Best way to visualize (in figures) the results from a non-normal GLM?

#1
Hi,

I've got a zero inflated, negative binomial GLM that I'm using and am curious what the best way to visually represent my results in figures would be. My model doesn't have a ton of factors; I'm looking at a blocking factor, a continuous predictor, the interaction between the two and two additional covariates.

Thoughts? Suggestions?
 

hlsmith

Omega Contributor
#2
I believe relative risks plotted as points with whiskers. Though if you have interactions, just the unique and higher ordered terms. It would help if you posted your output.
 
#3
Of course. So, this model here is actually just a negative binomial GLM (non-ZI) with a blocking factor, a continuous predictor (proportion of males), the interaction, and two additional covariates (male and female size). Here is the full model:

ScreenHunter_392 Jul. 19 11.35.jpg

And the reduced model:

ScreenHunter_393 Jul. 19 11.36.jpg

Here's a link to the outputs if the above images don't work. Prior to using penalized regression techniques (I used adaptive lasso in this case), I was just removing factors that I didn't think were necessarily relevant (in the context of this model) and that improved the AICc. I don't have a ton of experience with shrinkage and selection techniques but have been screwing around with lasso/elastic net techniques. Ultimately they both arrive at the same conclusion that the best model is the one with only the proportion of males as the predictor.

For such a simple model, what would be the best way to graph the results? Thank you!
 
#4
So, I've tried to find the best way to graph these. As far as I can tell, the two programs that I'm familiar with (JMP and SigmaPlot) don't have an option to fit a negative binomial curve. Any other programs you'd recommend I check out?