How to present GLM models when you want to show how x affects y? I am trying to figure out the best way to plot a gamma GLM.
Here is the data and model:
icecream <- data.frame(
  flavor=c(0, 1, 1, 0, 1),
  proportion=c(0.8, 0.78, 0.9, 0.66, 0.25)
)

icecream.model <- glmrob(proportion ~ flavor, data = icecream, family = Gamma)
summary(icecream.model)

It seems like my first option is plotting the predicted probabilities:
newdat2 <- data.frame(flavor=seq(min(icecream$flavor), max(icecream$flavor),len=100)) # is this the right way to look at it?
newdat2$vs = predict(icecream.model, newdata=newdat2, type="response")
plot(proportion ~ flavor, data = icecream, col="red4")
lines(vs ~ flavor, newdat2, col="green4", lwd=2)

I personally think that plotting the odds makes it easier to interpret:
plot_model(icecream.model, show.values = TRUE, show.p = TRUE)

Is there a best practice for plotting GLM models?
 A: If your response variable is a (continuous) proportion, why not model it using a beta regression model? The beta regression model is not a GLM model but shares some similarities with GLM models.  
For GLM models, the effects package in R provides ways to visualize the model effects on various scales (including the response scale).  For example, if the GLM model is a binary logistic regression model, effects can be visualized on the log odds scale or probability scale.  If the GLM model is a Gamma regression model, effects can be visualized on the log scale or response scale.  
Typical usage for the effects package is plot(allEffects(model)), where model is an appropriate fitted-model object. 
Note that the effects package will also work with beta regression models fitted via the betareg package in R.
The vignettes for the effects package available at https://cran.r-project.org/web/packages/effects/index.html will give you more insights into the package functionality.
Which scale you choose to visualize your effects depends both on your audience and your purposes.  An audience which is not statistically sophisticated may benefit from visualizing effects on a scale that is more easily understood (e.g., probability scale for binary logistic regression models, rather than log odds scale). 
Another R package which is helpful for visualizing effects of GLM models is the ggeffects package. See for instance https://cran.r-project.org/web/packages/ggeffects/vignettes/ggeffects.html.
