Entropy too high PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Jon Heron posted on Wednesday, August 05, 2015 - 11:24 am
Dear Bengt/ Linda

I have just been experimenting with the BCH three-step method.

I took a simple example with a single continuous outcome. I estimated within-class means using the auxiliary command and then repeated the same model after exporting the BCH weights following the method described in webnote-21.

This gives me the same estimated means (as of course it should) however in the second approach using training data I am quoted an entropy value of 1.501.

Is this anything to be concerned about? As I say, the model replicated with no other apparent problem.



best wishes, Jon
 Bengt O. Muthen posted on Thursday, August 06, 2015 - 1:31 pm
That sounds odd. Do you mind sending files to support so we can look into it?
 Jon Heron posted on Monday, August 10, 2015 - 6:48 am
Hi Bengt

in prepping a set of files to send to you, I thought as a quick check I would export "cprobs" from my third-step model which uses the BCH weights as training data.

I discovered that the cprobabilities are identical to these BCH weights. I think this explains why entropy is out of bounds and all my classification matrices are nonsensical.

Some BCH weights are negative - but this is fine as mentioned in Vermunt (2010). I obtain the correct answer from the model (correct = agrees with aux approach) but the classification summaries should probably be ignored.

This may be enough info for you, but I have the files here ready if you still need me to send.


best, Jon
 Bengt O. Muthen posted on Tuesday, August 11, 2015 - 12:40 am
Thanks.
 Bengt O. Muthen posted on Monday, August 17, 2015 - 6:15 pm
Jon - do you mind sending the files. Seems like we need them after all.
 Bengt O. Muthen posted on Tuesday, August 18, 2015 - 5:45 pm
The BCH weights use probabilities that are negative and >1 so the entropy is technically not even defined because of LOG of negative values. We will not print it in the next version for this case.
 Fred Danner posted on Tuesday, September 22, 2015 - 3:40 pm
I had exactly the same question (for a similar analysis that I had a colleague run for me with version 7.31)and am pleased to learn that I can ignore my entropy and classification probabilities > 1 output. However, I just purchased 7.31 and re-ran the same syntax but received an error message:
Unrecognized setting for SAVE option:
BCHWEIGHTS

is this because I did not first delete my version 7.1 when downloading 7.31?
 Linda K. Muthen posted on Tuesday, September 22, 2015 - 7:48 pm
Look at the top of the output to see whether 7.1 or 7.31 is being used. I suspect 7.1. You should not have more than one copy of Mplus.exe on your computer.
 Fred Danner posted on Tuesday, September 22, 2015 - 11:35 pm
That must be it -- I still have the old 7.1 Mplus.exe on my computer. I will uninstall everything and re-download. Thanks!
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: