1

I'm studying a colonial organism, and my hope is to compare differences in percent survival between three treatment groups. The results are clear, there is a 55% difference in survival between control and treatment groups and very little variance inside of treatments or seasons.

I have analyzed my data using a GLM, with a quasi-binomial error distribution, because of overdispersion (ugh). The trouble is that I have a considerable range of natural colony sizes (the denominator), but survival does not depend on colony size.

Describing this isometric relationship is a major point of my study, and I DO NOT wish to give more weight to large colonies in my model. Does a GLM minimize my small colonies and give more weight/importance to large colonies? Can someone explain to me how it works and what my alternatives might be (aside from arcsine transformation). I'm a novice and just want to do my proportion data justice.

I'm using R. Here is my model:

model1 <- glm(cbind(alive,dead) ~ treatment + season, family=quasibinomial)

I appreciate the help.

Disproportionately confused
  • 21
  • 4
  • Are your data counts of 'alive' & 'dead'? If so, an estimate of the odds of survival from a large colony (having, say, 100 alive & 200 dead) is more precise than an estimate from a small colony (having, say, 10 alive & 20 dead), & your GLM approach will take this into account. That's not the same as saying the odds of survival depend on colony size. Is it lack of independence between survival of different organisms in the same colony you're worried about? – Scortchi - Reinstate Monica Aug 18 '15 at 10:14
  • Thank you, this answers my question. I am dealing with thousands to hundreds of thousands of individuals per colony, so perhaps this is less of a problem than I thought. Also, I am not worried about the independence of individuals in each colony. Instead I am interested in the colony-level response. – Disproportionately confused Aug 18 '15 at 16:56

0 Answers0