R/effectsize.BFBayesFactor.R

In effectsize: Indices of Effect Size

#' @export
#' @rdname effectsize
#' @inheritParams bayestestR::describe_posterior
effectsize.BFBayesFactor <- function(
  model,
  type = NULL,
  ci = 0.95,
  test = NULL,
  verbose = TRUE,
  ...
) {
  insight::check_if_installed("BayesFactor")

  if (length(model) > 1) {
    if (verbose) {
      insight::format_warning("Multiple models detected. Using first only.")
    }
    model <- model[1]
  }

  if (inherits(model@numerator[[1]], "BFcontingencyTable")) {
    pars <- .effectsize_contingencyTableBF(
      model,
      type = type,
      verbose = verbose,
      ...
    )
  } else if (
    inherits(model@numerator[[1]], c("BFoneSample", "BFindepSample"))
  ) {
    pars <- .effectsize_ttestBF(model, type = type, verbose = verbose)
  } else if (inherits(model@numerator[[1]], "BFcorrelation")) {
    pars <- .effectsize_correlationBF(model, type = type, verbose = verbose)
  } else if (inherits(model@numerator[[1]], "BFproportion")) {
    pars <- .effectsize_proportionBF(model, type = type, verbose = verbose)
  } else {
    insight::format_error(
      "No effect size for this type of 'BayesFactor' object."
    )
  }

  # Clean up
  out <- bayestestR::describe_posterior(pars$res, ci = ci, test = test, ...)
  if (isTRUE(type == "cles")) {
    colnames(out)[2] <- "Coefficient"
  } else {
    colnames(out)[2] <- out$Parameter
    out$Parameter <- NULL
  }

  class(out) <- c(
    pars$xtra_class,
    "effectsize_table",
    "see_effectsize_table",
    class(out)
  )
  .someattributes(out) <- pars$attr
  .someattributes(out) <- list(
    ci = out$CI,
    approximate = FALSE,
    alternative = "two.sided"
  )
  out
}

#' @keywords internal
.effectsize_contingencyTableBF <- function(
  model,
  type = NULL,
  verbose = TRUE,
  adjust = TRUE,
  ...
) {
  if (is.null(type)) {
    type <- "cramers_v"
  }

  f <- switch(
    tolower(type),
    v = ,
    cramers_v = cramers_v,
    t = ,
    tschuprows_t = tschuprows_t,
    w = ,
    cohens_w = cohens_w,
    phi = phi,
    c = ,
    pearsons_c = pearsons_c,
    h = ,
    cohens_h = cohens_h,
    or = ,
    oddsratio = oddsratio,
    rr = ,
    riskratio = riskratio,
    arr = arr,
    nnt = nnt
  )
  data <- insight::get_data(model)
  posts <- insight::get_parameters(model, verbose = verbose)

  ES <- apply(posts, 1, function(a) {
    M <- matrix(a, nrow = nrow(data))
    f(M, ci = NULL, adjust = adjust)[[1]]
  })

  res <- data.frame(ES)
  colnames(res) <- colnames(f(data, ci = NULL, adjust = adjust))

  list(
    res = res,
    attr = NULL,
    xtra_class = NULL
  )
}


#' @keywords internal
.effectsize_ttestBF <- function(model, type = NULL, verbose = TRUE) {
  if (is.null(type) || tolower(type) == "cohens_d") {
    type <- "d"
  }

  samps <- as.data.frame(BayesFactor::posterior(
    model,
    iterations = 4000,
    progress = FALSE
  ))

  paired <- inherits(model@numerator[[1]], "BFoneSample")
  if (!paired) {
    mu <- 0
    D <- samps$delta
  } else {
    mu <- model@numerator[[1]]@prior$mu
    D <- (samps$mu - mu) / sqrt(samps$sig2)
  }

  res <- data.frame(Cohens_d = D)

  if (type == "d") {
    xtra_class <- "effectsize_difference"
  } else if (
    tolower(type) %in% c("p_superiority", "u1", "u2", "u3", "overlap")
  ) {
    if (paired && type != "p_superiority") {
      insight::format_error("CLES only applicable to two independent samples.")
    }

    converter <- match.fun(paste0("d_to_", tolower(type)))
    if (grepl("^(u|U)", type)) {
      type <- paste0("Cohens_", toupper(type))
    }

    res <- data.frame(converter(res$Cohens_d), check.names = FALSE)
    colnames(res) <- type
    xtra_class <- NULL
  }

  list(
    res = res,
    attr = list(mu = mu, paired = paired, pooled_sd = TRUE),
    xtra_class = xtra_class
  )
}


# Others ------------------------------------------------------------------
# Wrappers

#' @keywords internal
.effectsize_correlationBF <- function(model, type = NULL, verbose = TRUE) {
  rho <- insight::get_parameters(model, verbose = verbose)[["rho"]]
  res <- data.frame(rho = rho)

  list(
    res = res,
    attr = NULL,
    xtra_class = NULL
  )
}


#' @keywords internal
.effectsize_proportionBF <- function(model, type = NULL, verbose = TRUE) {
  res <- insight::get_parameters(model, verbose = verbose)

  p0 <- model@denominator@identifier[["p0"]]
  xtra_footer <- list(c(sprintf("\n- Against the null: p = %s.", p0), "cyan"))

  list(
    res = res,
    attr = list(xtra_footer = xtra_footer),
    xtra_class = NULL
  )
}