Forum: help

Negative binomial pairwise comparison [ Reply ]
By: Harry Peck on 2018-03-15 18:32

Hi all,

I'm trying comparing the levels of aggression in animals in response to different resources. I have fitted a negative binomial regression to my data:

NBLatency <- glm.nb(Latency_s ~ Resource, data = Bug)
summary(NBLatency)

Call:
glm.nb(formula = Latency_s ~ Resource, data = Bug, init.theta = 0.9811371849,
link = log)

Deviance Residuals:
Min 1Q Median 3Q Max
-2.124 -1.206 -0.466 0.438 1.799

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 4.7237 0.2927 16.138 <2e-16 ***
ResourceBurrow 0.2179 0.4231 0.515 0.606
ResourceFemale 0.5471 0.4817 1.136 0.256
ResourceFood 0.4057 0.4136 0.981 0.327
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for Negative Binomial(0.9811) family taken to be 1)

Null deviance: 49.908 on 41 degrees of freedom
Residual deviance: 48.306 on 38 degrees of freedom
AIC: 513.26

I wanted to compare each resource to each other resource, however, I was told that a tukey test is not appropriate when using a negative binomial. While I understand that in this particular example, there is likely to be no significant differences between resources, I am comparing a number of different factors which indicate aggression, in the same way, so knowing how to do this would be very useful.

I'd really appreciate any help.

Thanks!