utility to calculate annually for EJSCREEN the updated ACS data available at only tract resolution (% disability & language detail)

Usage

calc_blockgroupstats_from_tract_data(
  yr,
  tables = c("B18101", "C16001", "B27010"),
  formulas = NULL,
  dropMOE = TRUE,
  acs_raw = NULL,
  tract_weight_source = c("decennial2020", "acs")
)

Arguments

yr: endyear of ACS 5-year survey to use, inferred if omitted
tables: ACS tract tables used for tract-derived indicators, typically "B18101", "C16001", and "B27010" for disability, detailed language, and health insurance.
formulas: default includes formulas for disability-related and language-related indicators calculated from tract-level ACS variables. This is a vector of string formulas.
dropMOE: logical, whether to drop not retain the margin of error information for each ACS variable
acs_raw: optional raw ACS table list or bg_acs_raw pipeline object previously created by download_bg_acs_raw(). If supplied, no ACS download is performed for tract-resolution tables.
tract_weight_source: source for blockgroup-to-tract apportionment weights. "decennial2020" uses 2020 Decennial Census population weights, matching the legacy EJSCREEN approach. "acs" uses same-vintage ACS blockgroup population from acs_raw or downloads it when needed.

Value

data.table, one row per blockgroup (not tract)

Details

This is now typically orchestrated by calc_ejscreen_dataset() and the staged pipeline recipe/config helpers documented in data-raw/run_ejscreen_pipeline_*.R.

Relies on the function get_acs_new() which is available from the package ACSdownload (on github) as ACSdownload::get_acs_new()

Needs Census API key for tidycensus::get_decennial().

Takes some time to download data for every State!

First get tract counts, then apportion every tract count into blockgroup counts (the tract count times the blockgroup's share of the tract population, bgwt), then carry each tract's percentages onto its blockgroups.

Apportioning applies to the language counts from C16001 (lan_universe, lan_spanish, etc.) exactly as it does to disability and health insurance. map_headernames marks all of these as "sum of counts", so doaggregate() and calc_acs_by_geography() add them up across blockgroups. A tract total repeated on every blockgroup would be counted once per blockgroup, about three times over, in every buffer, county, and state total (issue EJAM#596). Apportioned counts add back up to the tract total, to rounding, which is also how EPA's EJScreen block group layer reports language counts.

Percentages such as pctlan_spanish are calculated at tract level and repeated on each blockgroup in the tract. That equals the ratio of the apportioned counts wherever bgwt > 0, and it keeps the tract rate on zero-population blockgroups instead of reporting 0 there, the same convention pctdisability uses. Apportioned counts are rounded to whole persons after the percentages are set.

For ACS 2022 and later, Connecticut ACS tract FIPS use planning-region county equivalents while 2020 Decennial blockgroup FIPS use the older county equivalents. When tract_weight_source = "decennial2020" and acs_raw is available, the function detects this mismatch and remaps the 2020 Decennial weights to the same ACS blockgroup suffixes, using same-vintage ACS blockgroup population weights only for ambiguous or unmatched cases.