NEW: API

[mattansb]

There are 3 major functions that need a reliable API. This is the goal for version 1.0.0

eta_squared()

Essentially any model that can produce one of 3 tables will be supported:

• Basic (aov)
• Multi-level (aovlist)
• Approx (anova-F)

And also:

• parameters_model - it should be possible to get the effect size without having to also call parameters!
  • Waiting for "object_name" attribute missing for ANOVA-like outputs parameters#624
  • split by response, if present

standardize.default()

• Nothing to do, really.
• Check standardize.mlm() - waiting for standardization of matrix converts it to vector datawizard#42
• Check standardize.coxph()

standardize_parameters()

• For "refit" need to be supported in standardize.
• For post hoc methods, needs to be supported by standardize_info
• Fix for multivariate models support for post hoc standardization for multivariate models parameters#707

And also:

• parameters_model - it should be possible to get the effect size without having to also call parameters!
  • Have standardize_parameters() use standardize_parameters.parameters_model() for complete usage.

standardize_info()

• Add more info to the vignette.

[bwiernik]

Should it be eta_squared()?

[mattansb]

hehe yeah

[mattansb]

@strengejacke I am working here on allowing full support for parameters_model objects. This is intended to somewhat "soften" the circular dependency between effectsize and parameters.

When I'm done, ideally, parameters will not have to pass regression / anova models to effectsize, and instead can (and should) pass the processed parameters_model data.frame object (instead of what we currently have - multiple calls to model_parameters, which is what is causing the "hard" circular dependency).

I'll let you know when this is done - then we can think about how to utilize these changes in parameters.

[bwiernik]

What does parameters need from effectsize? Can we eliminate that entirely?

[mattansb]

It gives users an option to append the effect sizes to the parameters' table.

Historically, all the effectsize functions were part of parameters, and at some point they broke off to do their own thing.

Perhaps, in the future, this can be handed off to report::report_table() to do the parameters + effectsize + performance binding.

[bwiernik]

Yeah, that's something that would be better in report I think, same with, eg, getting both raw and exponentiated coefficients in one table.

Is this specifically for ANOVA type contexts?

[mattansb]

For regression models (for standardized slopes), ANOVA tables, and some htest models.

[strengejacke]

and instead can (and should) pass the processed parameters_model data.frame object

This might be tricky for standardized coefficients. For standardize = "refit", I call effectsize::standardize() before I do all the stuff in model_parameters(). But at least we might reduce the circular dependency.

[mattansb]

Not only the standardization would neen to be moved to parameters, also the ANOVA effect size functions and the effectsize.htest() (which depends on cohens_d(), rank_biserial(), etc....). So literally everything, except for the interpret functions (which, as we all know, should be in report).

[mattansb]

Post-hoc standardize_parameters() API is getting too big - merging this for now.