Multiple comparisons in aster analysis and warning message with random effects models

[CassandraM]

Hello everybody,

 

I wonder if it is possible to apply post hoc tests (or another test for multiple comparisons) with aster analysis? In my models, one factor (treatment) have three different levels. At this stage, I repeated the same model by removing one of the three levels to make all the three possible comparisons but I am not sure this is the best option…

 

In another context, when I applied a model on the same dataset with a random effect (population that is nested within the treatment), I obtained the following warning message:

     “estimated Fisher information matrix not positive definite, making all standard errors infinite”

Can you explain me this warning message please and eventually how can I fix it?

Thank you very much for your help,

 

Cassandra

[geyer]

Ugh!  That error comes when the precison of computer arithmetic (about 16 decimal places) isn't enough to get accurate computation.  The issue is mentioned on the slides for the course on aster models and this is an unfortunate property of random effects (no matter what methodology is used to fit the model) as is explained later in the same deck of slides.

But that does not help you at all.  You have the data you have, and you want to fit the models you want to fit, those that address the questions of scientific interest.

So I apologize for not having a way around this, but no one knows if there is a way around this.

In short, that error message says the computer cannot handle that model.  As far as it can tell, there is no information in the data about the variance components in that model.  We have estimates, but the confidence intervals are zero to infinity (that is, any possible value).  So the estimates don't mean anything.

[CassandraM]

Hello, 

Thank you very much for your answer.

I wonder if you have any other advice about multiple comparison tests ? In my dataset, one of the main factors (the treatment in which mite populations evolved) have three levels. This factor has a significant effect on lifetime reproductive success and I would like to know which levels are different from each other.

Thank you in advance for your help, 

Cassandra