WARNING: THE BEST LOGLIKELIHOOD VALUE FOR THE 3 CLASS MODEL FOR THE GENERATED DATA WAS NOT REPLICATED IN 81 OUT OF 100 BOOTSTRAP DRAWS. THE P-VALUE MAY NOT BE TRUSTWORTHY DUE TO LOCAL MAXIMA. INCREASE THE NUMBER OF RANDOM STARTS USING THE LRTSTARTS OPTION.
Where 3 classes is the K-1 class model. In running the analysis without bootstrap and STARTS 40 8 the 3 (k-1) class model output looks ok (the best likelihood is replicated 8 times) - so what is going wrong in this bootstrap?
Also in all of the documentation I have seen, it is advised to use 0 0 in LRTSTARTS for the k-1 model - why is this the case? and what is going wrong in my case?
I have consulted webnote 14 and followed the procedure for determining number of lrstarts for the k class model and for determining the OPTSEED.
I am confused because the warning I get refers to the k-1 generated data and according to webnote 14 the generated data are generated according to the k-1 model, so what could be going wrong for k-1 likelihoods in the generated data to not be replicated?
There is a typo in that error message. It should have been
WARNING: THE BEST LOGLIKELIHOOD VALUE FOR THE 4 CLASS MODEL FOR THE GENERATED DATA WAS NOT REPLICATED IN 81 OUT OF 100 BOOTSTRAP DRAWS. THE P-VALUE MAY NOT BE TRUSTWORTHY DUE TO LOCAL MAXIMA. INCREASE THE NUMBER OF RANDOM STARTS USING THE LRTSTARTS OPTION.
In conducting growth mixture modeling, I got the following Tech 14 warning after specifying OPTSEED that is associated with the best loglikelihood value from previous analysis (K-1).
THE LIKELIHOOD RATIO TEST COMPUTATION FOR TECH14 DID NOT TERMINATE NORMALLY BECAUSE THE LOGLIKELIHOOD VALUE FOR THE MODEL WITH ONELESS CLASS IS LARGER THAN THE LOGLIKELIHOOD VALUE FOR THE ESTIMATED MODEL.INCREASE THE NUMBER OF RANDOM STARTS USING THE LRTSTARTS OPTION.
How should I approach this? I increased LRTSTARTS to 0 0 150 40.
RuoShui posted on Wednesday, May 07, 2014 - 1:17 am
Thank you very much! I have read Web Note 14. But I still had the warning when I used the first OPTSEED that is associated with the best loglikelihood value from K-1 analysis. I read from the discussion that you said any of the OPTSEED associated with the best loglikelihood value can be used. When I used the second OPTSEED, there was no warning. Is this normal? However, there are still a few questions.
1) I compared the model in which I used OPTSEED and Starts=0 to the model in which I did not use OPTSEED and used STARTS and LRTSTARTS. The shapes of the classes are different between the two models? What does this mean?
2) If I need to set a high LRTSTARTS to 0 0 2500 20 for the best likelihood to be replicated for BLRT; does this mean that my data does not actually support a K-class but a K-1 solution?