Option for non-count estimator (estimating equations interpretation)

[adw96]

Currently, corncob is set up to only take integer $W_ij$'s, which makes sense as it outputs MLEs from a beta-binomial model. However, we could consider a generalization for non-integer W_ij's, using the beta-binomial model to motivate estimating equations, which would still give us CANE estimates. Useful for modeling coverages! (and needed for a future project for which corncob is a dependency)

TODO: generalized corncob to estimate under non-integer W_ij's.

Must output a warning, because then people can throw anything in, without any guarantee that what they are doing is sensible, and previously this was often caught by the non-integer error.

Another option is to have a optional flag that defaults to FALSE allow_noninteger... maybe I like this idea better?

[adw96]

Implementation of this currently on testing 1c3dd4a. Some notes

• I believe that the existing functionality to throw an error with non integer accounts is maintained by default. The error message is "Double-check and correct your data, and/or force the model to fit with allow_noninteger"
• If allow_noninteger is TRUE the model will fit. Checks confirm that rounding non-integers gives similar results.

Still TODO

• Sandwich standard error should be returned by default if allow_noninteger is TRUE. Very grateful to @bryandmartin for making this easy (so it seems).
• Think about whether LRT should be banned given EE setting (is there a model-robust interpretation?)
• dbetabinom_cts_mod and relatives uses code from VGAM::dbetabinom.ab but relaxes checks. Best practice for crediting the code source? Perhaps @svteichman could do some research and help advise 🙏

No need to review yet, @svteichman -- I'll open a PR when it's ready.

[bryandmartin]

This is a great new feature!

[svteichman]

@adw96 I did some reading and I think a good practice could be adding the authors of VGAM as authors in the DESCRIPTION file with the tag "cph" for copyright holder and then add a comment for the author saying that they are the creator of the VGAM package. Additionally, I think we should add a comment for the dbetabinom_cts_mod function stating that it is a modified version of VGAM::betabinom.ab. Here's an example of attributing package authorship like this in the shiny package.

[svteichman]

Or if copyright holder is not quite right in this situation (I'm not totally sure what is appropriate given the GPL-3 license for VGAM), they could instead by listed as a contributor with "ctb"

[svteichman]

Closing because this is implemented in merged pull request #155