The dataverse package provides access to Dataverse APIs (versions 4+), enabling data search, retrieval, and deposit, thus allowing R users to integrate public data sharing into the reproducible research workflow.
# Install from CRAN install.packages("dataverse") # Install from GitHub # install.packages("remotes") remotes::install_github("iqss/dataverse-client-r")
Many features of the Dataverse API are public and require no authentication. This means in many cases you can search for and retrieve data without a Dataverse account or API key – you wil not need to worry about this.
For features that require a Dataverse account for the specific server installation of the Dataverse software, and an API key linked to that account. Instructions for obtaining an account and setting up an API key are available in the Dataverse User Guide. (Note: if your key is compromised, it can be regenerated to preserve security.) Once you have an API key, this should be stored as an environment variable called
DATAVERSE_KEY. It can be set as a default by adding
in your .Renviron file, where
examplekey12345 should be replace with your own key. The environment file can be opened by
Because there are many Dataverse installations, all functions in the R client require specifying what server installation you are interacting with. There are multiple ways to specify the server:
server argument in each function. e.g.,
server = "dataverse.harvard.edu" in the
Set the environment variable,
DATAVERSE_SERVER, in the script to be used throughout the session. e.g.,
Sys.setenv("DATAVERSE_SERVER" = "dataverse.harvard.edu")
.Renvironfile directly or open it by
usethis::edit_r_environ(). Then enter
DATAVERSE_SERVER = "dataverse.harvard.edu".
In all cases, values should be the Dataverse server, without the “https” prefix or the “/api” URL path, etc.
The dataverse package provides multiple interfaces to obtain data into R. Users can supply a file DOI, a dataset DOI combined with a filename, or a dataverse object. They can read in the file as a raw binary or a dataset read in with the appropriate R function.
get_dataframe_*() functions, depending on the input you have. For example, we will read a survey dataset on Dataverse, nlsw88.dta (
doi:10.70122/FK2/PPKHI1/ZYATZZ), originally in Stata dta form.
With a file DOI, we can use the
nlsw <- get_dataframe_by_doi( filedoi = "10.70122/FK2/PPIAXE/MHDB0O", server = "demo.dataverse.org" )
## Downloading ingested version of data with readr::read_tsv. To download the original version and remove this message, set original = TRUE. ## Rows: 2246 Columns: 17 ## ── Column specification ──────────────────────────────────────────────────────────────────────────────────────────────── ## Delimiter: "\t" ## dbl (17): idcode, age, race, married, never_married, grade, collgrad, south, smsa, c_city, industry, occupation, uni... ## ## ℹ Use `spec()` to retrieve the full column specification for this data. ## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
which by default reads in the ingested file (not the original dta) by the
Alternatively, we can download the same file by specifying the filename and the DOI of the “dataset” (in Dataverse, a collection of files is called a dataset).
nlsw_tsv <- get_dataframe_by_name( filename = "nlsw88.tab", dataset = "10.70122/FK2/PPIAXE", server = "demo.dataverse.org" )
Now, Dataverse often translates rectangular data into an ingested, or “archival” version, which is application-neutral and easily-readable.
read_dataframe_*() defaults to taking this ingested version rather than using the original, through the argument
original = FALSE.
This default is safe because you may not have the proprietary software that was originally used. On the other hand, the data may have lost information in the process of the ingestation.
Instead, to read the same file but its original version, specify
original = TRUE and set an
.f argument. In this case, we know that
nlsw88.tab is a Stata
.dta dataset, so we will use the
nlsw_original <- get_dataframe_by_name( filename = "nlsw88.tab", dataset = "10.70122/FK2/PPIAXE", .f = haven::read_dta, original = TRUE, server = "demo.dataverse.org" )
Note that even though the file prefix is “.tab”, we use
Of course, when the dataset is not ingested (such as a Rds file), users would always need to specify an
.f argument for the specific file.
Note the difference between
nls_original preserves the data attributes like value labels, whereas
nls_tsv has dropped this or left this in file metadata.
class(nlsw_tsv$race) # tab ingested version only has numeric data
attr(nlsw_original$race, "labels") # original dta has value labels
Dataverse provides two - basically unrelated - workflows for managing (adding, documenting, and publishing) datasets. The first is built on SWORD (v2.0). This means that to create a new dataset listing, you will have to first initialize a dataset entry with some metadata, add one or more files to the dataset, and then publish it. This looks something like the following:
# After setting appropriate dataverse server and environment, obtain SWORD # service doc d <- service_document() # create a list of metadata for a file metadat <- list( title = paste0("My-Study_", format(Sys.time(), '%Y-%m-%d_%H:%M')), creator = "Doe, John", description = "An example study" ) # create the dataset, where "mydataverse" is to be replaced by the name # of the already-created dataverse as shown in the URL ds <- initiate_sword_dataset("<mydataverse>", body = metadat) # add files to dataset readr::write_csv(iris, file = "iris.csv") # Search the initiated dataset and give a DOI and version of the dataverse as an identifier mydoi <- "doi:10.70122/FK2/BMZPJZ&version=DRAFT" # add dataset add_dataset_file(file = "iris.csv", dataset = mydoi) # publish new dataset publish_sword_dataset(ds) # dataset will now be published list_datasets("<mydataverse>")
The second workflow is called the “native” API and is similar but uses slightly different functions:
# create the dataset ds <- create_dataset("mydataverse") # add files tmp <- tempfile() write.csv(iris, file = tmp) f <- add_dataset_file(file = tmp, dataset = ds) # publish dataset publish_dataset(ds) # dataset will now be published get_dataverse("mydataverse")
Through the native API it is possible to update a dataset by modifying its metadata with
update_dataset() or file contents using
update_dataset_file() and then republish a new version using
The R client is current stable for data search and download. For more extensive features of uploading and maintaining data, see pyDataverse.
Currently, functions related to user management and permissions - are not currently exported in the package (but are drafted in the source code).
dataverse is the next-generation iteration of the now removed dvn package, which works with Dataverse 3 (“Dataverse Network”) applications.
Users interested in downloading metadata from archives other than Dataverse may be interested in Kurt Hornik’s OAIHarvester and Scott Chamberlain’s oai, which offer metadata download from any web repository that is compliant with the Open Archives Initiative standards. Additionally, rdryad uses OAIHarvester to interface with Dryad. The rfigshare package works in a similar spirit to dataverse with https://figshare.com/.
A 2021 talk demonstrating the Dataverse package is available at https://www.youtube.com/watch?v=-J-eiPnmoNE.