Skip to content
Prev 228219 / 398500 Next

Servreg $loglik

On Jul 20, 2010, at 11:20 AM, Charles Annis, P.E. wrote:

            
This has come up on r-help many times before (and probably on other  
lists as well), despite not being an R question at all. It is  
commonplace in modeling grouped data to see likelihoods reported  
differently from the result obtained when modeling ungrouped data  
representations with the same frequencies. The only valid statistical  
process is to compare differences in the likelihoods (or log(L) ),  
since the likelihood (or log(L) ) is only defined up to an arbitrary  
constant. You need to be comparing the result to some sort of "null  
model" for it to have any meaning. (... or perhaps that is your null  
model and you need to be looking at the impact of adding a covariate  
or two.)