fdabl

var sys  = require('sys');
var http = require('http');

function log(options) {
  var info = [options.method, options.path].join(' ');
  sys.log(info);
}

function proxy(options, req, res) {
  return http.request(options, function(proxy_res) {

fdabl

library('pwr') # for power calculations
library('foreign') # for dreadful spss .sav files
library('psych') # required by phack
source('http://rynesherman.com/phack.r')

SIM <- 10000
set.seed(42)
options(warn=-1) # avoid warning by read.spss
############################################
# Simulating p-hacking

fdabl

# models from p. 146 of
# http://www.uni-tuebingen.de//uni/sii/pw/heller/statistics_III_ws14/stat5www.pdf

# c(model1, model2, model3)
aic <- c(8128.519, 8249.935, 8163.826)
llh <- c(-4039.260, -4118.967, -4074.913)
params <- c(25, 6, 7)

# compute the relative distance of each model AIC to the
# model with the minimum AIC

fdabl

library('coda')
library('BayesFactor')
set.seed(1774) # Laplace easter egg :)


contains_zero <- function(interval) {
    return(interval[1] < 0 && interval[2] > 0)
}

fdabl

######
# Homework on Chi² Tests / Loglinear Models on 2x2 tables
# 
# instead of computing it by hand or in R, I wrote
# it in Julia, an awesome new language for technical
# computing. I discovered it yesterday, and it is
# truly intriguing. Do check it out! http://julialang.org/
######

# imports; functions in those modules are global now

fdabl

library('BayesFactor')
# Bayesian version of http://rynesherman.com/phack.r
# see also http://osc.centerforopenscience.org/2014/07/02/phack/

bhack <- function(n = 30, hackrate = 10, m1 = 0, m2 = 0, sd1 = 1, sd2 = 1,
                  max_n = 100, bf_cutoff = 3, sims = 1000, graph = TRUE) {

    out <- matrix(NA, nrow = sims, ncol = 4)
    for (i in 1:sims) {
        hackcount <- 0

fdabl

require(rvest)
require(dplyr)


scrape <- function(year, top) {
  possible <- c(10, 25, 100)
  if (!top %in% possible) stop("top must be 10, 25 or 100!")
  if (!year %in% 2012:2014) stop("data available only for years 2012 - 2014")
  
  targets <- top[1:which(possible == top)]

fdabl

/*
 * empirically tests the complexity of a bogosort based algorithm which
 * shuffles circles until they don't overlap (for stimulus presentation)
 * is a function of the circle radius, the canvas width and height
 *
 * call it with (first arg is radius, then width and height of canvas)
 * > node test.js 10 300 500
 * > node test.js 10
 * etc.
 */

fdabl

library('shiny')
# read this: http://alexanderetz.com/2015/04/15/understanding-bayes-a-look-at-the-likelihood/


shinyApp(
  ui = shinyUI(fluidPage(
    sidebarLayout(
      sidebarPanel(
        sliderInput("p1", label = "p1",
                    min = 0, max = 1, value = 0.7, step = 0.01),

fdabl

#' False Discovery Rate (FDR)
#' see also http://www.sciencedirect.com/science/article/pii/S1053811901910377
#'
#'          Declared active  Declared inactive
#' Active       V_aa               V_ai       T_a
#' Inactive     V_ia               V_ii       T_i
#'              D_a                D_i         V
#'
#' instead of controlling the overall family wise error, we control
#' the proportion of false discoveries; that is, we want (V_aa) / (V_aa + V_ia)