In eval(family$initialize, rho) : non-integer #successes in a binomial glm! warning because of survey weights

by

I am running glmer() in R.

Here is my empty model

empty <- glmer(DV ~ 
                      +(1|year), 
                    weights = weights,
                    
                    data= df,
                    family= binomial)
Warning message:
In eval(family$initialize, rho) : non-integer #successes in a binomial glm!

I got the warning message of a non-integer when using survey weights. This warning disappears when I remove the survey weights. My guess is that this warning is because the weights are numeric, not integers.

However, I don’t want to exclude weights in my model since AIC and BIC get much better with survey weights.
Is it OK to ignore the warning? OR are there any methods to fix the warning?

Thanks in advance.

  • Why not multiply your weights by, say, 1e6 and convert to integer. This will keep the weights correct to 6 figures. Mathematically, the absolute values of the weights don’t matter – only their relative size

    – 

  • Thanks. I tried it but got another error message…. “Error: (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate”

    – 

Leave a Comment