Compare models in cafe5

[Anezka Santolikova]

Hi,

I apologize for maybe-stupid question, I am not very experienced in statistics and have problem understanding how to decide which model to use in cafe5.

I have now run cafe5 using several models: single lambda for whole tree, three lambdas for tree and five lambdas/tree.

I somehow can't understand how to now evaluate if to use the simple one-lambda model or a more complex model. From Questions and answers posted here, I understand I need to install cafe version4 to run the likelihood-ratio test in cafe.

I do not understand why / if there is the need to perform simmulations for comparing models or if there is a way I can use the likelihoods (or the -lnL to be specific) of the models that I let run previously for the test.

Thank you in advance for any reply or clarification.

Anezka

[Hahn, Matthew]

Hi Anezka,

The problem is that the distribution of log-likelihoods may not be chi-squared distributed, so we recommend doing simulations. You can also do these in CAFE5, though you just have to run the simulations yourself. This is a better option for comparing a 3-p model to a 5-p model anyway (CAFE4 could not do this).

cheers,

Matt

--
You received this message because you are subscribed to the Google Groups "hahnlab-cafe" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hahnlabcafe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hahnlabcafe/29d94aee-152d-4a58-8c6d-50580f83726an%40googlegroups.com.