R code for paper Precision of Estimates of Nonresponse Bias in Means by Eckman, Unangst, Dever, and Antoun

All files also in OSF code repository

Published paper

Run code in this order (see master.R):

1_sims.R

Generates the populations and sample draws for the simulation study
Calls in 1.1_sims_continuous.R, 1.2_sims_binomial.R, and 1.3_sims_poisson.R. Each sims program (1.1, 1.2, 1.3) sets the parameters for the simulated populations. The corresponding functions programs (1.1_functions_cntns.R, 1.2_functions_binom.R, 1.3_functions_poisson.R) generate the populations, draw the samples, and calculate the standard error estimates for the nonresponse bias estimate from each sample draw

2_process.R

Processes and restructures the results from the simulation study to prepare for program 4_results.R
Calls in 2.1_rq1_datasets.R and 2.2_rq2_datasets.R

3_examples.R

Processes and produces estimates of nonresponse bias for GSS and LISS datasets 
Calls in 3.1_prepGSS.R and 3.2_prepLISS.R

4_results.R

Produces the graphs presented in the paper 

for_paper.Rmd -- numbers mentioned in paper

Subfolder structure -- programs expect these subfolders: \normal
\binomial
\poisson
\outputs\ -- graphs will be saved here \example_data\ -- should contain LISS and GSS data must be downloaded separately files needed: GSS.dat GSS.dct GSS_panel2010w123.dta jq13b_EN_2.0p.dta jf13a_EN_1.0p.dta see 3.1 and 3.2 R code for download information

0_setup.R is called by other programs

edit this file carefully, especially:

num.cores -- number of CPUs machine has. 16 cores will allow simultaneous processing of 16 populations

full or test sample sizes -- comment out test sizes if you want full sizes

full sizes reported in paper, but can take > 24 hours to run, depending on number of cores

Final run of code used 28 core machine and took ~30 hours to run the entire set of programs