Thanks for starting this. I'll give your package a look once I have a moment and see how easily we can combine.

I need to look back over your package as well. I looked at it prior to writing marfx and it was the only package that was good, but I think I had a few different ideas about the API. I'll look again and write up my thoughts.

TODO:

• Improve the value of eps
• Switch to central difference approximation method (f(x+h) - f(x - h))/2h
• Optionally use simulation approach from marfx
• Expose internal calculators with better names a la marfx
• identify "logical" class variables and handle as factor
• handle "ordered" class variables as factor
• Add ggplot2 examples; ensure all output is tidy (it should be, but need to check)
• Add example to README of calling mfx() directly
• Add example data beyond datasets (prob. examples from marfx vignette)

@jrnold I think I've effectively incorporated the core functionality of your package, including the simulation-based variance estimation and the addition of a low-level dydx() generic to calculate marginal effects or discrete differences for a single variable in a model. I'm finishing up the vignette explaining the computational approach and when that's done, I plan to push to CRAN. I've credited you as an author, given some of the code is from your package.

Awesome. Sorry I haven't had time to look at it. As I'm prepping/using it for my upcoming classes I'll see what I can add/contribute.