Automating the collection of U.S. Congress members' press releases.
pressR scrapes press releases from U.S. House (*.house.gov) and Senate
(*.senate.gov) members' websites over a date range and returns a tidy data
frame. Instead of guessing among dozens of CSS/XPath selectors per site, it
detects the content-management system behind each site and routes to a
dedicated extractor. Coverage is ~99% of the House and ~99% of the Senate.
Member sites cluster into a few vendor families. The House skews to the official Drupal template; the Senate skews to WordPress (often with a custom press-release post type). Approximate House shares:
| CMS | Share | How it's scraped |
|---|---|---|
| Drupal (official House template) | ~54% | /media/press-releases listing, ?page=N pagination |
| ASP.NET ("DocumentID") | ~20% | documentquery.aspx listing, documentsingle.aspx items |
| WordPress | ~12% (most of the Senate) | wp-json REST API — the congress_press_release category, a press-release custom post type, or an Elementor HTML fallback when REST is blocked |
| press-releases vendor | small | /press-releases (or /media-center) listing with ?ID=<GUID> (year-less dates inferred) or /YYYY/M/slug item links; pager param auto-probed |
| headless WordPress (Next.js SPA) | small | public WPGraphQL endpoint (/graphql); releases at /posts/<slug> |
| Other | ~14% | generic heuristic extractor (heading-link or long-text titles + nearest date, incl. /YYYY/M/slug fallback) |
Five vendor extractors plus one generic fallback cover both chambers.
Pages are fetched statically by default; a headless-browser fallback
([render_html()], via the suggested chromote package) is available for the
rare JS-rendered site.
# install.packages("pak")
pak::pak("zaynesember/pressR")library(pressR)
# Current members and their sites
members <- list_members()
#> # A tibble: 437 × 7
#> name state district party committee url chamber
#> <chr> <chr> <chr> <chr> <chr> <chr> <chr>
#> 1 Moore, Barry Alabama 1st R Agriculture;Judici… https://barrymoore.house.gov house
#> ...
# One member
moore <- scrape_member("barrymoore.house.gov", from = "2026-01-01")
# A set of members (metadata carried through)
res <- scrape_pressers(
members[1:10, ],
from = "2026-01-01",
to = Sys.Date(),
log_fails = TRUE # writes fails.csv
)
# The whole House (optionally capped for a quick sample)
all <- scrape_house(from = "2026-01-01", max_members = 50)
# The Senate works the same way (same extractors, plus a `chamber` column)
senators <- list_senators()
sen <- scrape_senate(from = "2026-01-01", max_members = 50)Every release-returning function yields columns date, title, body,
tags, url, cms (plus any member metadata). scrape_pressers() /
scrape_house() also attach a failures table:
attr(res, "failures") # tibble(url, stage, message)list_members()parses https://www.house.gov/representatives.detect_cms()classifies a homepage via its<meta generator>tag and markup fingerprints.- The matching extractor finds the press-release listing, walks its pages newest-first until the window's start, and pulls each release's body.
scrape_member()ties these together for one site;scrape_pressers()runs many, isolating per-site failures.
Requests carry an identifying user agent and are throttled and retried via httr2. Tunable via options:
options(pressR.throttle = 20) # requests/minute (default 20)
options(pressR.cache_dir = "~/.cache/pressR") # enable on-disk HTTP cacheScrapes return in-memory tibbles; to build historical coverage, append runs to
a local, year-partitioned, de-duplicated store (xz-compressed RDS, keyed on
url so re-scrapes refresh rather than duplicate):
res <- scrape_house(from = "2026-01-01")
archive_releases(res) # -> tools::R_user_dir("pressR","data")
read_archive(from = "2026-01-01", to = Sys.Date())Set options(pressR.archive_dir = "~/pressR-archive") to choose the location.
Because the corpus grows (~33 MB/year compressed, body text included) it lives
on disk, not in the package. Prebuilt snapshots are published as GitHub release
assets; fetch them without scraping via download_archive() (and, for
maintainers with write access, publish_archive()). Both need the suggested
piggyback package.
Archiving accumulates a tidy, one-row-per-release corpus across both chambers.
The published snapshot — fetch it with download_archive() — currently holds:
- 436,201 releases from 537 member offices, 2010 through mid-2026
- ~94% carry full body text; issue tags wherever the CMS exposes them (~41% overall)
- House — 282,687 · Senate — 153,514 · by party ≈ 262k D · 173k R · 0.7k I
A reproducible recent slice (what scrape_house(from = "2026-01-01") yields):
read_archive(from = "2026-01-01")
#> # A tibble: 24,122 × 12
#> name state party chamber date title tags cms
#> <chr> <chr> <chr> <chr> <date> <chr> <chr> <chr>
#> 1 Latta, Robert Ohio R house 2026-06-17 Latta Applauds FDA Approval of New … Veterans aspx
#> 2 Hyde-Smith, Cindy MS R senate 2026-04-23 Hyde-Smith Backs Bill to Reauthoriz… Health Care drupal
#> 3 Clark, Katherine Mass… D house 2026-06-18 Whip Clark Celebrates Reopening of … Health Care wordpress
#> # … plus district, committee, body, urlEvery release is dated, attributed (member, state, party, chamber, committee), and topic-tagged, so the corpus is ready for quick analysis:
library(dplyr); library(tidyr)
a <- read_archive(from = "2026-01-01")
# Most-used issue tags
a |> filter(!is.na(tags)) |> separate_rows(tags, sep = ";") |> count(tags, sort = TRUE)
#> tags n
#> Education 1198
#> Veterans 1088
#> Immigration 811
#> Health Care 775
#> Agriculture 584
#> # …Issue tags and state names are recorded as each chamber/office formats them, so synonyms ("Health Care" vs "Healthcare") and mixed state forms (House full names, Senate two-letter codes) appear — normalize before aggregating if needed.
An exploratory analysis layer built on the archived corpus — near-duplicate
"message family" detection, issue-tag completion, structural topic models, and
sentiment, plus a Shiny dashboard — lives in nlp/. It also folds in two
external congressional press-release datasets (Stout 114–117; Wang & Tucker
109–115) for historical depth, taking the combined corpus to ~894k releases back
to 2004. This is research code, kept separate from the installable package; see
nlp/README.md for the pipeline and how to run it.
devtools::load_all()
devtools::test() # offline unit tests run against saved HTML fixtures
devtools::check()The original grad-school notebook implementation is preserved under
legacy/ for reference.
Some of the original legacy code (under legacy/) was generously
provided by Chris Stout.
MIT © Zayne Sember