I am analysing data using a GLM. In some GLM-s data is binary while in others the response variable is count data. I want to know if it is necessary to use an ANOVA after constructing the GLM model. Or is it enough to report the p-values of the GLM-s?
1 Answer
$\begingroup$
$\endgroup$
4
The ANOVA test, of the regression equation, is essentially a test to determine if your proposed model is "better" than simply modelling the outcome using only the mean (of the outcome).
As per your comments nothing is forcing you to run ANOVA style testing, but most standard software will compute it for you regardless. In your case it does not seem in particular relevant.
-
$\begingroup$ Thank you @Repmat. So ANOVA can also be conducted on GLM models which are constructed on binary data? $\endgroup$– ChzMCommented Jun 7, 2017 at 18:43
-
1$\begingroup$ @kamel yes indeed, typically this is referred to as the chi square test of the (logit) model. $\endgroup$– RepmatCommented Jun 7, 2017 at 18:56
-
1$\begingroup$ Usually it would be called an analysis of deviance. $\endgroup$ Commented Jun 7, 2017 at 19:16
-
1$\begingroup$ Thank you for the help. I will then remove the ANOVA analysis from the paper since the GLM already proved the effect of one variable on the other. $\endgroup$– ChzMCommented Jun 8, 2017 at 10:23
anovain R to get a p-value). Well you can do something quite similar with a GLM, but it's analysis of deviance and based on asymptotic chi-square tests rather than small-sample F results. You're not very clear about what things you want to find out. $\endgroup$