jepusto’s gists

jepusto / simulation-in-R-2016

Created September 28, 2016 18:16

Rmd for my presentation on simulation studies in R, Quant Methods brownbag colloquium 2016/09/28

	---
	title: "Designing simulation studies in R"
	author: "James E. Pustejovsky"
	date: "September 28, 2016"
	output:
	ioslides_presentation:
	css: custom.css
	widescreen: true
	transition: faster
	---

jepusto / bug-in-nlme::getVarCov.R

Last active August 11, 2016 14:37


	# Demonstrate the problem with gls model

	library(nlme)
	data(Ovary)

	gls_raw <- gls(follicles ~ sin(2piTime) + cos(2piTime), data = Ovary,
	correlation = corAR1(form = ~ 1 \| Mare),
	weights = varPower())

jepusto / Fatal crashes by city, 2006-2015

Created August 24, 2015 02:17

	library(tidyr)
	library(dplyr)
	library(stringr)
	library(ggplot2)

	cities_select <- c("HOUSTON","SAN ANTONIO","DALLAS","AUSTIN","FORT WORTH","EL PASO")

	#----------------------------------------
	# get population estimates
	#----------------------------------------

jepusto / Austin crashes - annual data figures

Last active August 29, 2015 14:27

Code for figures displaying annual number of fatal automobile crashes, fatalities, etc. in Austin and Travis County, 2003-2015

	library(tidyr)
	library(dplyr)
	library(stringr)
	library(ggplot2)

	#--------------------------------
	# format the data for graphing
	#--------------------------------

	crash_dat <- read.csv("http://blogs.edb.utexas.edu/pusto/files/2015/08/Yearly_crash_data_Austin_and_Travis_County.csv")

jepusto / metafor-BRL.R

Created April 25, 2014 20:33

Bias-reduced linearization covariance estimator and degrees of freedom for multi-variate meta-analysis models

	require(Formula)
	require(metafor)
	require(sandwich)
	require(zoo)
	require(lmtest)

	#-----------------------------------------------
	# Identify outer-most clustering variable
	#-----------------------------------------------

jepusto / RVE-correlated.R

Created April 21, 2014 16:10

Correlated effects example from Tanner-Smith & Tipton (2013)

	# robumeta calculations
	library(grid)
	library(robumeta)
	data(corrdat)
	rho <- 0.8

	HTJ <- robu(effectsize ~ males + college + binge,
	data = corrdat,
	modelweights = "CORR", rho = rho,
	studynum = studyid,

jepusto / metafor-sandwich.R

Created April 21, 2014 14:16

Calculate sandwich covariance estimators for multi-variate meta-analysis models

	require(Formula)
	require(metafor)
	require(sandwich)
	require(zoo)
	require(lmtest)

	#-----------------------------------------------
	# Functions for making sandwich standard errors
	#-----------------------------------------------

jepusto / RVE-hierarchical.R

Last active August 29, 2015 14:00

Hierarchical effects example from Tanner-Smith & Tipton (2013)

	# robumeta calculations
	library(grid)
	library(robumeta)

	data(hierdat)

	HTJ <- robu(effectsize ~ males + binge,
	data = hierdat, modelweights = "HIER",
	studynum = studyid,
	var.eff.size = var, small = FALSE)

jepusto / Simulate Welch t-test in parallel

Last active December 31, 2015 23:28

An adaptation of the Welch t-test simulation, for running in parallel on the Stampede server of the Texas Advanced Computing Cluster.

	#----------------------------------------------
	# data-generating model
	#----------------------------------------------

	two_group_data <- function(iterations, n, p, var_ratio, delta) {
	Group <- c(rep("C", n * p), rep("T", n * (1 - p)))
	Y_C <- matrix(rnorm(iterations * n * p, mean = 0, sd = 1), n * p, iterations)
	Y_T <- matrix(rnorm(iterations * n * (1 - p), mean = delta, sd = sqrt(var_ratio)), n * (1 - p), iterations)
	dat <- data.frame(Group, rbind(Y_C, Y_T))
	return(dat)

jepusto / Simulate Welch t-test

Last active December 29, 2015 14:49

A simulation evaluating the confidence interval coverage for a difference in means, allowing for unequal variances, based on Welch's approximation for the degrees of freedom. The code demonstrates the basic structure of a simulation study in R.

	#----------------------------------------------
	# data-generating model
	#----------------------------------------------

	two_group_data <- function(iterations, n, p, var_ratio, delta) {
	Group <- c(rep("C", n * p), rep("T", n * (1 - p)))
	Y_C <- matrix(rnorm(iterations * n * p, mean = 0, sd = 1), n * p, iterations)
	Y_T <- matrix(rnorm(iterations * n * (1 - p), mean = delta, sd = sqrt(var_ratio)), n * (1 - p), iterations)
	dat <- data.frame(Group, rbind(Y_C, Y_T))
	return(dat)

James E. Pustejovsky jepusto