0

I am studying the factors that influence mosquito feeding behavior. In the experiments, N mosquitoes are exposed to a host for a duration t. At the end of this exposure, we count how many mosquitoes (k) fed. This is repeated across several hosts, but the duration of exposure varied. I want to incorporate the duration in the response, a bit like an offset, so that the model considers that a longer exposure increases the proportion of fed mosquitoes we observe at the end. I was thinking of doing (syntax with glmmTMB package from R for example):

model <- glmmTMB(k/(N*t) ~ x1 + x2 + ...,
                 family = binomial(link = "logit"),
                 weights = N*t)

But I feel like that might be wrong. My response is no longer a proportion that can reach one (even if k = N), and that might mean that I cannot interpret odd ratios for instance : the probability of success and failure is not straightforward.

What would be your suggestion?

kjetil b halvorsen
  • 77,844
alpagarou
  • 51
  • 1
    why would you not introduce t as an independent variable? – seanv507 Sep 08 '23 at 14:38
  • Short version : because I want to force the effect and not estimate it. Long version : for the first part of the experiment, the proportions of mosquitoes fed were lower than expected, so for the second part of the experiment, it was decided to expose them a little longer, hoping to increase their feeding. It did not happen and the proportions of fed mosquitoes were even a bit lower than in the first part. So basically when I include duration as a variable, it can seem that it has a negative effect, but this is not biologically plausible, which is why I'd rather force the effect. – alpagarou Sep 11 '23 at 08:12
  • 1
    I think this thread may be relevant: Modelling a binary outcome when census interval varies. – dipetkov Sep 24 '23 at 15:30

0 Answers0