convert_oddsratio_to_riskratio() / convert_oddsratio_to_riskdiff() and vice versa

[bwiernik]

convert_oddsratio_to_odds <- function(OR, log = FALSE, base_rate = NULL, base_odds = NULL) {
  if (is.null(base_odds)) {
    if (is.null(base_rate)) {
      insight::format_error("Either `base_rate` or `base_odds` must be supplied.")
    }
    base_odds <- convert_probs_to_odds(base_rate, log = log)
  }
  compare_odds <- ifelse(isTRUE(log), OR + base_odds, OR * base_odds)
  return(compare_odds)
}

convert_oddsratio_to_probs <- function(OR, log = FALSE, base_rate = NULL, base_odds = NULL) {
  compare_odds <- convert_oddsratio_to_odds(OR = OR, log = log, base_rate = base_rate, base_odds = base_odds)
  compare_rate <- convert_odds_to_probs(compare_odds, log = log)
  return(compare_rate)
}

convert_oddsratio_to_riskratio <- function(OR, log = FALSE, base_rate = NULL, base_odds = NULL) {
  compare_rate <- convert_oddsratio_to_probs(OR = OR, log = log, base_rate = base_rate, base_odds = base_odds)
  if (is.null(base_rate)) {
    base_rate <- convert_odds_to_probs(base_odds, log = log)
  }
  riskratio <- compare_rate / base_rate
}

convert_oddsratio_to_riskdiff <- function(OR, log = FALSE, base_rate = NULL, base_odds = NULL) {
  compare_rate <- convert_oddsratio_to_probs(OR = OR, log = log, base_rate = base_rate, base_odds = base_odds)
  if (is.null(base_rate)) {
    base_rate <- convert_odds_to_probs(base_odds, log = log)
  }
  riskratio <- compare_rate - base_rate
}

[mattansb]

We have the OR to RR one here https://easystats.github.io/effectsize/reference/oddsratio_to_riskratio.html

Should we have all possible pairings?

[bwiernik]

Oh, then I guess the issue is that some of our functions start with convert and some don't. That's confusing and we should consider standardizing

[mattansb]

I think the verbosity can be reduced (:

[strengejacke]

Do any functions start with convert_*?

[strengejacke]

Just saw, that seem to be aliases, right? Maybe we can deprecate those?

[strengejacke]

But for convenience reasons, we could probably add those conversion functions, too? And also add log-odds functions that call the oddsratio-functions with log = TRUE? Some colleagues were using the effectsize package and "complaining" that you can only convert to/from odds ratio, but not log-odds. They missed the log argument, which also isn't clearly documented (people need "keywords" like "log-odds" or similar).

[mattansb]

Okay, so I think we need two sets of functions:

1. Between prob-change metrics

• [x] OR to RR (oddsratio_to_riskratio)
• [x] OR to ARR (oddsratio_to_arr)
• [x] OR to NNT
• [x] logOR to RR
• [x] logOR to ARR
• [x] logOR to NNT
• [x] RR to OR (riskratio_to_oddsratio)
• [x] RR to logOR
• [x] RR to ARR (riskratio_to_arr)
• [x] RR to NNT
• [x] ARR to OR (arr_to_oddsratio)
• [x] ARR to logOR
• [x] ARR to RR (arr_to_riskratio)
• [x] ARR to NNT (arr_to_nnt)
• [x] NNT to OR
• [x] NNT to logOR
• [x] NNT to RR
• [x] NNT to ARR (nnt_to_arr)

Do we need??

• [ ] OR to logOR
• [ ] logOR to OR

2. Change to Probs

• [ ] OR to prob
• [ ] logOR to prob
• [ ] RR to prob
• [ ] ARR to prob
• [ ] NNT to prob

@strengejacke @bwiernik

• Do we also want the inverse of these? (e.g., prob to OR)

• Do we also want conversion to odds? (e.g., NNT to odds)

Also

• [x] Deprecate all the convert_ functions (sorry @DominiqueMakowski )
• [x] d to logOR

@strengejacke started on some of these in #569