Merge branch 'main' into update_outlier_vignette

DESCRIPTION

@@ -1,8 +1,8 @@
 Type: Package
 Package: performance
 Title: Assessment of Regression Models Performance
-Version: 0.12.0.4
-Authors@R: 
+Version: 0.12.0.9
+Authors@R:
     c(person(given = "Daniel",
              family = "Lüdecke",
              role = c("aut", "cre"),
@@ -39,18 +39,18 @@ Authors@R:
              email = "[email protected]",
              comment = c(ORCID = "0000-0003-4315-6788", Twitter = "@rempsyc")),
       person(given = "Vincent",
-             family = "Arel-Bundock", 
-             email = "[email protected]", 
+             family = "Arel-Bundock",
+             email = "[email protected]",
              role = "ctb",
              comment = c(ORCID = "0000-0003-2042-7063")),
       person(given = "Martin",
              family = "Jullum",
              role = "rev"),
       person(given = "gjo11",
              role = "rev"),
-      person("Etienne", 
-             "Bacher", , 
-             "[email protected]", 
+      person("Etienne",
+             "Bacher", ,
+             "[email protected]",
              role = "ctb",
              comment = c(ORCID = "0000-0002-9271-5075")))
 Maintainer: Daniel Lüdecke <[email protected]>
@@ -70,8 +70,8 @@ Depends:
     R (>= 3.6)
 Imports:
     bayestestR (>= 0.13.2),
-    insight (>= 0.20.1),
-    datawizard (>= 0.11.0),
+    insight (>= 0.20.2),
+    datawizard (>= 0.10.0),
     stats,
     utils
 Suggests:
@@ -93,14 +93,15 @@ Suggests:
     DHARMa,
     estimatr,
     fixest,
+    flextable,
     forecast,
     gamm4,
     geomtextpath,
     ggplot2,
     glmmTMB,
     graphics,
     Hmisc,
-    httr,
+    httr2,
     ICS,
     ICSOutlier,
     ISLR,
@@ -130,6 +131,8 @@ Suggests:
     quantreg,
     qqplotr (>= 0.0.6),
     randomForest,
+    RcppEigen,
+    rempsyc,
     rmarkdown,
     rstanarm,
     rstantools,
@@ -152,4 +155,3 @@ Config/Needs/website:
     r-lib/pkgdown,
     easystats/easystatstemplate
 Config/rcmdcheck/ignore-inconsequential-notes: true
-Remotes: easystats/insight

NAMESPACE

@@ -148,6 +148,7 @@ S3method(display,test_performance)
 S3method(fitted,BFBayesFactor)
 S3method(format,compare_performance)
 S3method(format,performance_model)
+S3method(format,performance_rmse)
 S3method(format,test_performance)
 S3method(logLik,cpglm)
 S3method(logLik,iv_robust)
@@ -319,6 +320,7 @@ S3method(print,performance_hosmer)
 S3method(print,performance_model)
 S3method(print,performance_pcp)
 S3method(print,performance_pp_check)
+S3method(print,performance_rmse)
 S3method(print,performance_roc)
 S3method(print,performance_score)
 S3method(print,performance_simres)
@@ -451,6 +453,7 @@ S3method(r2_coxsnell,survreg)
 S3method(r2_coxsnell,svycoxph)
 S3method(r2_coxsnell,truncreg)
 S3method(r2_efron,default)
+S3method(r2_ferrari,default)
 S3method(r2_kullback,default)
 S3method(r2_kullback,glm)
 S3method(r2_loo_posterior,BFBayesFactor)
@@ -597,6 +600,7 @@ export(r2)
 export(r2_bayes)
 export(r2_coxsnell)
 export(r2_efron)
+export(r2_ferrari)
 export(r2_kullback)
 export(r2_loo)
 export(r2_loo_posterior)

NEWS.md

@@ -14,6 +14,13 @@
 * `icc()` and `r2_nakagawa()` get a `model_component` argument indicating the 
   component for zero-inflation or hurdle models.
 
+* `performance_rmse()` (resp. `rmse()`) can now compute analytical and 
+  bootstrapped confidence intervals. The function gains following new arguments:
+  `ci`, `ci_method` and `iterations`.
+
+* New function `r2_ferrari()` to compute Ferrari & Cribari-Neto's R2 for 
+  generalized linear models, in particular beta-regression.
+
 # performance 0.12.0
 
 ## Breaking

R/check_distribution.R

@@ -191,33 +191,33 @@ check_distribution.numeric <- function(model) {
   # validation check, remove missings
   x <- x[!is.na(x)]
 
-  mode <- NULL
+  mode_value <- NULL
   # find mode for integer, or MAP for distributions
   if (all(.is_integer(x))) {
-    mode <- datawizard::distribution_mode(x)
+    mode_value <- datawizard::distribution_mode(x)
   } else {
     # this might fail, so we wrap in ".safe()"
-    mode <- tryCatch(
+    mode_value <- tryCatch(
       as.numeric(bayestestR::map_estimate(x, bw = "nrd0")),
       error = function(e) NULL
     )
-    if (is.null(mode)) {
-      mode <- tryCatch(
+    if (is.null(mode_value)) {
+      mode_value <- tryCatch(
         as.numeric(bayestestR::map_estimate(x, bw = "kernel")),
         error = function(e) NULL
       )
     }
   }
 
-  if (is.null(mode)) {
+  if (is.null(mode_value)) {
     mean_mode_diff <- mean(x) - datawizard::distribution_mode(x)
     msg <- "Could not accurately estimate the mode."
     if (!is.null(type)) {
       msg <- paste(msg, "Predicted distribution of the", type, "may be less accurate.")
     }
     insight::format_alert(msg)
   } else {
-    mean_mode_diff <- .safe(mean(x) - mode)
+    mean_mode_diff <- .safe(mean(x) - mode_value)
   }
 
   data.frame(

R/check_homogeneity.R

@@ -165,7 +165,7 @@ check_homogeneity.afex_aov <- function(x, method = "levene", ...) {
     between <- between[!is_covar]
   }
 
-  form <- stats::formula(paste0(dv, "~", paste0(between, collapse = "*")))
+  form <- stats::formula(paste0(dv, "~", paste(between, collapse = "*")))
   test <- car::leveneTest(form, ag_data, center = mean, ...)
 
   p.val <- test[1, "Pr(>F)"]

R/helpers.R

@@ -1,6 +1,6 @@
 # small wrapper around this commonly used try-catch
 .safe <- function(code, on_error = NULL) {
-  if (getOption("easystats_erros", FALSE) && is.null(on_error)) {
+  if (isTRUE(getOption("easystats_errors", FALSE) && is.null(on_error))) {
     code
   } else {
     tryCatch(code, error = function(e) on_error)

R/icc.R

@@ -1,5 +1,7 @@
-#' Intraclass Correlation Coefficient (ICC)
+#' @title Intraclass Correlation Coefficient (ICC)
+#' @name icc
 #'
+#' @description
 #' This function calculates the intraclass-correlation coefficient (ICC) -
 #' sometimes also called *variance partition coefficient* (VPC) or
 #' *repeatability* - for mixed effects models. The ICC can be calculated for all
@@ -15,32 +17,32 @@
 #' Else, for instance for nested models, name a specific group-level effect
 #' to calculate the variance decomposition for this group-level. See 'Details'
 #' and `?brms::posterior_predict`.
-#' @param ci Confidence resp. credible interval level. For `icc()` and `r2()`,
-#' confidence intervals are based on bootstrapped samples from the ICC resp.
-#' R2 value. See `iterations`.
-#' @param by_group Logical, if `TRUE`, `icc()` returns the variance
-#' components for each random-effects level (if there are multiple levels).
-#' See 'Details'.
+#' @param ci Confidence resp. credible interval level. For `icc()`, `r2()`, and
+#' `rmse()`, confidence intervals are based on bootstrapped samples from the
+#' ICC, R2 or RMSE value. See `iterations`.
+#' @param by_group Logical, if `TRUE`, `icc()` returns the variance components
+#' for each random-effects level (if there are multiple levels). See 'Details'.
 #' @param iterations Number of bootstrap-replicates when computing confidence
-#' intervals for the ICC or R2.
+#' intervals for the ICC, R2, RMSE etc.
 #' @param ci_method Character string, indicating the bootstrap-method. Should
-#' be `NULL` (default), in which case `lme4::bootMer()` is used for
-#' bootstrapped confidence intervals. However, if bootstrapped intervals cannot
-#' be calculated this was, try `ci_method = "boot"`, which falls back to
-#' `boot::boot()`. This may successfully return bootstrapped confidence intervals,
-#' but bootstrapped samples may not be appropriate for the multilevel structure
-#' of the model. There is also an option `ci_method = "analytical"`, which tries
-#' to calculate analytical confidence assuming a chi-squared distribution.
-#' However, these intervals are rather inaccurate and often too narrow. It is
-#' recommended to calculate bootstrapped confidence intervals for mixed models.
+#' be `NULL` (default), in which case `lme4::bootMer()` is used for bootstrapped
+#' confidence intervals. However, if bootstrapped intervals cannot be calculated
+#' this way, try `ci_method = "boot"`, which falls back to `boot::boot()`. This
+#' may successfully return bootstrapped confidence intervals, but bootstrapped
+#' samples may not be appropriate for the multilevel structure of the model.
+#' There is also an option `ci_method = "analytical"`, which tries to calculate
+#' analytical confidence assuming a chi-squared distribution. However, these
+#' intervals are rather inaccurate and often too narrow. It is recommended to
+#' calculate bootstrapped confidence intervals for mixed models.
 #' @param verbose Toggle warnings and messages.
 #' @param null_model Optional, a null model to compute the random effect variances,
 #' which is passed to [`insight::get_variance()`]. Usually only required if
 #' calculation of r-squared or ICC fails when `null_model` is not specified. If
 #' calculating the null model takes longer and you already have fit the null
 #' model, you can pass it here, too, to speed up the process.
 #' @param ... Arguments passed down to `lme4::bootMer()` or `boot::boot()`
-#' for bootstrapped ICC or R2.
+#' for bootstrapped ICC, R2, RMSE etc.; for `variance_decomposition()`,
+#' arguments are passed down to `brms::posterior_predict()`.
 #'
 #' @inheritParams r2_bayes
 #' @inheritParams insight::get_variance
@@ -247,7 +249,12 @@ icc <- function(model,
     # iccs between groups
     # n_grps <- length(vars$var.intercept)
     # level_combinations <- utils::combn(1:n_grps, m = n_grps - 1, simplify = FALSE)
-    # icc_grp <- sapply(level_combinations, function(v) vars$var.intercept[v[1]] / (vars$var.intercept[v[1]] + vars$var.intercept[v[2]]))
+    # icc_grp <- sapply(
+    #   level_combinations,
+    #   function(v) {
+    #     vars$var.intercept[v[1]] / (vars$var.intercept[v[1]] + vars$var.intercept[v[2]])
+    #   }
+    # )
     #
     # out2 <- data.frame(
     #   Group1 = group_names[sapply(level_combinations, function(i) i[1])],
@@ -273,11 +280,11 @@ icc <- function(model,
       # this is experimental!
       if (identical(ci_method, "analytical")) {
         result <- .safe(.analytical_icc_ci(model, ci))
-        if (!is.null(result)) {
+        if (is.null(result)) {
+          icc_ci_adjusted <- icc_ci_unadjusted <- NA
+        } else {
           icc_ci_adjusted <- result$ICC_adjusted
           icc_ci_unadjusted <- result$ICC_unadjusted
-        } else {
-          icc_ci_adjusted <- icc_ci_unadjusted <- NA
         }
       } else {
         result <- .bootstrap_icc(model, iterations, tolerance, ci_method, ...)
@@ -321,8 +328,6 @@ icc <- function(model,
 
 
 
-#' @param ... Arguments passed down to `brms::posterior_predict()`.
-#' @inheritParams icc
 #' @rdname icc
 #' @export
 variance_decomposition <- function(model,
@@ -428,7 +433,7 @@ print.icc <- function(x, digits = 3, ...) {
   }
 
   # separate lines for multiple R2
-  out <- paste0(out, collapse = "\n")
+  out <- paste(out, collapse = "\n")
 
   cat(out)
   cat("\n")
@@ -591,7 +596,11 @@ print.icc_decomposed <- function(x, digits = 2, ...) {
 .boot_icc_fun <- function(data, indices, model, tolerance) {
   d <- data[indices, ] # allows boot to select sample
   fit <- suppressWarnings(suppressMessages(stats::update(model, data = d)))
-  vars <- .compute_random_vars(fit, tolerance, verbose = FALSE)
+  vars <- .compute_random_vars(
+    fit,
+    tolerance,
+    verbose = isTRUE(getOption("easystats_errors", FALSE))
+  )
   if (is.null(vars) || all(is.na(vars))) {
     return(c(NA, NA))
   }
@@ -604,7 +613,11 @@ print.icc_decomposed <- function(x, digits = 2, ...) {
 
 # bootstrapping using "lme4::bootMer"
 .boot_icc_fun_lme4 <- function(model) {
-  vars <- .compute_random_vars(model, tolerance = 1e-05, verbose = FALSE)
+  vars <- .compute_random_vars(
+    model,
+    tolerance = 1e-10,
+    verbose = isTRUE(getOption("easystats_errors", FALSE))
+  )
   if (is.null(vars) || all(is.na(vars))) {
     return(c(NA, NA))
   }
@@ -685,10 +698,10 @@ print.icc_decomposed <- function(x, digits = 2, ...) {
   }
 
   model_rank <- tryCatch(
-    if (!is.null(model$rank)) {
-      model$rank - df_int
-    } else {
+    if (is.null(model$rank)) {
       insight::n_parameters(model) - df_int
+    } else {
+      model$rank - df_int
     },
     error = function(e) insight::n_parameters(model) - df_int
   )