# ropercenter: Reproducible Retrieval of Roper Center Datasets

#### 2018-06-28

The Roper Center for Public Opinion Research, in its own words, works “to collect, preserve, and disseminate public opinion data; to serve as a resource to help improve the practice of survey research; and to broaden the understanding of public opinion through the use of survey data in the United States and around the world.” It maintains the largest archive of public opinion data in existence, holding data dating back to the 1930s and from over 100 countries. Researchers taking advantage of these datasets, however, are caught in a bind. The terms and conditions for downloading any Roper Center dataset state that datasets “may not be resold or re-disseminated.”1 But to ensure that one’s work can be reproduced, assessed, and built upon by others, one must provide access to the raw data one employed.

The ropercenter package cuts this knot by providing programmatic, reproducible access to specified Roper Center datasets from within R for registered users at the Roper Center’s member institutions.

Please remember that by using Roper Center services, you accept all of the Center’s Terms and Conditions.

## Setup

When used interactively, the roper_download function will ask for the login information required by the Roper Center: the registered user’s email and password. After that information is input once, it will be entered automatically for any other download requests made in the same session. To change this contact information within a session, one may set the argument reset to TRUE when running roper_download again, and the function will again request the required information.

An optional, but highly recommended, setup step is to add the information the Roper Center requires to your .Rprofile as in the following example:

options("roper_email" = "juanita-herrera@uppermidwest.edu",
"roper_password" = "password123!")

The roper_download function will then access the information it needs to pass on to the Roper Center by default. This means that researchers will not have to expose their info in their R scripts and that others reproducing their results later—given that they have registered as users with the Roper Center—will be able to execute those R scripts without modification. (They will, however, need to enter their own information either interactively or in their own .Rprofiles, a detail that should be noted in the reproducibility materials to avoid confusion.)

## Use

The roper_download function (1) simulates a visit to the Roper Center’s sign-in page, (2) enters the required information to sign in, (3) navigates to a specified dataset and downloads the dataset’s files, and, optionally but by default, (4) converts the dataset’s SPSS-formated files to .Rdata format.

Datasets are specified using the file_id argument. The Roper Center uses a unique number to identify each of its datasets; this number is consistently listed alongside the dataset’s name. For this CNN/ORC poll on the 2016 presidential election, for example, the file id is USORCCNN2015-010:

roper_download(file_id = "USORCCNN2015-010")

Multiple datasets may be downloaded from the same research area in a single command by passing a vector of ids to file_id. The following downloads the above-described CNN/ORC poll along with two similar polls conducted earlier in the campaign:

roper_download(file_id = c("USORCCNN2015-010", "USORCCNN2015-009", "USORCCNN2015-008"))

After the needed datasets are downloaded, if they are in SPSS format, they are by default converted to .RData format (via haven::read_por if possible, foreign::read.spss otherwise) and ready to be loaded into R using load() or rio::import().

orccnn2015_010 <- rio::import("roper_data/USORCCNN2015-010/USORCCNN2015-010.RData")

If no SPSS-formatted dataset is available, the .dat ASCII file is downloaded. The next section describes how to use read_ascii to make this data usable.

Many older Roper Center datasets are available only in ASCII format, which is notoriously difficult to work with. The read_ascii function facilitates the process of extracting selected variables from ASCII datasets. For single-card files, such as this Gallup Poll from June 1982, one can simply identify the names, positions, and widths of the needed variables from the codebook and pass them to read_ascii’s var_names, var_positions, and var_widths arguments. The resulting data frame will include these variables plus a variable for the respondent id number and one that encodes the raw data as a single string.

roper_download("USAIPO1982-1197G")            # Gallup Poll for June 25-28, 1982 (ASCII only)
var_names = c("q09j", "weight"),
var_positions = c(38, 1),
var_widths = c(1, 1))

Multicard datasets are more complicated. In the best case, the file contains one line per card per respondent; then, the user can extract the needed variables by adding only the var_cards and total_cards arguments. When this condition is violated—there is not a line for every card for every respondent, or there are extra lines—the read_ascii function will throw an error and request the additional arguments card_pattern and respondent_pattern. These take regular expressions that match the card and respondent identifiers on each line in the original file (note that look-behind assertions are often particularly handy for constructing these regexs). Either way, the resulting data frame will include the variables specified in the var_ arguments, a variable for the respondent id number, and as many additional variables as cards in the file, each of which encodes the raw data on that card as a single string.

library(ropercenter)
respondent_pattern = "(?<=^\\s{2})\\d{4}")