m-Py

# Get % of variance explained by Principal Component Analysis

library(psych)

pca_variance_explained <- function(data, n_components) {
  pca <- psych::principal(data, n_components, rotate = "none")
  list(
    by_variable = colSums(cor(pca$scores, data)^2),
    total = summary(prcomp(data))$importance["Cumulative Proportion", paste0("PC", n_components)]
  )

m-Py

# compare Cohen's h and phi coefficient as effect sizes for comparing proportions

library(effectsize)

# test data (have to be contingency tables; one column is always the same)
matrices <- lapply(1:999, function(i) matrix(c(i, 1000-i, 999, 1), ncol = 2))
phis <- sapply(matrices, function(x) effectsize::phi(x)$phi)
hs <- sapply(matrices, function(x) effectsize::cohens_h(x)$Cohens_h)

plot(abs(hs), phis, type = "l")

m-Py

# Test if splitting data via anticlustering leads to closer groups means to the *true* population means,
# as compared to a random split (e.g., for cross validation
simulate <- function(N = 100, split = c(1, 3) / 4) { # default: split 75/25
  data <- rnorm(N)
  groups <- anticlustering(
    data,
    K = round(N * split),
    objective = "variance"
  )
  c(

m-Py

## 1. Load - and, if required, install - package `anticlust`

if (!requireNamespace("remotes")) {
  install.packages("remotes")
}
remotes::install_github("m-Py/anticlust")

library(anticlust)

m-Py

# Show that interaction in glm() changes nature of main effect
# (only if a categorical predictor is dummy coded - not contrast coded)

# Returns the p-value associated with a predictor main effect, once
# with and once without interaction with a (non-predictive) categorical
# independent variable

simulate_glm <- function(N = 100, contrast_coding = FALSE) {
  iv1 <- rnorm(N) # related to DV

m-Py

# Author: Martin Papenberg
# Year: 2019

# Perform fast KNN classifier using RANN for nearest neighbour search

library("RANN")
library("data.table")

# param data: The numeric data matrix used 
# param labels: the labels to predict

m-Py

## This document illustrates that type 1 sum of squares lead to increased alpha
## error rates when a predictive covariate is included in the regression model. 


# Estimate p-value for treatment (null) effect via linear regression, 
# including a covariate that is predictive of the outcome
#
# param N: sample size, default 100

m-Py

## Year 2019 - 2020
## Author: Martin Papenberg

## This code is in the public domain, do with it whatever you like.
# Generate bivariate normal data with specified correlation

# param n: how many data points
# param mx: the mean of the first variable
# param my: the mean of the second variable
# param sdx: the standard deviation of the first variable

m-Py

## Warning: This code is just for fun / educational purposes; the file contains functions 
## to find out how severely the p value in a t-test can be minimized by systematic removal of data points.

## SIX OUT OF THIRTY - Martin's approach
## Based on @juli_tkotz's (https://twitter.com/juli_tkotz/status/1085446224117985281) 
## idea that removing from the most extreme values is the best apporach.


#' Simulate t-tests and store best p values 
#' 

m-Py

## Author Martin Papenberg
## Year 2018 

## This code is released into the public domain. Anybody may use, alter
## and distribute the code without restriction. The author makes no
## guarantees, and takes no liability of any kind for use of this code.

#' Compute ordinal scores from continuous data
#'
#' Might be useful for data exploration with highly skewed data