Portable Routines for Preparing CCES and ACS data for MRP
Cite as:
- Shiro Kuriwaki (2020). ccesMRPprep: Functions and Data to Prepare CCES data for MRP. R package. https://www.github.com/kuriwaki/ccesMRPprep
Purpose and Contribution
Multilevel Regression and Poststratification (MRP) is an increasingly popular method for analyzing surveys, and can be implemented on public datasets such as the CCES and ACS. Several helpful tutorials give introductions with sample R code (Kastellec, Lax, and Phillips, 2019; Hanretty, 2019),
But despite its increasingly popularity, doing one’s own MRP entails
considerable upfront costs: downloading the appropriate survey and
contextual data, recoding survey values to match with their Census
counterparts, and generating population frames to post-stratify on,
potentially by merging different datasets. While there already exist
some packages for MRP
(e.g. gelman/mrp,
stan-dev/rstanarm,
kuriwaki/sparseregMRP),
these often define generic functions and leave users to prepare the
cleaned data to use those functions with specific requirements.
The ccesMRPprep package instead provides data loading, processing,
and formatting functions for a particular task: using CCES data for
MRP. Limiting its usage to a fixed (but fairly widespread) set of survey
data has several benefits. Its key contributions are functions that are
calibrated to a consistent syntax, pre-built lookup tables and value-key
pairs of data that are based upon a careful reading of data sources, and
data loading functions that use APIs
(IQSS/dataverse-client-r
and walkerke/tidycensus) to
reduce the dependency on downloading large files. Model fitting and
visualization of MRP itself is handled in the companion package,
kuriwaki/ccesMRPrun.
This package is focused on the preparation to get there.
Installation
# remotes::install_github("kuriwaki/ccesMRPprep")
library(ccesMRPprep)Getting Started
See vignette("overview") for a overview of the steps involved.
For documentation of the data sources, see vignette("acs") for the
Census and vignette("derived") for CCES variables.
This vignette also covers more advanced techniques to expand population
tables. See vignette("synth") for an overview and demonstration.
Each function and built-in data provides documentation as well.
Workflow
See the overview vignette (vignette("overview")) from a illustrative
workflow.
Data Sources
Function-specific pages will detail the documentation used in each function. Here is a manual compilaiton:
| Information | Source | Citation and URL (if public) |
|---|---|---|
| CCES Covariates | Cumulative CCES | Shiro Kuriwaki, “Cumulative CCES Common Content”. https://doi.org/10.7910/DVN/II2DB6 |
| CCES Outcomes | Each Year’s CCES | Stephen Ansolabehere, Sam Luks, and Brian Schaffner. “CCES Common Content” (varies by year). https://cces.gov.harvard.edu/ |
| Poststratification | Census Bureau ACS | American Community Survey. Extracted via tidycensus package. See ACS vignette |
| District-level Contestedness and Incumbency | Collected mainly by Jim Snyder | |
| CD-level Presidential Voteshare | Daily Kos | Daily Kos, The ultimate Daily Kos Elections guide to all of our data sets |
| State-level Presidential Voteshare | MEDSL | MIT Election Data and Science Lab, 2017, “U.S. President 1976–2016”. https://doi.org/10.7910/DVN/42MVDX |
Related Packages
- kuriwaki/rcces has another set of CCES related functions, but these are either my own personal functions in development (not for production), or specific to non-MRP projects.
- kuriwaki/CCES_district-opinion is a private package that uses (among others) this package to process large CCES data for MRP at scale.
Support
This package is a part of the CCES MRP project, supported by NSF Grant 1926424: Bayesian analytical tools to improve survey estimates for subpopulations and small areas. The contents are based on collaborations and discussions with Ben Bales, Lauren Kennedy, Mitzi Morris, and Soichiro Yamauchi.