Repository navigation
This is the final ACS 2018-2022 release; it is frozen from here on.
The ACS 2020-2024 vintage version will be released as 4.2024.0,
which is where development continues.
New Features
-
Zip code analysis:
ejamit(zipcode = 10605), for example, works. Zip codes are converted
to Census ZCTA polygons by the newshapes_from_zip()helper and analyzed like
any other shapefile, andejam2report()describes the places as zip codes and
maps their boundaries without needing theshpparameter. This automates the
workflow that the Zipcodes article documented as manual steps (#482). -
Updated FRS facility data to the September 2026 EPA snapshot, with refreshed
examples and update instructions. -
Web-app completion estimates now reflect measured live performance, and report a
lower bound instead of an expected time beyond the measured range (#513).
Dataset Fix
- Language-spoken counts (number of people speaking given language) were shown
incorrectly and are now fixed. The counts from the tract-only ACS table C16001 (lan_universe,
lan_spanish, and the otherlan_*counts) are now correctly apportioned to block groups
by population share instead of repeating each tract's total on every block
group, so sums across block groups (inejamit()results and the pipeline's
acs_by_tract,acs_by_county, andacs_by_statelayers) no longer come out
about three times too high. The packagedblockgroupstatsdata and saved
example outputs now carry the corrected counts (#596).
Bug Fixes
-
Population-share summaries are now correct for small numbers of sites (#137).
-
Percentiles no longer depend on which operating system the analysis runs on.
A raw score is normally a population-weighted average, so it carries a few
units in the last place of rounding error, and how much differs between macOS,
Linux, and Windows. Because percentile lookup is a step function, a score that
lands exactly on a cutoff could fall either side of it, and where the lookup
table has several percentiles tied at one cutoff the reported percentile could
move by the whole width of that tie block. One test site's state asthma
percentile differed by 7 points between operating systems for this reason.
Scores that sit within floating-point noise of a cutoff are now snapped onto
it, so every platform reports the same percentile, and a tied cutoff correctly
reports the lowest of the tied percentiles as documented (#555).This corrects saved percentiles in the eight
testoutput_ejamit_*and
testoutput_doaggregate_*datasets, which had recorded the higher end of a tie
block. Only percentile columns changed. It also meanslookup_pctile()handles
the ACS22pctdisabilityboundary case without needing thesignif_digits
argument. -
A custom report title is now kept in the downloaded report, not just in the app (#458).
-
The notes tab of the Excel workbook now says how the sites were selected. It
had never done so:buffer_desc_from_sitetype()only appended that detail when
the description so far was empty, which none of its branches can produce, and
the test inside was inverted as well. A SIC analysis of latitude/longitude
sites now reads "Locations defined by latitude, longitude and radius, based on
EPA-regulated facilities by SIC code (industry type)" instead of stopping at
the radius. The detail is left off when it would only restate the site type,
so plain shapefile analyses do not read "Polygons defined by shapefile, based
on shapefile". -
SIC and MACT analyses are now described again in reports and the Excel notes.
site_method2text()lowercases its input, but its SIC and MACT branches
compared against the uppercase spellings. -
Built-in state and county boundaries now work offline (#527).
-
County boundaries from
shapes_from_fips()no longer fail with "missing value
where TRUE/FALSE needed" when the sf package was not yet loaded. -
The web app no longer requests about a dozen files it does not serve (a web
manifest, extra favicons, and Esri map-widget images and fonts from the report
stylesheet), which had shown as 404 errors in the browser console (#588). -
ejam2report()now fetches FIPS boundaries whensite_methodis given as
"fips" rather than "FIPS". The two gates that rebuild those polygons were
case-sensitive, so a lowercase spelling silently produced an unmapped report.
Matches how the zip code gates added for #482 already behave. -
The release workflow now refuses to publish a draft Release whose target is not
the head ofmain, and its dry run shows that target and fails on a mismatch. -
A GitHub outage no longer looks like missing data. When the API cannot list a
release's assets,download_latest_arrow_data()now says so and retries,
instead of treating the empty answer as an empty release and reporting a
missingquaddata.arrowmuch later. -
Docker build and deployed app now verifies PDF report download works (#510).
-
Trying to upload a shapefile that is point-based now tells user to use
lat/lon point upload option, and geometry rule is centralized & tested (#550). -
Web-app article text showing URLs now shows them as clickable links (#599).
-
Uploaded polygon data stay in
sfform through the web-app map path, avoiding
anst_geometry()console error (#136). -
Count of sites with N "high" scores was confusing.
count_sites_with_n_high_scores()now starts at 1.05 ratio cutoff (#546). -
R CMD check problems exposed by full CI are now resolved across examples,
dependencies, line endings, and portable fixtures (#548). -
NAICS function unit tests of
naics_from_any(21112, website_scrape = TRUE)
now skip when naics.com blocks the CI runner (#560). -
Installed-package vignettes now use working documentation URLs (#565).
-
Quick-install setup steps now time out promptly instead of hanging for an
hour (#585). -
The Basics - Quick Start Guide article now links to a Community Report example (#600).
-
In Using EJAM for Analysis in R
and other articles, fixed about 20 broken or incomplete links and text (#602). -
Rewrote the developer articles Releasing a New Version of EJAM
and Deploying the Web App to AWS
to match the current release and deploy process (#631, #632).
Full changelog: NEWS.md