Using global options

[mattansb]

We don't use options() or getOption() in {easystats} - is there a reason?

I was thinking of implementing these for control of CIs (for the ci and alternative arguments), something along the line of:

# in zzz.R
.onLoad <- function(libname, pkgname) {
  op <- options()
  op.es <- list(
    es.ci.mag = 0.95,
    es.ci.diff = 0.95,
    es.alt.mag = "greater",
    es.alt.diff = "two.sided"
  )

  toset <- !names(op.es) %in% names(op)
  if(any(toset)) options(op.afex[toset])

  invisible(NULL)
}

Then functions would be defined:

eta_squared <- function(..., 
                        ci = getOption("es.ci.mag", 0.95), 
                        alternative = getOption("es.alt.mag", "greater"),
                        ...) {
  # ...
}


cohens_d <- function(..., 
                     ci = getOption("es.ci.diff", 0.95), 
                     alternative = getOption("es.alt.diff", "two.sided"),
                     ...) {
  # ...
}

This could also be useful in bayestestR (where I personally am a fan of the 89% CI... 🙃)

[strengejacke]

Yeah, I was thinking about this, too, e.g. I often use model_parameters(summary = TRUE), so customizing settings would be nice. It requires a good documentation, so it's accessible or visible to users. Maybe also in our meta-package?

[mattansb]

We can have them where ever we need them!

My only concern is that users might find it opaque :/ So good clear documentation is indeed required...

[strengejacke]

[strengejacke]

ôr

[strengejacke]

The question is: do we want getOption() in the function definition (like in your examples), which might be confusing to new users and really looks a bit ugly...

Or do we want to check inside a function, like:

foo <- function(ci = .95) {
  if (missing(ci)) {
    ci <- getOption("ci", .95)
  }
}

[bwiernik]

I think ci = getOption("easystats.ci", defaultValue = .95) probably isn't too alarming or confusing. I would say we should have just a single global CI option.