Download American Community Survey (ACS) 5-Year Data Tables from Census Bureau Nationwide in Bulk • ACSdownload

Bulk-download tables from the U.S. Census Bureau’s American Community Survey (ACS) 5-year summary file – entire-USA-at-once, no API key, one file per table.

ACSdownload reads the Census Bureau’s table-based summary file (introduced for the 2018-2022 vintage). The 3.x series only supports this format; for the older sequence-file format, use the 2.x series or tidycensus.

Why this package

The Census Bureau makes it easy to download one State at a time. The tidycensus package makes it easy to pull modest slices via the Census API. But if you want every block group in the US for a dozen tables – something EJSCREEN and similar tools need – both options are slow and awkward. There are over 240,000 block groups and 85,000 tracts.

ACSdownload does it as ~one HTTP fetch per table to the Census Bureau’s public bulk files, with retry / timeout / optional on-disk caching baked in. A 16-table block-group pull is typically a few minutes wall-clock on a residential connection, and the second run is instant when caching is enabled.

Installation

# install.packages("devtools")
devtools::install_github("ejanalysis/ACSdownload")

Quick start

The function you want is get_acs_new():

library(ACSdownload)

# Just one table, for one geography type
b25034 <- get_acs_new(
  yr     = 2024,
  tables = "B25034",
  fips   = "county"
)
b25034[["B25034"]][1:3, 1:5]

# Full EJSCREEN block-group pull, merged into one wide table
bg <- get_acs_new(
  yr     = 2024,
  fips   = "blockgroup",
  tables = setdiff(ejscreen_acs_tables, c("C16001", "B18101")),
  return_list_not_merged = FALSE,
  cache_dir = tools::R_user_dir("ACSdownload", "cache")
)

C16001 and B18101 are tract-only tables; passing them with fips = "blockgroup" filters them to zero rows. With return_list_not_merged = FALSE they get dropped from the merge with a warning, which is usually what you want.

What you get back

Each downloaded table arrives as a data.table with:

GEO_ID, fips, SUMLEVEL – bookkeeping
<TABLE>_<NNN> – estimate columns (the leading _E is stripped)
<TABLE>_M<NNN> – margin-of-error columns

Want a human-readable label for B01001_001? Use acs_label():

acs_label(c("B01001_001", "B01001_M001", "C16001_002"))
#>   variable_id table_id              label                                                table_title                    universe
#> 1: B01001_001   B01001             Total:                                                 Sex by Age            Total population
#> 2: B01001_M001  B01001             Total:                                                 Sex by Age            Total population
#> 3: C16001_002   C16001 Speak only English Language Spoken at Home for the Population 5 Years and Over Population 5 years and over

acs_label() is backed by acs_table_shells, a 28k-row lookup bundled with the package, parsed from the official Census table shells file.

Caching, retry, timeout, parallelism

get_acs_new() accepts:

arg	default	what it does
`cache_dir`	`getOption("ACSdownload.cache_dir", NULL)`	persist .dat files; second call returns instantly
`timeout_sec`	`getOption("ACSdownload.timeout", 300)`	per-request timeout in seconds
`max_retries`	`getOption("ACSdownload.retries", 3)`	retry-with-backoff on HTTP 429 / 5xx
`parallel`	`FALSE`	concurrent downloads via `future.apply` (caller sets `future::plan()`)
`variables`	`NULL`	keep only these estimate columns
`keep_moe`	`TRUE`	keep `_M<nnn>` margin-of-error columns
`keep_annotations`	`FALSE`	keep `_EA<nnn>` / `_MA<nnn>` annotation columns

Recommended for repeat work:

options(ACSdownload.cache_dir = tools::R_user_dir("ACSdownload", "cache"))

future::plan(future::multisession, workers = 4)
bg <- get_acs_new(parallel = TRUE, ...)

tidycensus – API-driven, easier for small slices, requires a key.
EJAM – the EPA EJSCREEN / EJAM toolchain that consumes the data this package produces. EJAM’s tables_ejscreen_acs is the authoritative list of EJSCREEN tables and is mirrored here as ejscreen_acs_tables.

Documentation

Function reference

Acknowledgements

The 3.0.0 refactor – rewriting the package around the Census Bureau’s table-based summary file format, and the documentation and tooling updates that followed – was developed with extensive use of Anthropic’s Claude (via Claude Code).

ACSdownload