I have a dataset in which I compare the total number (the summed number of birds over a couple of weeks) of birds across different distances (continuous factor). I hypothesize that the numbers of birds is the highest at the first distance and declines. The model I used was a GLM.

I now have a significant effect of the Distance, which is what I was expecting. But what can I use to see which of the distances differ from each other? I've tried this but that didn't work.

post1 <- glht(m5, family = poisson())
summary(m5, test = adjusted())

The emmeans package does not work.

This question can be deleted, the answer did not work and I did not conduct a post-hoc.

  • $\begingroup$ What do you mean by a "continuous factor"? Is the variable numeric (continuous) or a factor (categorical)? $\endgroup$ – Mark White Apr 26 '18 at 0:30
  • $\begingroup$ Please don't erase questions, at least not by overwriting them. If you really want to, you can delete the question. If you found a workable solution to your problem, you can post it as an answer. $\endgroup$ – Ben Bolker Jul 25 '18 at 14:17

You want the emmeans package. It is remarkably broad and effective. Below, I simulate data that might look like yours. Then I calculate all pairwise comparisons and then just the contrast of interest that you are interested in.

# generate fake data that might look like what you want ------------------------
set.seed(1839) # for replicability
n <- 400 # sample size
Distance <- factor(sample(1:5, n, TRUE)) # make distance from 1 to 5 and factor
# make rain and method covariates that are unrelated to outcome: 
Rain <- rbeta(400, 4, 2)
Method <- rbinom(n, 1, .4)
# calculate number that distance 1 is higher than the other 4:
Number <- rpois(n, ifelse(Distance == "1", 4, 2))

# fit model --------------------------------------------------------------------
model <- glm(Number ~ Distance + Rain + Method, family = "poisson")

# use emmeans ------------------------------------------------------------------
# all post-hoc comparisons:
pairs(emmeans(model, ~ Distance))

# just the contrast you wanted:
contrast(emmeans(model, ~ Distance), method = "trt.vs.ctrl", ref = 1)
# see ?contrast-methods for more information on types of methods of contrasts
# as well as how to specify the reference (that is, it should be the level value
# and not the label)

There are tons of great vignettes to help learn emmeans, e.g.:

  • 1
    $\begingroup$ This question looks entirely off topic--ie, only about code & packages. What saves it is the OP's apparent confusion about some of the underlying statistical concepts (eg, about continuous vs factor variables, about how distance as used in the formula relates to the idea that distance 1 might differ from the others, but the others not differ w/i themselves, etc). Can you address any of that in your answer? $\endgroup$ – gung May 16 '18 at 15:28

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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