lrberge
commented
1 month ago Hi Elias!
I must say that your comment strikes a chord. This is exactly what I wanted to do: establish a frameworks that makes user contributions seamless. I actually started some work on it two years ago (see here) but never had the time to finish, although the bulk of the work is done.
The steps were:
- [X] create a system to easily tell when the fitstat is available for a given model
- [X] easily integrate a short (on user request with a new arg. help = 'keyword-regex') and an extended documentation
- [ ] implement the short documentation display
- [ ] automatic integration in the docs of the extended documentation of each fitstat
- [ ] create a function that checks if a new fitstat never returns an error for any model (first stage validation of user-provided fistats)
- [ ] refactor the internal fitstat function to work only with a list of existing fitstats
- [ ] write guidelines on how users can contribute new fitstats to the fixest package
The most difficult steps were to define a design and the function telling which model is which (that was a pain), and this is done.
So this is definitely planned! When it will be done... that's another question however. My fixest to-do list is pretty big and the time I can devote to it is pretty slim. I think I will be able to get a couple of weeks of work on fixest at the beginning of next year, I hope I can code that up at that time.
Thanks for the reminder :-)
Hello,
first of all, thanks for the package @lrberge!
This is not an issue per se, but rahter a question/ suggestion: Would it be possible to establish a common pool for individual fitstat creations? (Or maybe if there exists something like that, please lead me towards it but I could not find anything.)
I am thinking of something like common thread on here or similar where people could share their functions for customized tests (using fitstat_register) so other people may be able to use them as well? For instance, I would be very interested in the code by @24thronin on the Sanderson and Windmeijer (2014) first stage F statistic. I think, in general, their lays potential benefit in offering a sort of institionalized way of sharing customized fitstat statistics!
Even better would be of course an easy way to make the command available publicaly directly through fitstat but I am guessing this will then often also lead to buggy results due to unique customizations and model kinds...
Is there anything like that planned by any chance?
Best wishes, Elias