CODEWITHSUNDEEP
rlme4
Di Zeng
Di Zeng

Reputation: 3

Error: (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate in lme4 for ratio data

My data is ratio data. So I'm trying to use lme4() with a binomial model to analyze it.

Here is my code:

fate.reP = glmer(predated~type+(1|island),data=fate.rate,family="binomial")

Here is a sample set of data:

type    cluster  tree   predated
 B        B7-1    1       0.48  
 B        B7-1    2       0.66
 B        B7-2    3       0.18
 M         I63    8       0.55
 M         I63    9       0.6
 M         I63   20       0.41
 M         I63   21       0.42
 S         I14    5       0.75
 S         I14   17       0.53
 S         I15    6       0.23
 S         I15    7       0.03

When I run the model, it shows that :

Error: (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate
In addition: Warning message:
In eval(expr, envir, enclos) : non-integer #successes in a binomial glm!

Are there errors in my data, or any other errors? I am using R 3.0.3 for Windows.

Upvotes: 0

Views: 2259

Answers (1)

Ben Bolker
Ben Bolker

Reputation: 226577

The warning non-integer #successes in a binomial glm! gives you the hint that binomial responses in lme4 must be integers. I don't see the denominators (i.e. total number of individuals exposed to predation) anywhere here: if you have them in your data set as (e.g.) total_exposed, you can use

fate.reP <- glmer(predated~type+(1|island),
     data=fate.rate,family="binomial",
     weights=total_exposed)

The behaviour of lme4 is a little bit different from glm in base R, which will warn you but still produce results if glm is non-integer.

Upvotes: 0

Related Questions