Visualising MRP: Alternatives of The Current Practice

📋 About

This is a repository for my master degree research project about Multilevel Regression and Poststratification (MRP) visualisation. This study aims to:

1. Discuss the current practice of visualisation of MRP models.
2. Understand the implication of existing visualisation choices with real-world data.
3. Explore possible improvements of the current practice of MRP visualisation.

🔧 Method

According to those objectives, this study is performed through:

1. Systematic literature review on peer-reviewed articles applied MRP.
2. Case study (The application of MRP in estimating Trump's vote share in the 2016 U.S. presidential election.

📋 Data

The data used in the case study are not uploaded in this repository but they are publicly available and can be obtained with the following steps:

CCES data

The survey data used is the 2016 Cooperative Congressional Election Study (CCES). This data is publicly available in Harvards dataverse, but in this study it is accessed through an R-package called ccesMRPprep(Kuriwaki, 2021). The codes used to get the data is:

# remotes::install_github("kuriwaki/ccesMRPprep")
library(ccesMRPprep)

cces <- get_cces_dataverse("cumulative")

# filter to only have 2016 data
cces_2016 <- cces %>%
  filter(year == "2016") %>%
  ccc_std_demographics()

ACS data

This study uses 3-years of ACS data, i.e., 2015-2017 Public Use Microdata Sample (PUMS), which can be obtained through:

1. 2015 PUMS data, download the file named: csv_pus.zip
2. 2016 PUMS data, download the file named: csv_pus.zip
3. 2017 PUMS data, download the file named: csv_pus.zip

The complete codes to do data preparation is available in this file.

📁 File Structure

The file stucture of this repository is:

🗂️ case_study

This folder includes all of the files needed to reproduce the analysis in the case study (assume that the downloaded ACS data is saved in folder data/data_raw).

• analysis
  • analysis_code (folder contains codes used in the case study)
    • cces_acs_wrangling.R (data wrangling of CCES and ACS data)
    • mrp_build.Rmd (code used when prepare and fitting MRP models)
    • mrp_fitting.R (code used to prepare and specifiy MRP models, used in the report for Chapter 3)
    • mrp_vis.R (code used to visualise MRP estimatesm used in report for Chapter 3)
  • hpc_code (folder contains R codes and slurm submission for High Performance Cluster)
    • The results of model fitting in the HPC are saved in a folder called results.

🗂️ presentation

This folder contains slides for presenting the thesis (preliminary presentation and final presentation). Slides are made using Xaringan in rmarkdown.

🗂️ report

This folder contains the thesis/report. The full report (pdf version) is located in folder: "_book"

🗂️ sys_literature_review

This folder includes all of the files needed to reproduce the systematic literature review in this study.

• analysis (folder contains the analysis of extracted data from systematic literature review).
  • preliminary_result.Rmd (rough result of systematic literature review)
  • slr_analysis.R (code used in Chapter 2)
• endnote_lib
  • This folder contains endnote files to manage reviewed literatures.
• lit_review_documentation
  • Database_search.docx (documentation of paper identification in each research database)
  • inclusion_mismatch_notes.docx (documentation of screening mismatch by DA and LK)
• paper_metadata
  • Folder contains csv files of papers' metadata extracted from slr.

🔧 Packages used

If you want to reproduce this analysis, make sure you have these packages installed: mrpkit; ccesMRPprep ; brms; cmdstanr; survey; ddi; tidyverse; forcats; Metrics; data.table; kableExtra; janitor; scales; ggplot2; patchwork; flipPlots; igraph; urbnmapr; ggstance; ggpmisc; wacolors; rmarkdown; knitr; MonashEBSTemplates.