cpsvote: A Social Science Toolbox for Using the Current Population Survey’s Voting and Registration Supplement

cpsvote helps you work with data from the Current Population Survey’s (CPS) Voting and Registration Supplement (VRS), published by the U.S. Census Bureau and Bureau of Labor Statistics. This high-quality, large-sample survey has been conducted after every federal election (in November of even years) since 1964, surveying Americans about their voting practices and registration. The raw data, archived by the National Bureau of Economic Research, is spread across several fixed-width files with different question locations and formats.

This package consolidates common questions and provides the data in a structure that is much easier to work with and interpret, since much of the basic factor recoding has already been done. We also calculate alternative sample weights based on demonstrated changes in non-response bias over the decades, recommended by several elections researchers as a best practice. Documentation of this reweighting is provided in vignette("voting").

We have provided access to VRS data from 1994 to 2024.

Installing and Loading the Package

Version 0.2 is on CRAN!

install.packages('cpsvote')
library(cpsvote)

You can also install the development version from our GitHub repository.

# remotes::install_github("Reed-EVIC/cpsvote")
install.packages("cpsvote", repos = c('https://reed-evic.r-universe.dev', 'https://cloud.r-project.org'))
library(cpsvote)

Basic Use (AKA Tips if You Don’t Like Reading Documentation)

We have written several functions to transform the VRS from its original format into a more workable structure. The easiest way to access the data is with the cps_load_basic() command:

# Load All Years
# May take some time to download and process files the first time! 
cps <- cps_load_basic()  

# Just load 2006 and 2008
cps <- cps_load_basic(years = c(2006, 2008))

This will load the prepared VRS data into your environment as a tibble called cps. The first time you try to load a given year of data, the raw data file will be downloaded to your computer. This can take some time depending on your internet speeds. In future instances, R will just read from the data files that have already been downloaded. See ?cps_allyears_100k for a description of the columns and fields that cps_load_basic() outputs.

Data directory: By default, CPS files are stored in ~/cps_data (your home directory). You can check the current default with cps_data_dir(). To use a persistent custom location, add this to your .Rprofile:

options(cpsvote.datadir = “~/my/path/to/cps_data”)

After restarting R, all cpsvote data-loading functions will use that path automatically. If you specify a location that does not have the correct files, these functions will attempt to re-download the data from NBER, which can take up noticeable time and storage space.

We have also included a 100,000 row sample of the full VRS data, which comes with the package as cps_allyears_100k. This is particularly useful for planning out a given analysis before you download the full data sets.

library(dplyr)
data("cps_allyears_100k")

cps_allyears_100k %>%
  select(1:3, VRS_VOTE:VRS_REG, VRS_VOTEMETHOD_CON, turnout_weight) %>%
  sample_n(10)

FILE	YEAR	STATE	VRS_VOTE	VRS_REG	VRS_VOTEMETHOD_CON	turnout_weight
cps_nov2018.zip	2018	NJ	NO RESPONSE	NO RESPONSE	NA	3378.0841
cps_nov1994.zip	1994	MI	NO	YES	NA	1784.3059
cps_nov2000.zip	2000	CT	NO RESPONSE	NO RESPONSE	NA	2444.6010
cps_nov2014.zip	2014	VT	YES	NA	ELECTION DAY	253.6059
cps_nov2024.zip	2024	KS	NA	NA	NA	2041.8559
cps_nov2000.zip	2000	AK	NA	NA	NA	495.0288
cps_nov2020.zip	2020	IA	YES	NA	ELECTION DAY	1783.2662
cps_nov2006.zip	2006	VT	YES	NA	ELECTION DAY	261.6553
cps_nov2002.zip	2002	NY	NO	YES	NA	3623.6069
cps_nov2008.zip	2008	AR	YES	NA	ELECTION DAY	1789.3102

The CPS has survey weights that are necessary to calculate accurate estimates about the US population. Two R packages that work with survey weighting are survey and srvyr (a tidyverse-compatible wrapper for survey). You can see more examples and details on weighting in vignette("voting"), but here is one example of using srvyr to calculate state-level voter turnout among eligible voters in 2020.

library(srvyr)

cps20_weighted <- cps_load_basic(years = 2020, datadir = here::here('cps_data')) %>%
  as_survey_design(weights = turnout_weight)

turnout20 <- cps20_weighted %>%
  group_by(STATE) %>%
  summarize(turnout = survey_mean(hurachen_turnout == "YES", na.rm = TRUE))

head(turnout20, 10)

STATE	turnout	turnout_se
AL	0.4827207	0.0123850
AK	0.5670961	0.0178198
AZ	0.5997274	0.0154234
AR	0.4270782	0.0131359
CA	0.5554578	0.0066525
CO	0.5998372	0.0173296
CT	0.5806014	0.0191721
DE	0.5926931	0.0174831
DC	0.5937632	0.0234803
FL	0.5534077	0.0085969

These estimates follow closely Dr. Michael McDonald’s estimates of turnout among eligible voters in the November 2020 General Election. For a detailed examination of how non-response bias has affected the use of CPS for estimating turnout, see vignette("voting"). We thank the U.S. Elections Project at the University of Florida for the turnout estimates.

Advanced Use

In addition to the basic function listed above, you can customize several steps in the process of reading in the VRS data. If you’ve worked with the CPS before, you may already have some code to read in analyze this survey data. We still hope that this package can help you organize your workflow or ease some of the more tedious steps necessary to work with the CPS.

Be sure to refer to the CPS documentation files when working with alternative versions of the VRS data. We have included the function cps_download_docs() to provide the documentation versions that match this data. These are all in PDF format (and several are not text-based), so they are not easy to search through.

cps_load_basic() is a wrapper for several constituent steps that have their own parameters and assumptions. We’ve detailed the changes made to get from the raw data file to the cleaned file in vignette("add-variables").

cps_download_data(path = "cps_data",
                  years = seq(1994, 2020, 2))
cps_download_docs(path = "cps_data",
                  years = seq(1994, 2020, 2))

cps_read(years = seq(1994, 2020, 2),
         dir = "cps_data",
         cols = cpsvote::cps_cols,
         names_col = "new_name",
         join_dfs = TRUE) %>%
    cps_label(factors = cpsvote::cps_factors,
              names_col = "new_name",
              na_vals = c("-1", "BLANK", "NOT IN UNIVERSE"),
              expand_year = TRUE,
              rescale_weight = TRUE,
              toupper = TRUE) %>%
    cps_refactor(move_levels = TRUE) %>%
    cps_recode_vote(vote_col = "VRS_VOTE",
                    items = c("DON'T KNOW", "REFUSED", "NO RESPONSE")) %>%
    cps_reweight_turnout()

• cps_download_data() will download the data files from NBER according to years into the folder at path. This is automatically called by cps_read() when the CPS data files are not found in the provided dir - it will search for files with the 4-digit year associated with their data.
• cps_download_docs() will downlaod the pdf documentation into path for each year supplied in years.The documentation here is aligned with the NBER data, and other data sources (such as ICPSR) may have edited the data such that their data or documentation does not line up with the NBER data and documentation. By using the NBER data through cps_download_docs(), you can make sure that the fields you look up in documentation are the proper fields referenced in the data.
• cps_read() is the function that actually loads in the original, (mostly) numeric data from files defined by the arguments years and dir. Since the raw data is in fixed-width files, you have to define the range of characters that are read. You can see the default set of columns in the included data set cps_cols, or supply cols with your own specifications of columns (for details on adding other columns, see vignette("add-variables")). The names_col argument details which variable in cols will become the column names for the output; we have provided the original CPS names as cps_name, but recommend using new_name as it is more informative and accounts for questions changing names (“PES5”, “PES6”, etc.) across multiple years. join_dfs lets you join multiple years into one tibble, and should only be used if you’re sure that a column name (like “PES5”) refers to the same question across all years you read in.
• cps_label() replaces the numeric entries from the raw data with appropriate factor levels (as given by the data documentation; see cps_download_docs()). We have taken the factor levels as written from the PDFs, including capitalization, typos, and differences across years. This is provided in the included cps_factors dataset, but you can supply the factors argument with your own coding (for details on changing factor levels or adding them for a new column, see vignette("add-variables")). The names_col argument defines which column of factors contains the column names that match the incoming data set to be labelled. Further: na_vals defines which factor levels should be marked as NA, expand_year turns the two-digit years in some files into four-digit years (e.g. “94” becomes “1994”), and rescale_weight divides the given weight by 10,000 (as noted by the data documentation) to ensure accurate population sums. toupper will make all the factor levels upper case, which is useful because as-is the factors are a mix of sentence case and upper case.
• cps_refactor deals with all of the typos, capitalization, and shifting questions across years. We have attempted here to consolidate factor levels and variables in a way that makes sense. For example, one common method of assessing vote mode (in-person on Election Day, early in-person, or by mail) has been split between two separate questions from 2004 onwards, and this function consolidates those two questions (and the one question of previous surveys) into one VRS_VOTEMETHOD_CON variable. Note that this function will only work with certain column names in the data; see ?cps_refactor for more details.
• cps_recode_vote() recodes the variable VRS_VOTE according to two different assessments of voter turnout. The new variable cps_turnout will calculate turnout the same way that the Census does, while another new variable hurachen_turnout will calculate turnout according to Hur & Achen (2013). These two methods differ in how they count responses of “Don’t know”, “Refused”, and “No response”; see vignette("background") for more details.
• cps_reweight_turnout() adds a new variable, turnout_weight, that reweights the original WEIGHT according to Hur & Achen (2013) to account for the adjusted turnout measure. This corrects for increased nonresponse to the VRS over time, as well as a general pattern of respondents overreporting their personal voting history (though the CPS sees less overreporting than other surveys). See vignette("background") for details.

You can use different combinations of these functions to customize which CPS data is read in. For example, this code would load the 2014 VRS data with the original column names and numeric data.

cps14 <- cps_read(2014, names_col = "cps_name")

You can then apply factor labels to this data.

cps14_lab <- cps_label(cps14, names_col = "cps_name")

Note that some features (like cps_refactor()) won’t work on certain customized versions of the data, because they are relatively hard-coded based on specific column names. For example, correcting “HIPSANIC” to “HISPANIC” only works if you know which column represents the Hispanic flag. Feel free to take the code from functions like this and adapt based on your own column names.

Examples, Background Reading, and Data Sources

• Vignettes:
  • vignette("basics") provides an intro to the package with some basic instructions for use, and mirrors our GitHub README
  • vignette("background") describes our intellectual rationale for creating this package
  • vignette("add-variables") describes how additional variables from the CPS can be merged with the default dataset
  • vignette("voting") does a deep dive into how to use the CPS and the default datasets from cpsvote to look at voter turnout and mode of voting
• Aram Hur, Christopher H. Achen. Coding Voter Turnout Responses in the Current Population Survey. Public Opinion Quarterly, Volume 77, Issue 4, Winter 2013, Pages 985–993. https://doi.org/10.1093/poq/nft042
• Michael McDonald. What’s Wrong with the CPS? Presented at the 2014 American Political Science Association Conference, Washington, D.C., August 27-31. https://www.electproject.org/home/voter-turnout/cps-methodology
• The Current Population Survey is conducted monthly by the U.S. Census Bureau and the Bureau of Labor Statistics, and the Voting and Registration Supplement is administered as part of this survey after each federal election. The CPS data that this package downloads is provided by the National Bureau of Economic Research.
• This is an animated ternary plot made using vote mode data from cpsvote. The code that created this in available in the “The Move Away From Election Day Voting in America: The Snowglobe Plot” vignette.

Acknowledgements

The cpsvote package was originally created at the Elections & Voting Information Center at Reed College, now colocated at Center for Public Service at Portland State University.