v0.3.8

[mountainMath]

Release branch for v0.3.8.

Breaking changes

• get_tongfen_ca_census now honours its base_geo, na.rm, tolerance, crs and data_transform arguments, all of which were silently ignored. Most visibly, the documented default base_geo = NULL now returns data without geographic information, where previously the geography of the first dataset came back regardless.
• Removed area_mismatch_cutoff from get_tongfen_ca_census and get_tongfen_correspondence_ca_census, it never had any effect. check_tongfen_areas is the way to inspect area mismatches.

Performance

• Correspondence tables are built via a vectorised connected components pass instead of a row-by-row union-find. Dissemination blocks for a large province go from minutes to seconds.
• The "statcan" method no longer downloads census geometries it does not use, dissolving skips regions that don't need merging, and check_tongfen_areas / aggregate_correspondences got faster.

US census

• New get_tongfen_correspondence_us_census, for correspondence tables without also fetching the data.

• Tract correspondence reaches back to the 1990 census (dec1990). Correspondence layer only, the Census Bureau retired the 1990 API endpoint so the data itself has to come from elsewhere (NHGIS) and be combined via tongfen_aggregate.

• County subdivisions can be matched across the 2010 and 2020 censuses, previously only 2000 and 2010.

• Correspondence tables no longer chain regions together over slivers. The relationship files are geometric overlays listing every intersection, including boundaries that only shifted slightly, and matching those up merged unrelated regions into one common geography. New min_area_share argument, default 0.01, applied in memory at parse time so cached files stay byte-identical and min_area_share = 0 reproduces the old behaviour. No region is ever dropped: if all of a region's parts are slivers, its largest part is kept.

  Common geographies at cutoff 0 / 0.01 / 0.05, region counts constant across all three:

  	regions	0	0.01	0.05
  RI 1990→2000 tract	232	130	179	200
  RI 2000→2010 tract	234	132	192	207
  RI 2010→2020 tract	248	60	198	234
  RI 2010→2020 cousub	40	8	40	40
  VT 1990→2000 tract	181	106	152	166
  VT 2000→2010 tract	179	42	162	174
  VT 2010→2020 tract	187	26	151	168
  VT 2010→2020 cousub	255	62	238	252

  The default is calibrated on Rhode Island, whose towns provably did not change between 2010 and 2020: 0.01 is the value that recovers exactly its 40 one-to-one county subdivision matches.

• Bug fix uncovered by the above: the 2010→2020 tract relationship file has 18 columns but was read with a 15-character col_types string, so TRACT_2020 was parsed as a number and lost its leading zeros (051000 → 51000). Those malformed GEOIDs cannot join against tidycensus, so most tracts were silently dropped from any result tongfen'd into 2020. 246 of Rhode Island's 250 and 53 of Vermont's 193 2020 tract identifiers were affected.

Other fixes

• get_tongfen_correspondence_ca_census gained a crs argument for the spatial intersections, default 3347.
• Missing geographic identifiers no longer merge unrelated regions into one common geography.
• Fixed a crash when tongfen-ing census tracts across non-adjacent censuses.
• Fixed get_tongfen_census_ct, get_tongfen_census_da and get_tongfen_ca_census_ct_from_da erroring out with geo_format=NA.
• US county subdivision data errors out up front on censuses it can't be matched across, instead of failing with "Did not find matching geographic identifiers" after downloading the comparability file and the census data.

Checks

R CMD check --as-cran: 1 NOTE (HTML Tidy toolchain artifact, not package content). All tests pass.

🤖 Generated with Claude Code