0
$\begingroup$

I'm looking at the influence of pollen type on whether a flower sets fruit (i.e., yes or no = 1 or 0). Then looking at number of seeds per fruit (1-6 possible).

I was told I should use lmer, however it gave me an error:

calling lmer with 'family' is deprecated; please use glmer() instead

I've used glmer(FruitYesNo~Pollentype+(1|Plantnumber),family=binomial) for the fruit and glmer(Seednumber ~ Pollentype+(1|Plantnumber), family=poisson) for the seeds. But I've read that glmms won't perform well like that. Can anyone give me any advice on what model I should be using?

Improve this question
$\endgroup$
3
  • 3
    $\begingroup$ Where did you read that "glmm's won't perform well like that"? What was the argument? On a different note, be aware that questions that are only about how to use software are generally considered off-topic here. This Q strikes me as on-topic, but near the borderline; I don't know if others will disagree. $\endgroup$
    – gung - Reinstate Monica
    Commented Apr 24, 2014 at 14:07
  • $\begingroup$ Sorry, I didn't realise. I can delete the question if it's off topic. What I read was about the number of random effects $\endgroup$
    – Emilyt
    Commented Apr 24, 2014 at 14:18
  • $\begingroup$ No need to apologize, @Emilyt, I think your Q is on-topic here; I was just letting you know. Can you say where you read this, & what it said about the number of random effects? $\endgroup$
    – gung - Reinstate Monica
    Commented Apr 24, 2014 at 14:45

1 Answer 1

Reset to default
1
$\begingroup$

This is just a confusion of terminology. Sometimes people refer to using mixed models to mean either linear mixed models or generalized linear mixed models. The "mixed" part refers to model-based approaches to handling dependent data with random effects.

If you used a linear mixed model for binary outcomes, you would have the problems of predicted fruiting probabilities outside the 0-1 interval which is nonsensical, and prediction intervals would be homoscedastic which is generally not true of proportions.

The binomial mixed model is probably the one you want. There's no reason against choosing this model provided it answers your question of interest. The syntax is (now) glmer and not lmer as you had previously written,.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.