Merge branch 'dev' into FindBestModel

DESCRIPTION

@@ -1,17 +1,19 @@
 Package: psycho
 Type: Package
 Title: Efficient and Publishing-Oriented Workflow for Psychological Science
-Version: 0.3.5
+Version: 0.3.6
 Authors@R: c(
     person("Dominique", 
         "Makowski", 
         email = "[email protected]",
         role = c("aut", "cre", "cph"), 
         comment = c(ORCID = "0000-0001-5375-9967")),
+    person("Hugo",
+        "Najberg",
+        role = "ctb"),
     person("Viliam",
         "Simko",
-        role = "ctb",
-        email = "[email protected]"),
+        role = "ctb"),
     person("Sasha", 
         "Epskamp", 
         role = "rev",
@@ -26,7 +28,7 @@ Description: The main goal of the psycho package is to provide tools for psychol
 License: MIT + file LICENSE
 Encoding: UTF-8
 LazyData: true
-RoxygenNote: 6.0.1
+RoxygenNote: 6.1.0
 Depends: 
     R (>= 3.4.0)
 Imports:

NAMESPACE

@@ -12,7 +12,7 @@ S3method(analyze,lm)
 S3method(analyze,lmerModLmerTest)
 S3method(analyze,stanreg)
 S3method(as.data.frame,density)
-S3method(find_best_model,merModLmerTest)
+S3method(find_best_model,lmerModLmerTest)
 S3method(find_best_model,stanreg)
 S3method(find_combinations,formula)
 S3method(get_R2,glm)
@@ -22,6 +22,9 @@ S3method(get_R2,stanreg)
 S3method(get_contrasts,glmerMod)
 S3method(get_contrasts,lmerModLmerTest)
 S3method(get_contrasts,stanreg)
+S3method(get_data,lm)
+S3method(get_data,merMod)
+S3method(get_data,stanreg)
 S3method(get_formula,glm)
 S3method(get_formula,glmerMod)
 S3method(get_formula,lm)
@@ -50,6 +53,9 @@ S3method(standardize,numeric)
 S3method(standardize,stanreg)
 S3method(summary,psychobject)
 export(.fa_variance_text)
+export(HDI)
+export(HDImax)
+export(HDImin)
 export(R2_LOO_Adjusted)
 export(R2_nakagawa)
 export(R2_tjur)
@@ -82,7 +88,6 @@ export(get_formula)
 export(get_info)
 export(get_loadings_max)
 export(get_predicted)
-export(hdi)
 export(interpret_R2)
 export(interpret_R2_posterior)
 export(interpret_RMSEA)
@@ -115,6 +120,7 @@ export(remove_empty_cols)
 export(reorder_matrix)
 export(rnorm_perfect)
 export(rope)
+export(simulate_data_regression)
 export(standardize)
 export(values)
 import(broom)
@@ -173,6 +179,7 @@ importFrom(stats,dnorm)
 importFrom(stats,ecdf)
 importFrom(stats,family)
 importFrom(stats,formula)
+importFrom(stats,getCall)
 importFrom(stats,hclust)
 importFrom(stats,mad)
 importFrom(stats,median)
@@ -204,5 +211,6 @@ importFrom(stringr,str_trim)
 importFrom(tibble,rownames_to_column)
 importFrom(utils,capture.output)
 importFrom(utils,combn)
+importFrom(utils,data)
 importFrom(utils,head)
 importFrom(utils,tail)

NEWS.md

@@ -1,8 +1,10 @@
 # Current Dev
 
 ### Breaking changes
+- Changed `hdi` to `HDI`
 - Removed Mean and SD from summary in `bayes_cor.test`
 ### New functions / parameters
+- Added `simulate_data_regression`
 - Extended `standardize` method
 - Added `get_R2` method
 - Added `interpret_odds` and logistic regression effect size interpretation
@@ -163,7 +165,7 @@
 
 ### Breaking changes
 ### New functions / parameters
-- `hdi`: Compute highest density intervals
+- `HDI`: Compute highest density intervals
 - `format_string`: A tidyverse friendly version of `sprintf`
 ### Major changes
 - Changed credible interval computation in analyze.stanreg

R/analyze.htest.R

@@ -48,6 +48,8 @@ analyze.htest <- function(x, effsize_rules="cohen1988", ...) {
 
   # Text
   # -------------
+
+  # CORRELATION
   if (grepl("correlation", values$method)) {
     text <- paste0(
       "The ",
@@ -68,6 +70,8 @@ analyze.htest <- function(x, effsize_rules="cohen1988", ...) {
       format_p(values$p, stars = FALSE),
       ")."
     )
+
+  # T-TEST
   } else if (grepl("t-test", values$method)) {
     if (names(x$null.value) == "mean") {
       means <- paste0(
@@ -93,10 +97,6 @@ analyze.htest <- function(x, effsize_rules="cohen1988", ...) {
       vars <- paste0(values$names, " (", means, ")")
     }
 
-
-
-
-
     values$effect <- values$effect[1] - values$effect[2]
 
     text <- paste0(
@@ -117,6 +117,7 @@ analyze.htest <- function(x, effsize_rules="cohen1988", ...) {
       format_p(values$p, stars = FALSE),
       ")."
     )
+  # OTHER
   } else {
     stop(paste0("The ", values$method, " is not implemented yet."))
   }

R/analyze.lm.R

@@ -50,7 +50,7 @@ analyze.lm <- function(x, CI=95, effsize_rules="cohen1988", ...) {
   summary$p <- summary$`Pr...t..`
 
   # standardized coefficients
-  standardized <- tibble::rownames_to_column(standardize(fit, method = "refit"), "Variable")
+  standardized <- tibble::rownames_to_column(standardize(fit, method = "refit", data=data), "Variable")
   summary <- merge(summary, standardized, by = "Variable", all.x = TRUE, sort = FALSE)
   summary$Effect_Size <- c(NA, interpret_d(tail(summary$Coef_std, -1), rules = effsize_rules))
 

R/analyze.stanreg.R

@@ -4,6 +4,9 @@
 #'
 #' @param x A stanreg model.
 #' @param CI Credible interval bounds.
+#' @param index Index of effect existence to report. Can be 'overlap' or 'ROPE'.
+#' @param ROPE_bounds Bounds ot the ROPE. If NULL and effsize is TRUE, than the ROPE.
+#' will have default values c(-0.1, 0.1) and computed on the standardized posteriors.
 #' @param effsize Compute Effect Sizes according to Cohen (1988). For linear models only.
 #' @param effsize_rules Grid for effect size interpretation. See \link[=interpret_d]{interpret_d}.
 #' @param ... Arguments passed to or from other methods.
@@ -15,6 +18,7 @@
 #'  \item{the Credible Interval (CI) (by default, the 90\% CI; see Kruschke, 2018), representing a range of possible parameter.}
 #'  \item{the Maximum Probability of Effect (MPE), the probability that the effect is positive or negative (depending on the median’s direction).}
 #'  \item{the Overlap (O), the percentage of overlap between the posterior distribution and a normal distribution of mean 0 and same SD than the posterior. Can be interpreted as the probability that a value from the posterior distribution comes from a null distribution.}
+#'  \item{the ROPE, the proportion of the 95\% CI of the posterior distribution that lies within the region of practical equivalence.}
 #'  }
 #'
 #' @examples
@@ -62,7 +66,7 @@
 #' @importFrom broom tidy
 #' @importFrom stringr str_squish str_replace
 #' @export
-analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988", ...) {
+analyze.stanreg <- function(x, CI=90, index="overlap", ROPE_bounds=NULL, effsize=FALSE, effsize_rules="cohen1988", ...) {
   fit <- x
 
   # Info --------------------------------------------------------------------
@@ -164,7 +168,9 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
         CI = CI,
         effsize = effsize,
         effsize_rules = effsize_rules,
-        fit = fit
+        fit = fit,
+        index=index,
+        ROPE_bounds=ROPE_bounds
       )
     }
   }
@@ -377,7 +383,7 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
   values$mad <- mad(posterior)
   values$mean <- mean(posterior)
   values$sd <- sd(posterior)
-  values$CI_values <- hdi(posterior, prob = CI / 100)
+  values$CI_values <- HDI(posterior, prob = CI / 100)
   values$CI_values <- c(values$CI_values$values$HDImin, values$CI_values$values$HDImax)
   values$MPE <- NA
   values$MPE_values <- NA
@@ -454,7 +460,7 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
   values$mad <- mad(posterior)
   values$mean <- mean(posterior)
   values$sd <- sd(posterior)
-  values$CI_values <- hdi(posterior, prob = CI / 100)
+  values$CI_values <- HDI(posterior, prob = CI / 100)
   values$CI_values <- c(values$CI_values$values$HDImin, values$CI_values$values$HDImax)
   values$MPE <- NA
   values$MPE_values <- NA
@@ -507,7 +513,18 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
 
 
 #' @keywords internal
-.process_effect <- function(varname, posteriors, posteriors_std, info_priors, predictors, CI=90, effsize=FALSE, effsize_rules=FALSE, fit) {
+.process_effect <- function(varname,
+                            posteriors,
+                            posteriors_std,
+                            info_priors,
+                            predictors,
+                            CI=90,
+                            effsize=FALSE,
+                            effsize_rules=FALSE,
+                            fit,
+                            index="overlap",
+                            ROPE_bounds=NULL) {
+
   values <- .get_info_priors(varname, info_priors, predictors)
   posterior <- posteriors[, varname]
 
@@ -518,10 +535,12 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
   values$mad <- mad(posterior)
   values$mean <- mean(posterior)
   values$sd <- sd(posterior)
-  values$CI_values <- hdi(posterior, prob = CI / 100)
+  values$CI_values <- HDI(posterior, prob = CI / 100)
   values$CI_values <- c(values$CI_values$values$HDImin, values$CI_values$values$HDImax)
   values$MPE <- mpe(posterior)$MPE
   values$MPE_values <- mpe(posterior)$values
+
+  # Index
   values$overlap <- 100 * overlap(
     posterior,
     rnorm_perfect(
@@ -531,6 +550,48 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
     )
   )
 
+  if(!is.null(ROPE_bounds)){
+    rope <- rope(posterior, bounds=ROPE_bounds)
+    values$ROPE_decision <- rope$rope_decision
+    values$ROPE <- rope$rope_probability
+  }
+
+  if(index == "overlap"){
+    index <- paste0("Overlap = ",
+                    format_digit(values$overlap, null_treshold = 0.01),
+                    "%).")
+  } else if(index == "ROPE"){
+    if(!is.null(ROPE_bounds)){
+      index <- paste0("ROPE = ",
+                      format_digit(values$ROPE, null_treshold = 0.001),
+                      ").")
+    } else{
+      if(effsize == TRUE){
+        rope <- rope(posteriors_std[, varname], bounds=c(-0.1, 0.1))
+        values$ROPE_decision <- rope$rope_decision
+        values$ROPE <- rope$rope_probability
+        index <- paste0("ROPE = ",
+                        format_digit(values$ROPE, null_treshold = 0.001),
+                        ").")
+      } else{
+        warning("you need to specify ROPE_bounds (e.g. 'c(-0.1, 0.1)'). Computing overlap instead.")
+        index <- paste0("Overlap = ",
+                        format_digit(values$overlap, null_treshold = 0.01),
+                        "%).")
+      }
+    }
+
+  } else{
+    warning("Parameter 'index' should be 'overlap' or 'ROPE'. Computing overlap.")
+    index <- paste0("Overlap = ",
+                    format_digit(values$overlap, null_treshold = 0.01),
+                    "%).")
+  }
+
+
+
+
+
   # Text
   if (grepl(":", varname)) {
     splitted <- strsplit(varname, ":")[[1]]
@@ -564,9 +625,7 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
     "% CI [",
     format_digit(values$CI_values[1], null_treshold = 0.0001), ", ",
     format_digit(values$CI_values[2], null_treshold = 0.0001), "], ",
-    "Overlap = ",
-    format_digit(values$overlap, null_treshold = 0.01),
-    "%)."
+    index
   )
 
 
@@ -579,7 +638,7 @@ analyze.stanreg <- function(x, CI=90, effsize=FALSE, effsize_rules="cohen1988",
     values$std_mad <- mad(posterior_std)
     values$std_mean <- mean(posterior_std)
     values$std_sd <- sd(posterior_std)
-    values$std_CI_values <- hdi(posterior_std, prob = CI / 100)
+    values$std_CI_values <- HDI(posterior_std, prob = CI / 100)
     values$std_CI_values <- c(values$std_CI_values$values$HDImin, values$std_CI_values$values$HDImax)
 
     if (fit$family$family == "binomial" & fit$family$link == "logit") {

R/bayes_cor.R

@@ -69,8 +69,8 @@ bayes_cor.test <- function(x, y, CI=90, iterations = 10000, effsize_rules_r="coh
   values$mad <- mad(posterior)
   values$mean <- mean(posterior)
   values$sd <- sd(posterior)
-  values$CI <- hdi(posterior, prob = CI / 100)$text
-  values$CI_values <- hdi(posterior, prob = CI / 100)
+  values$CI <- HDI(posterior, prob = CI / 100)$text
+  values$CI_values <- HDI(posterior, prob = CI / 100)
   values$CI_values <- c(values$CI_values$values$HDImin, values$CI_values$values$HDImax)
   values$MPE <- mpe(posterior)$MPE
   values$MPE_values <- mpe(posterior)$values

R/crawford.test.R

@@ -95,7 +95,7 @@ crawford.test <- function(patient,
 
   pvalues <- pvalues / 2
   p <- mean(pvalues)
-  CI <- hdi(pvalues, prob = CI / 100)
+  CI <- HDI(pvalues, prob = CI / 100)
   # CI_1 <- sort(pvalues)[iter * (100 - CI) / 100]
 
 

R/find_best_model.R

@@ -4,7 +4,7 @@
 #' documentation for your model's class:
 #' \itemize{
 #'  \item{\link[=find_best_model.stanreg]{find_best_model.stanreg}}
-#'  \item{\link[=find_best_model.merModLmerTest]{find_best_model.merModLmerTest}}
+#'  \item{\link[=find_best_model.lmerModLmerTest]{find_best_model.lmerModLmerTest}}
 #'  }
 #'
 #' @param fit Model

R/find_best_model.lmerModLmerTest.R

@@ -79,4 +79,4 @@ find_best_model.lmerModLmerTest <- function(fit, interaction=TRUE, fixed=NULL, .
 
   best <- list(formula = best, by_criterion = by_criterion, table = comparison)
   return(best)
-}
+}