I'm having hard time interpreting results from glm.diag.plots (package 'boot'). I read through some questions and answers (this one was very helpful, thank you! Interpreting glm.diag.plots) but didn't find relevant answers to the question I have. I'd like to apologize first just in case my tacky questions may nag at your nerves..
I have analysed my field-sampled data, which is very skewed, by a Gamma family GLM (using a log link). By the help of glm.diag.plots, I gained diagnostics plots as shown below:
result plots from glm.diag.plots]3
- I guess the upper right plot is similar to a normal Q-Q plot, which assures me I can use a Gamma glm. Am I right?
- The problem is the bottom two panels. Is the Cook statistics the same as Cook's distance? According to the package's document, Points above and to the right of the horizontal and vertical line, respectively, are the points with high influence on the model. Do they have be to omitted from the dataset? As the data were obtained from field experiment, I recognize there should be some outliers, but hesitate to leave them out..!
- Then what about the final plot? As it is shown, the only influential point, which is above the line, is #30 from the dataset. What is the difference between the bottom two panels?
If anybody can help, it would be fantastic. Please forgive the noob question! Thank you.
