I'm experiencing some trouble when using the polr function.
Here is a subset of the data I have:
# response variable rep = factor(c(0.00, 0.04, 0.06, 0.13, 0.15, 0.05, 0.07, 0.00, 0.06, 0.04, 0.05, 0.00, 0.92, 0.95, 0.95, 1, 0.97, 0.06, 0.06, 0.03, 0.03, 0.08, 0.07, 0.04, 0.08, 0.03, 0.07, 0.05, 0.05, 0.06, 0.04, 0.04, 0.08, 0.04, 0.04, 0.04, 0.97, 0.03, 0.04, 0.02, 0.04, 0.01, 0.06, 0.06, 0.07, 0.08, 0.05, 0.03, 0.06,0.03)) # "rep" is discrete variable which represents proportion so that it varies between 0 and 1 # It is discrete proportions because it is the proportion of TRUE over a finite list of TRUE/FALSE. example: if the list has 3 arguments, the proportions value can only be 0,1/3,2/3 or 1 # predicted variable set.seed(10) pred.1 = sample(x=rep(1:5,10),size=50) pred.2 = sample(x=rep(c('a','b','c','d','e'),10),size=50) # "pred" are discrete variables # polr polr(rep~pred.1+pred.2)
The subset I gave you works fine ! But my entire data set and some subset of it does not work ! And I can't find anything in my data that differ from this subset except the quantity. So, here is my question: Is there any limitations in terms of the number of levels for example that would yield to the following error message:
Error in optim(s0, fmin, gmin, method = "BFGS", ...) : the initial value in 'vmin' is not finite
and the notification message:
glm.fit: fitted probabilities numerically 0 or 1 occurred
(I had to translate these two messages into english so they might no be 100% correct)
I sometimes only get the notification message and sometimes everything is fine depending on the what subset of my data I use.
My rep variable have a total of 101 levels for information (and contain nothing else than the kind of data I described)
So it is a terrible question that I am asking becaue I can't give you my full dataset and I don't know where is the problem. Can you guess where my problem comes from thanks to these informations ?