search

easystats / parameters

:bar_chart: Computation and processing of models' parameters
https://easystats.github.io/parameters/
GNU General Public License v3.0
438 stars 36 forks source link

Effect size package? #138

Closed strengejacke closed 5 years ago

strengejacke commented 5 years ago

We have quite some stuff dealing with effect sizes, conversion and ANOVA stuff. What do you think, does it make sense to separate these functions from parameters and put them into an own package?

DominiqueMakowski commented 5 years ago

IMO effect sizes for models are closely related to parameters (being just scaleless parameters), so, for now, I'd say we can keep the handful of conversion functions here (as they are quite straightforward to maintain) and are a core aspect of the parameters engineering scope of the package. That being said, if we start to extend this aspect (for instance metaanalysis tools etc.) then that might become necessary indeed...

(but let's get the 3 remaining packages on CRAN before adding more 😅)

DominiqueMakowski commented 5 years ago

Mmh after spending my morning overthinking standardized parameters and all, I think you might be right. We could create a separate package containing methods for standardizing, converting and interpreting parameters (mainly move functions from here and from report). That would ease their subsequent inclusion in parameters and report. It would also be easier to document it there.

I'll create the new package, and copy the relevant functions without removing them from parameters and report until the package in on CRAN. We'll see how it works having this package separate.

This could go fast, as most functions work already. @strengejacke are you okay to take over parameters and be the one listed as maintainer?

strengejacke commented 5 years ago

I would probably leave standardization in parameters? Making effectsize a similar small package like correlation?

DominiqueMakowski commented 5 years ago

standardization is needed for effet sizes, but not needed for parameters. And it better fits there IMO. But as effectsize would be super light (no dependencies (not even suggested) aside from insight and (?) bayestestR), it's not a problem to import it in the others

strengejacke commented 5 years ago

True.

strengejacke commented 5 years ago

Although standardization is something I often use in modelling, while I actually never use effect size (conversion).