R Client for Dataverse Repositories

The dataverse package provides access to Dataverse APIs (versions 4-5), enabling data search, retrieval, and deposit, thus allowing R users to integrate public data sharing into the reproducible research workflow. dataverse is the next-generation iteration of the dvn package, which works with Dataverse 3 (“Dataverse Network”) applications. dataverse includes numerous improvements for data search, retrieval, and deposit, including use of the (currently in development) sword package for data deposit and the UNF package for data fingerprinting.

Getting Started

You can find a stable 2017 release on CRAN, or install the latest development version from GitHub:

library("dataverse")

Keys

Some 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 for that a specific Dataverse installation. But, other features 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 within R using:

Sys.setenv("DATAVERSE_KEY" = "examplekey12345")

Server

Because there are many Dataverse installations, all functions in the R client require specifying what server installation you are interacting with. This can be set by default with an environment variable, DATAVERSE_SERVER. This should be the Dataverse server, without the “https” prefix or the “/api” URL path, etc. For example, the Harvard Dataverse can be used by setting:

Sys.setenv("DATAVERSE_SERVER" = "dataverse.harvard.edu")

Note: The package attempts to compensate for any malformed values, though.

Currently, the package wraps the data management features of the Dataverse API. Functions for other API features - related to user management and permissions - are not currently exported in the package (but are drafted in the source code).

Data and Metadata Retrieval

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.

Reading data as R objects

Use the 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 get_dataframe_by_doi function:

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.

## 
## ── Column specification ────────────────────────────────────────────────────────────────────────────────────────────────
## cols(
##   idcode = col_double(),
##   age = col_double(),
##   race = col_double(),
##   married = col_double(),
##   never_married = col_double(),
##   grade = col_double(),
##   collgrad = col_double(),
##   south = col_double(),
##   smsa = col_double(),
##   c_city = col_double(),
##   industry = col_double(),
##   occupation = col_double(),
##   union = col_double(),
##   wage = col_double(),
##   hours = col_double(),
##   ttl_exp = col_double(),
##   tenure = col_double()
## )

which by default reads in the ingested file (not the original dta) by the readr::read_tsv function.

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 haven::read_dta function.

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 haven::read_dta.

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_tsv and nls_original. 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

## [1] "numeric"

attr(nlsw_original$race, "labels") # original dta has value labels

## white black other 
##     1     2     3

Reading a dataset as a binary file.

In some cases, you may not want to read in the data in your environment, perhaps because that is not possible (e.g. for a .docx file), and you want to simply write these files your local disk. To do this, use the more primitive get_file_* commands. The arguments are equivalent, except we no longer need an .f argument

nlsw_raw <-
  get_file_by_name(
    filename    = "nlsw88.tab",
    dataset     = "10.70122/FK2/PPIAXE",
    server      = "demo.dataverse.org"
  )
class(nlsw_raw)

## [1] "raw"

Reading file metadata

The function get_file_metadata() can also be used similarly. This will return a metadata format for ingested tabular files in the ddi format. The function get_dataset() will retrieve the list of files in a dataset.

get_dataset(
  dataset = "10.70122/FK2/PPIAXE",
  server  = "demo.dataverse.org"
)

## Dataset (182162): 
## Version: 1.1, RELEASED
## Release Date: 2020-12-30T00:00:24Z
## License: CC0
## 22 Files:
##                   label version      id               contentType
## 1 nlsw88_rds-export.rds       1 1734016  application/octet-stream
## 2            nlsw88.tab       3 1734017 text/tab-separated-values

Data Discovery

Dataverse supplies a robust search API to discover Dataverses, datasets, and files. The simplest searches simply consist of a query string:

dataverse_search("Gary King")

More complicated searches might specify metadata fields:

dataverse_search(author = "Gary King", title = "Ecological Inference")

And searches can be restricted to specific types of objects (Dataverse, dataset, or file):

dataverse_search(author = "Gary King", type = "dataset")

The results are paginated using per_page argument. To retrieve subsequent pages, specify start.

Data Archiving

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:

# retrieve your service document
d <- service_document()

# create a list of metadata
metadat <-
  list(
    title       = "My Study",
    creator     = "Doe, John",
    description = "An example study"
  )

# create the dataset
ds <- initiate_sword_dataset("mydataverse", body = metadat)

# add files to dataset
tmp <- tempfile()
write.csv(iris, file = tmp)
f <- add_file(ds, file = tmp)

# 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 publish_dataset().

Other Installations

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/.