Introduction to jstor

  collapse = TRUE,
  comment = "#>"

The tool Data for Research (DfR) by JSTOR is a valuable source for citation analysis and text mining. jstor provides functions and suggests workflows for importing datasets from DfR.

When using DfR, requests for datasets can be made for small excerpts (max. 25,000 records) or large ones, which require an agreement between the researcher and JSTOR. jstor was developed to deal with very large datasets which require an agreement, but can be used with smaller ones as well.

The most important set of functions is a group of jst_get_* functions:

I will demonstrate their usage using the sample dataset which is provided by JSTOR on their website.

General Concept

All functions from the jst_get_* family which are concerned with meta data operate along the same lines:

  1. The file is read with xml2::read_xml().
  2. Content of the file is extracted via XPATH or CSS-expressions.
  3. The resulting data is returned in a tidy tibble.

The functions are similar in that all operate on single files (article, book, research report or pamphlet). Depending on the content of the file, the output of the functions might have one or multiple rows. jst_get_article always returns a tibble with one row: the core meta data (like title, id, or first page of the article) are single items, and only one article is processed at a time. Running jst_get_authors for the same article might give you a tibble with one or multiple rows, depending on the number of authors the article has. The same is true for jst_get_references and jst_get_footnotes. If a file has no data on references (they might still exist, but JSTOR might not have parsed them), the output is only one row, with missing references. If there is data on references, each entry gets its own row. Note however, that the number of rows does not equal the number of references. References usually start with a title like "References", which is obviously not a reference to another article. Be sure to think carefully about your assumptions and to check the content of your data before you make inferences.

Books work a bit differently. Searching for data on lets you filter for books, which are actually book chapters. If you receive data from DfR on a book chapter, you always get one xml-file with the whole book, including data on all chapters. Ngram or full-text data for the same entry however is processed only from single chapters^[See the technical specifications for more detail.]. Thus, the output of jst_get_book for a single file is similar to the one from jst_get_article: it is one row with general data about the book. jst_get_chapters gives you data on all chapters, and the resulting tibble therefore might have multiple rows.

The following sections showcase the different functions separately.


Apart from jstor we only need to load dplyr for matching records and knitr for printing nice tables.



The basic usage of the jst_get_* functions is very simple. They take only one argument, the path to the file to import:

meta_data <- jst_get_article(file_path = jst_example("article_with_references.xml"))

The resulting object is a tibble with one row and 17 columns. The columns correspond to most of the elements documented here:

The columns are:

Since the output from all functions are tibbles, the result is nicely formatted:

meta_data %>% kable()


Extracting the authors works in similar fashion:

authors <- jst_get_authors(jst_example("article_with_references.xml"))

Here we have the following columns:

The number of rows matches the number of authors -- each author get its' own row.


references <- jst_get_references(jst_example("article_with_references.xml"))

# # we need to remove line breaks for knitr::kable() to work properly for printing
references <- references %>%
  mutate(ref_unparsed = stringr::str_remove_all(ref_unparsed, "\\\n"))

We have two columns:

Here I display 5 random entries:

references %>% 
  sample_n(5) %>% 

This example shows several things: file_name is identical among rows, since it identifies the article and all references came from one article. The the sample file doesn't follow a typical convention (it was published in 1922), therefore there are several different headings (ref_title). Usually, this is only "Bibliography" or "References".

Since the references were not parsed by JSTOR, we only get an unparsed version. In general, the content of references (unparsed_refs) is in quite a raw state, quite often the result of digitising scans via OCR. For example, the last entry reads like this: MACHADO, A.1911 Zytologische Untersuchungen fiber Trypanosoma rotatorium .... There is an error here: fiber should be ├╝ber. The language of the source is German, but the OCR-software assumed English. Therefore, it didn't recognize the Umlaut. Similar errors are common for text read via OCR.

For other files, we can set parse_refs = TRUE, so references will be imported in their parsed form, whenever they are available.

  parse_refs = TRUE
) %>% 

Note, that there might be other content present like endnotes, in case the article used endnotes rather than footnotes.


jst_get_footnotes(jst_example("article_with_references.xml")) %>% 

Very commonly, articles either have footnotes or references. The sample file used here does not have footnotes, therefore a simple tibble with missing footnotes is returned.

I will use another file to demonstrate footnotes.

footnotes <- jst_get_footnotes(jst_example("article_with_footnotes.xml"))

footnotes %>% 
  mutate(footnotes = stringr::str_remove_all(footnotes, "\\\n")) %>% 

In general, you might need to combine jst_get_footnotes() with jst_get_references() to get all available information on citation data.


The function to extract full texts can't be demonstrated with proper data, since the full texts are only supplied upon special request with DfR. The function guesses the encoding of the specified file via readr::guess_encoding(), reads the whole file and returns a tibble with file_name, full_text and encoding.

I created a file that looks similar to files supplied by DfR with sample text:

full_text <- jst_get_full_text(jst_example("full_text.txt"))
full_text %>% 
  mutate(full_text = stringr::str_remove_all(full_text, "\\\n")) %>% 

Combining results

Different parts of meta-data can be combined by using dplyr::left_join().

Matching with authors

meta_data %>% 
  left_join(authors) %>%
  select(file_name, article_title, pub_year, given_name, surname) %>% 

Matching with references

meta_data %>% 
  left_join(references) %>% 
  select(file_name, article_title, volume, pub_year, ref_unparsed) %>%
  head(5) %>% 


Quite recently DfR added book chapters to their stack. To import metadata about the books and chapters, jstor supplies jst_get_book and jst_get_chapters.

jst_get_book is very similar to jst_get_article. We obtain general information about the complete book:

jst_get_book(jst_example("book.xml")) %>% knitr::kable()

A single book might contain many chapters. jst_get_chapters extracts all of them. Due to this, the function is a bit slower than most of jstor's other functions.

chapters <- jst_get_chapters(jst_example("book.xml"))


Without the abstracts (they are rather long) the first 10 chapters look like this:

chapters %>% 
  select(-abstract) %>% 
  head(10) %>% 

Since extracting all authors for all chapters needs considerably more time, by default authors are not extracted. You can import them like so:

author_chap <- jst_get_chapters(jst_example("book.xml"), authors = TRUE) 

The authors are supplied in a list column:


You can expand this list with tidyr::unnest:

author_chap %>% 
  tidyr::unnest(authors) %>% 
  select(part_id, given_name, surname) %>% 
  head(10) %>% 

You can learn more about the concept of list-columns in Hadley Wickham's book R for Data Science.

Try the jstor package in your browser

Any scripts or data that you put into this service are public.

jstor documentation built on Dec. 11, 2021, 9:56 a.m.