I have fit a negative binomial model in R, and would like to report the findings, but I'm unsure how (or if) I should convert the estimates to reportable coefficients. Here is my output:
> summary(negbinom)
Call:
glm.nb(formula = Frequency ~ Groups, data = Data, init.theta = 4.577879741,
link = log)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.3978 -0.8346 0.0000 0.4622 2.4414
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.32176 0.14217 9.297 <2e-16 ***
Group1 -0.47446 0.20440 -2.321 0.0203 *
Group2 -0.01117 0.20137 -0.055 0.9558
Group3 0.06454 0.19221 0.336 0.7370
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for Negative Binomial(4.5779) family taken to be 1)
Null deviance: 139.05 on 105 degrees of freedom
Residual deviance: 129.82 on 102 degrees of freedom
AIC: 478.1
Number of Fisher Scoring iterations: 1
Theta: 4.58
Std. Err.: 1.70
2 x log-likelihood: -468.104
My questions are:
1) is it okay to simply report that Group 1 reported stuff I'm looking at with a lower frequency than Group 0 (the reference group), B = - .475, z = -2.32, p = .020? Or do I need to do something (exp() maybe?) with the coefficients to be able to report? Even then is this the right notation to use (B, z, etc.)?
2) can I use Anova() or anova() to get the omnibus effect of Groups? e.g., can I do this:
> Anova(negbinom)
Analysis of Deviance Table (Type II tests)
Response: Frequency
LR Chisq Df Pr(>Chisq)
Groups 9.2233 3 0.02646 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
...and then report it as: ChiSq(3) = 9.22, p = .026?