Additional models to support with summ

[jacob-long]

My general philosophy goes like this:

• The model's output needs to be relatively predictable (this is why I have not supported lavaan, which can be endlessly complicated and used for very different purposes)
• The model should be regression or similar — summ will not handle other kinds of input, like data.frames or the like. skimr is a package that does those things well.
• summ should be able to offer added value above and beyond summary

With that said, models I definitely plan to support are:

• lme

Still thinking about/auditing:

• brmsfit — worried about variation in output due to wide variety of options, unsure if summ can add benefit since refitting models isn't feasible.
• stanreg — less concern about variation than with brmsfit, but "added value" concern remains
• polr — Need to look more closely at the interface, make sure I know enough to make a good summary. Need to think about how to plot predictions from these models (same goes for ordinal package models), but that isn't essential.

Checklist of models I plan to add barring complications as I implement them (and outside contributors may feel free to do a pull request for one of these):

• [ ] lme (others in nlme?)
• [ ] glmmTMB

[jacob-long]

Suggestions are welcome. Hopefully soon I'll better document to outsiders how to make a summ method, but it's been pretty ad hoc to this point.

[some-r-user]

How about adding the support for packages supporting mixed models with specified residual covariance, like nlme glmmTMB, glmPQL? lme4 cannot do that, which eliminates it from the analysis of longitudinal trials, where the assumptions made by lme4 are totally unrealistic.

[jacob-long]

I will look into that, @some-r-user. I had in the past looked into nlme support in particular and can't recall right now why I ended up not going forward with it. I have also gotten a couple of other requests for glmmTMB and will check into it and glmmPQL as well. This will depend somewhat on how easy their developers have made it to access useful information from their models.

[ngreifer]

It might be cool to see support for vglm objects in VGAM. I don't know if vgam objects have a similar enough form (I've never used them).

[nagbech]

Hello everybody, I'm not sure it is the right place to post it. I'm surprised nobody mentionned basicivreg objects: is this type of object already compatible? Because on my side, I'm not able to use jtools functions with my ivreg objects. Could someone give me an example of how to use jtools Library with such type of objects?

Best, Nagui

[hliebert]

It would be great if you could support felm from library(lfe). lfe supports linear regression with high-dimensional fixed effects, instrumental variables regression and clustering. Having ivreg support, as mentioned above, would also be nice. Thank you for your work on jtools!