So I get the non-integer #successes in a binomial glm! warning, which has been asked about many times and I understand what it is. My dv is a % measure of accuracy, and these have weights so that they can be modeled using glmer(family=binomial). Sometimes however, people received fractional scores-- so although it is correct to think of them having 5 trials, they might have earned a 0.9 when the weights say it is 5, because they were awarded 4.5/5 as correct. Hence the non-integer successes.
Of course R only gives you a warning here, the models converge and the results look sensible. If I round up or down (to 4/5 or 5/5) to force the data to have only integer successes, the estimates change only slightly-- so it's clear to me that whatever R does with these non-integer successes, it isn't crazy.
However I'm now super worried about understanding WHAT it does with these non-integer successes, and why the warning is there if the estimates seem fine... do I need to be doing something else to model these data even though the results look valid?
r-sig-mixed-models
... $\endgroup$