Merge pull request #51 from neuropsychology/dev

0.2.3 - CRAN

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: psycho
 Type: Package
 Title: Efficient and Publishing-Oriented Workflow for Psychological Science
-Version: 0.2.2
+Version: 0.2.3
 Authors@R: c(
     person("Dominique", 
         "Makowski", 

NAMESPACE

@@ -103,10 +103,12 @@ importFrom(stats,qnorm)
 importFrom(stats,quantile)
 importFrom(stats,rchisq)
 importFrom(stats,rnorm)
+importFrom(stats,runif)
 importFrom(stats,sd)
 importFrom(stats,terms)
 importFrom(stats,update)
 importFrom(stats,var)
+importFrom(stringr,str_squish)
 importFrom(stringr,str_to_title)
 importFrom(tibble,rownames_to_column)
 importFrom(utils,combn)

NEWS.md

@@ -8,6 +8,7 @@
 - `crawford.test` now computes the Bayesian version
 - `crawford.test` has been renamed to `crawford.test.freq`
 ### New functions / parameters
+- Added `i_am_cheating` parameter to `correlation` to prevent p-hacking
 - Added `percentile` and `percentile_to_z` functions
 - Added `as.data.frame` method for density objects
 - Added Crawford-Garthwaite (2007) Bayesian test for single-case analysis

R/assess.R

@@ -54,8 +54,8 @@ assess <- function(patient, mean = 0, sd = 1, n = NULL, controls = NULL, CI=95,
       warning("Multiple scores were provided. Returning a list of results.")
     }
     results <- list()
-    for (i in patient) {
-      results[i] <- crawford.test(patient, controls, mean, sd, n, CI, treshold, iter, color_controls, color_CI, color_score, color_size, alpha_controls, alpha_CI)
+    for (i in 1:length(patient)) {
+      results[[i]] <- crawford.test(patient[i], controls, mean, sd, n, CI, treshold, iter, color_controls, color_CI, color_score, color_size, alpha_controls, alpha_CI)
       return(results)
     }
   } else {

R/correlation.R

@@ -17,6 +17,7 @@
 #'   "hochberg", "hommel", "bonferroni", "BH", "BY", "fdr", "none"). See
 #'   \link[stats]{p.adjust} for details about why to use "holm" rather than
 #'   "bonferroni").
+#' @param i_am_cheating Set to TRUE to run many uncorrected correlations.
 #'
 #' @return output
 #'
@@ -37,18 +38,20 @@
 #'
 #' @author \href{https://dominiquemakowski.github.io/}{Dominique Makowski}
 #'
-#' @importFrom stats na.omit p.adjust cor
+#' @importFrom stats na.omit p.adjust cor runif
 #' @importFrom psych corr.test
 #' @importFrom ggplot2 theme element_text
 #' @importFrom stringr str_to_title
 #' @import ggcorrplot
 #' @import ppcor
+#' @import dplyr
 #' @export
 correlation <- function(df,
                         df2 = NULL,
                         type = "full",
                         method = "pearson",
-                        adjust = "holm") {
+                        adjust = "holm",
+                        i_am_cheating = FALSE) {
 
   # Processing
   # -------------------
@@ -60,6 +63,21 @@ correlation <- function(df,
   if (is.null(df2) == FALSE) {
     df2 <- df2[, sapply(df2, is.numeric)]
   }
+
+  # P-fishing prevention
+  if (ncol(df) > 10 && adjust == "none" && i_am_cheating == FALSE) {
+    warning("We've detected that you are running a lot (> 10) of correlation tests without adjusting the p values. To help you in your p-fishing, we've added some interesting variables: You never know, you might find something significant!\nTo deactivate this, change the 'i_am_cheating' argument to TRUE.")
+    df_complete <- dplyr::mutate_all(df, dplyr::funs_("replace(., is.na(.), 0)"))
+    df$Local_Air_Density <- svd(df_complete)$u[, 1]
+    df$Reincarnation_Cycle <- runif(nrow(df), max = 100)
+    df$Communism_Level <- -1 * svd(df_complete)$u[, 2]
+    df$Alien_Mothership_Distance <- rnorm(nrow(df), mean = 50000, sd = 5000)
+    df$Schopenhauers_Optimism <- svd(df_complete)$u[, 3]
+    df$Hulks_Power <- runif(nrow(df), max = 10)
+  }
+
+
+
   # Compute r coefficients
   if (type == "full") {
     corr <- psych::corr.test(df, y = df2, use = "pairwise", method = method, adjust = "none")

docs/_posts/preparation/correlation.Rmd

@@ -0,0 +1,107 @@
+---
+title: "Formatted and Flexible Correlation Tables"
+layout: post
+output:
+  html_document:
+    df_print: paged
+    toc: yes
+  md_document:
+    toc: yes
+    variant: markdown_github
+author: "Dominique Makowski"
+date: "`r Sys.Date()`"
+tags:
+- r
+- psychology
+- neuroscience
+- neuropsychology
+---
+
+
+```{r message=FALSE, warning=FALSE, include=FALSE}
+library(knitr)
+```
+
+# Another *correlation* function?!
+
+Yes, the `correlation` function from the [`psycho`](https://github.com/neuropsychology/psycho.R) package.
+
+```{r fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='hide'}
+library(psycho)
+library(tidyverse)
+
+cor <- psycho::affective %>% 
+  correlation()
+```
+
+This function automatically select numeric variables and run a correlation analysis. It returns a `psychobject`.
+
+
+# A table
+
+We can then extract a **formatted table** that can be saved and pasted into reports and manuscripts.
+
+```{r fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='hide'}
+summary(cor)
+# write.csv(summary(cor), "myformattedcortable.csv")
+```
+```{r echo=FALSE, fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='markup'}
+knitr::kable(summary(cor))
+```
+
+# A Plot
+
+It integrates a **plot** done with [`ggcorplot`](https://github.com/kassambara/ggcorrplot).
+
+```{r, eval = TRUE, results='markup', fig.align='center'}
+plot(cor)
+```
+
+
+# A print
+
+It also includes a **pairwise correlation printing** method.
+
+```{r fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='markup'}
+print(cor)
+```
+
+# Options
+
+You can also cutomize the **type** (pearson, spearman or kendall), the **p value correction method** (holm (default), bonferroni, fdr, none...) and run **partial, semi-partial or glasso** correlations.
+
+
+```{r fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='hide'}
+psycho::affective %>% 
+  correlation(method = "pearson", adjust="bonferroni", type="partial") %>% 
+  summary()
+```
+```{r echo=FALSE, fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='markup'}
+psycho::affective %>% 
+  correlation(method = "pearson", adjust="bonferroni", type="partial") %>% 
+  summary() %>% 
+  knitr::kable()
+```
+
+# Fun with *p*-hacking
+
+In order to **prevent people for running many uncorrected correlation tests** (promoting *p*-hacking and result-fishing), we included the `i_am_cheating` parameter. If FALSE (default), the function will help you finding interesting results!
+
+```{r, eval = TRUE, results='hide', fig.align='center'}
+df_with_11_vars <- data.frame(replicate(11, rnorm(1000)))
+cor <- correlation(df_with_11_vars, adjust="none") 
+summary(cor)
+```
+```{r echo=FALSE, fig.align='center', message=FALSE, warning=FALSE, comment=NA, results='markup'}
+knitr::kable(summary(cor)[,1:11])
+```
+
+
+**As we can see, Schopenhauer's Optimism is strongly related to many variables!!!**
+
+
+# Credits
+
+This package was useful? You can cite [`psycho`](https://github.com/neuropsychology/psycho.R) as follows:
+
+- Makowski, (2018). *The psycho Package: an Efficient and Publishing-Oriented Workflow for Psychological Science*. Journal of Open Source Software, 3(22), 470. https://doi.org/10.21105/joss.00470

docs/_posts/preparation/package_downloads.Rmd

@@ -50,12 +50,12 @@ blogposts_names <- list.files("../") %>%
 
 events <- data.frame(date = c("2018-01-10", "2018-02-05", "2018-03-21"),
                      label = c("JOSS Review", "JOSS Publication", "Blog Creation"),
-                     color = c("orange", "green", "blue")) %>% 
+                     color = c("#2196F3", "#4CAF50", "#FF9800")) %>% 
    mutate_all(as.character) %>% 
   bind_rows(
     data.frame(date = blogposts_dates,
                label = blogposts_names,
-               color = "red") %>% 
+               color = "#E91E63") %>% 
       mutate_all(as.character)
     ) %>% 
   mutate(date = lubridate::ymd(date)) %>% 

tests/testthat/test-assess.R

@@ -16,4 +16,13 @@ test_that("It works", {
   )
 
   testthat::expect_equal(x$values$p, 0.18, tol = 0.02)
+
+  x <- assess(
+    patient = c(10, 12),
+    mean = 8,
+    sd = 2,
+    verbose=FALSE
+  )
+
+  testthat::expect_equal(x[[1]]$values$p, 0.16, tol = 0.05)
 })

tests/testthat/test-odds_to_probs.R

@@ -3,4 +3,12 @@ context("odds_to_probs")
 test_that("Correct", {
   testthat::expect_equal(odds_to_probs(-1.6), 0.17, tolerance = 0.01)
   testthat::expect_equal(odds_to_probs(-1.6, log = F), 2.66, tolerance = 0.01)
+
+  testthat::expect_equal(
+    ncol(odds_to_probs(
+      psycho::affective,
+      subset=c("Life_Satisfaction"),
+      except=c("Sex"))),
+    8)
 })
+