I conducted a glm.nb by
glm1<-glm.nb(x~factor(group))
with group being a categorial and x being a metrical variable. When I try to get the summary of the results, I get slightly different results, depending on if I use summary()
or summary.glm
. summary(glm1)
gives me
...
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 0.1044 0.1519 0.687 0.4921
factor(gruppe)2 0.1580 0.2117 0.746 0.4555
factor(gruppe)3 0.3531 0.2085 1.693 0.0904 .
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for Negative Binomial(0.7109) family taken to be 1)
whereas summary.glm(glm1) gives me
...
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.1044 0.1481 0.705 0.4817
factor(gruppe)2 0.1580 0.2065 0.765 0.4447
factor(gruppe)3 0.3531 0.2033 1.737 0.0835 .
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for Negative Binomial(0.7109) family taken to be 0.9509067)
I understand the meaning of the dispersion parameter, but not of the line
(Dispersion parameter for Negative Binomial(0.7109) family taken to be 0.9509067)
.
In the handbook it says, it would be the estimated dispersion, but it seems to be a bad estimate, as 0.95 is not close to 0.7109, or is the estimated dispersion something different than the estimated dispersion parameter?
I guess, I have to set the dispersion in the summary.nb(x, dispersion=)
to something, but I'm not sure, if I have to set the dispersion to 1 (which will yield the same result as summary()
or if I should insert an estimate of the dispersion parameter, in this case leading to summary.nb(glm1, dispersion=0.7109)
or something else? Or am I fine with just using the summary(glm1)
?