jslejko]

Thanks for your message!  I tried installing the hotfix and got this message:

"SAS Threaded Kernel Extensions for Statistics 12.1 J15001 English

The maintenance package you are trying to install requires a maintenance version between 0 and 1, which could not be located on your computer

---

No changes need to be made. Your system is up to date"

I will try FMM again and open a ticket if the problem persists.

Thanks again,

Julia