support for quantile regression and mixed effects quantile regression

DESCRIPTION

@@ -1,7 +1,7 @@
 Type: Package
 Package: performance
 Title: Assessment of Regression Models Performance
-Version: 0.10.8.14
+Version: 0.10.8.15
 Authors@R: 
     c(person(given = "Daniel",
              family = "Lüdecke",
@@ -128,6 +128,7 @@ Suggests:
     patchwork,
     pscl,
     psych,
+    quantreg,
     qqplotr (>= 0.0.6),
     randomForest,
     rempsyc,

NAMESPACE

@@ -89,6 +89,9 @@ S3method(check_outliers,metagen)
 S3method(check_outliers,numeric)
 S3method(check_outliers,rma)
 S3method(check_outliers,rma.uni)
+S3method(check_outliers,rq)
+S3method(check_outliers,rqs)
+S3method(check_outliers,rqss)
 S3method(check_overdispersion,default)
 S3method(check_overdispersion,fixest)
 S3method(check_overdispersion,fixest_multi)

NEWS.md

@@ -42,6 +42,8 @@
 * Fixed edge cases in `check_collinearity()` and `check_outliers()` for models
   with response variables of classes `Date`, `POSIXct`, `POSIXlt` or `difftime`.
 
+* Fixed issue with `check_model()` for models of package *quantreg*.
+
 # performance 0.10.8
 
 ## Changes

R/check_outliers.R

@@ -310,7 +310,7 @@
 #' group_iris <- datawizard::data_group(iris, "Species")
 #' check_outliers(group_iris)
 #'
 #' @examplesIf require("see") && require("bigutilsr") && require("loo") && require("MASS") && require("ICSOutlier") && require("ICS") && require("dbscan")
 #' \donttest{
 #' # You can also run all the methods
 #' check_outliers(data, method = "all", verbose = FALSE)
@@ -1192,7 +1192,7 @@
 
   # Isolation Forest
   # if ("iforest" %in% method) {
   #   out <- c(out, .check_outliers_iforest(x, threshold = thresholds$iforest))
   # }
 
   # Local Outlier Factor
@@ -1576,7 +1576,7 @@
     out <- cbind(out, ID.names)
   }
 
   # out$Distance_IQR <- Distance_IQR
 
   out$Distance_IQR <- vapply(as.data.frame(t(Distance_IQR)), function(x) {
     ifelse(all(is.na(x)), NA_real_, max(x, na.rm = TRUE))
@@ -1758,7 +1758,7 @@
   }
 
   # check whether N to p ratio is not too large, else MCD flags too many outliers
   # See #672: This does seem to be a function of the N/p (N = sample size; p =
   # number of parameters) ratio. When it is larger than 10, the % of outliers
   # flagged is okay (in well behaved data). This makes sense: the MCD looks at
   # the cov matrix of subsamples of the data - with high dimensional data, small
@@ -1899,21 +1899,21 @@
 
 
 # .check_outliers_iforest <- function(x, threshold = 0.025) {
 #   out <- data.frame(Row = seq_len(nrow(x)))
 #
 #   # Install packages
 #  insight::check_if_installed("solitude")
 #
 #   # Compute
 #   if (utils::packageVersion("solitude") < "0.2.0") {
 #     iforest <- solitude::isolationForest(x)
 #     out$Distance_iforest <- stats::predict(iforest, x, type = "anomaly_score")
 #   } else if (utils::packageVersion("solitude") == "0.2.0") {
 #     stop(paste("Must update package `solitude` (above version 0.2.0).",
 #                "Please run `install.packages('solitude')`."), call. = FALSE)
 #   } else {
 #     iforest <- solitude::isolationForest$new(sample_size = nrow(x))
 #     suppressMessages(iforest$fit(x))
 #     out$Distance_iforest <- iforest$scores$anomaly_score
 #   }
 #
@@ -1987,3 +1987,12 @@
 
 #' @export
 check_outliers.glmrob <- check_outliers.glmmTMB
+
+#' @export
+check_outliers.rq <- check_outliers.glmmTMB
+
+#' @export
+check_outliers.rqs <- check_outliers.glmmTMB
+
+#' @export
+check_outliers.rqss <- check_outliers.glmmTMB

R/model_performance.bayesian.R

@@ -40,7 +40,7 @@
 #'
 #'   - **PCP**: percentage of correct predictions, see [performance_pcp()].
 #'
-#' @examplesIf require("rstanarm") && require("rstantools") && require("BayesFactor")
+#' @examplesIf require("rstanarm") && require("rstantools")
 #' \donttest{
 #' model <- suppressWarnings(rstanarm::stan_glm(
 #'   mpg ~ wt + cyl,
@@ -59,12 +59,6 @@
 #'   refresh = 0
 #' ))
 #' model_performance(model)
-#'
-#' model <- BayesFactor::generalTestBF(carb ~ am + mpg, mtcars)
-#'
-#' model_performance(model)
-#' model_performance(model[3])
-#' model_performance(model, average = TRUE)
 #' }
 #' @seealso [r2_bayes]
 #' @references Gelman, A., Goodrich, B., Gabry, J., and Vehtari, A. (2018).

R/r2_bayes.R

@@ -30,7 +30,7 @@
 #'   `r2_posterior()` is the actual workhorse for `r2_bayes()` and
 #'   returns a posterior sample of Bayesian R2 values.
 #'
-#' @examplesIf require("rstanarm") && require("rstantools") && require("BayesFactor") && require("brms")
+#' @examplesIf require("rstanarm") && require("rstantools") && require("brms")
 #' library(performance)
 #' \donttest{
 #' model <- suppressWarnings(rstanarm::stan_glm(
@@ -53,24 +53,6 @@
 #' r2_bayes(model)
 #' }
 #'
-#' BFM <- BayesFactor::generalTestBF(mpg ~ qsec + gear, data = mtcars, progress = FALSE)
-#' FM <- BayesFactor::lmBF(mpg ~ qsec + gear, data = mtcars)
-#'
-#' r2_bayes(FM)
-#' r2_bayes(BFM[3])
-#' r2_bayes(BFM, average = TRUE) # across all models
-#'
-#' # with random effects:
-#' mtcars$gear <- factor(mtcars$gear)
-#' model <- BayesFactor::lmBF(
-#'   mpg ~ hp + cyl + gear + gear:wt,
-#'   mtcars,
-#'   progress = FALSE,
-#'   whichRandom = c("gear", "gear:wt")
-#' )
-#'
-#' r2_bayes(model)
-#'
 #' \donttest{
 #' model <- suppressWarnings(brms::brm(
 #'   mpg ~ wt + cyl,

tests/testthat/test-check_model.R

@@ -43,3 +43,11 @@ test_that("`check_model()` for invalid models", {
   m1 <- lm(y ~ 1, data = dd)
   expect_error(check_model(m1))
 })
+
+test_that("`check_model()` works for quantreg", {
+  skip_if_not_installed("quantreg")
+  data(engel, package = "quantreg")
+  qm <- quantreg::rq(foodexp ~ income, data = engel)
+  x <- check_model(qm, verbose = FALSE)
+  expect_s3_class(x, "check_model")
+})

vignettes/r2.Rmd

@@ -29,14 +29,11 @@ knitr::opts_chunk$set(
 )
 options(digits = 2)
 
-pkgs <- c(
-  "effectsize", "BayesFactor", "lme4", "rstanarm"
-)
+pkgs <- c("effectsize", "lme4", "rstanarm")
 successfully_loaded <- sapply(pkgs, requireNamespace, quietly = TRUE)
 if (all(successfully_loaded)) {
   library(performance)
   library(effectsize)
-  library(BayesFactor)
   library(lme4)
   library(rstanarm)
 }
@@ -147,27 +144,6 @@ model <- stan_lmer(Petal.Length ~ Petal.Width + (1 | Species), data = iris, refr
 r2(model)
 ```
 
-Let's look at another regression analysis carried out with `{BayesFactor}` package.
-
-```{r, eval=successfully_loaded["BayesFactor"] && utils::packageVersion("BayesFactor") >= package_version("0.9.12-4.3")}
-library(BayesFactor)
-data(puzzles)
-
-m1 <- anovaBF(extra ~ group + ID,
-  data = sleep,
-  whichRandom = "ID", progress = FALSE
-)
-
-r2(m1)
-
-m2 <- generalTestBF(RT ~ shape * color + ID,
-  data = puzzles, whichRandom = "ID",
-  neverExclude = "ID", progress = FALSE
-)
-
-r2(m2)
-```
-
 # Comparing change in R2 using Cohen's *f*
 
 Cohen's $f$ (of [ANOVA fame](https://easystats.github.io/effectsize/articles/anovaES.html)) can be used as a measure of effect size in the context of sequential multiple regression (i.e., [**nested models**](https://easystats.github.io/performance/reference/test_performance.html)).