Fixed-width column dictionary for RESTRICTED-use MCOD files
Source:R/data-mcod_fwf_dicts.R
mcod_fwf_dicts.RdByte positions for parsing the NCHS Multiple Cause of Death restricted-use (All-County) fixed-width files, one set of rows per data year (1979-2024). Positions were verified against the raw NCHS bytes (see the narcan_c verification pipeline). Some geography fields are intentionally nested/ overlapping (e.g. `stateoc` 21-22 sits inside `countyoc` 21-25; `staters`/ `statersr` inside `countyrs`) – `readr::read_fwf()` tolerates this via explicit `fwf_positions()`.
Format
A data frame with 5 columns
- name
string, variable name (NBER short names, kept stable for backward compatibility)
- type
string, single-letter readr column type ("c" character, "n" numeric)
- start
integer, 1-indexed starting byte position; `NA` if the field is named for that year but has no readable bytes (see below)
- end
integer, ending byte position; `NA` under the same condition
- year
integer, data year of this dictionary
A few field-years are named but carry no position (`start` and `end` both `NA`), so the column exists for parity across years but imports as all-`NA`. `start` and `end` are always missing together. This is the restricted-tier counterpart of the public tier's `suppressed` flag; see [mcod_public_fwf_dicts].
Position notes
A few columns keep their historical NBER names but change meaning across a coding-era boundary – the name is stable, the semantics are not:
`ucr130` is a general all-ages recode in 1999-2001 and becomes infant-only only from data year 2002; do not filter it as infant-only for 1999-2001.
`racer5` at byte 450 holds the bridged Race Recode 5 from 2003 through 2020 but the single-race (1997-OMB) Race Recode 6 from 2022; the two are not comparable across the 2021 boundary. Before the 2003/2004 layout fix the 2003 entry was mislabeled `racer2`; the byte was always right, the name was not. `racer2` proper is a different field and ends at data year 2002.
`certifier` (110), `tobacco_use` (142) and `pregnancy_status` (143) run from data year 2003, not 2005. Before that same fix the dictionary omitted them for 2003-2004 even though the bytes are populated.
`certifier` and `tobacco_use` are blank for data year 2013 in the **restricted** tier – a one-year gap in the NCHS source, not a coding change (2012 and 2014 are populated). The positions stay declared and the columns import as `NA`. The public tier blanks both fields in *every* year as a standing suppression, so 2013 reads `NA` there for a different reason.
`hspanicr` (Hispanic Origin/Race Recode) is 1 byte at 488 through **2021** and 2 bytes at 487-488 from 2022 (14-category scheme). 2021 is declared at 488 like the years before it; an earlier version of this note said "through 2020" and omitted 2021 entirely.
`racer40` (single-race detail) appears to carry real data from data year **2011**, though that onset is established only by exclusion – the bytes are neither blank nor a state distribution from 2011 – which is a weaker standard than the 2003-2007 finding below, and 2012-2017 has not been probed at all. Bytes 489-490 are blank 2008-2010, and for **2003-2007** they hold the expanded state of occurrence in the legacy NCHS numeric coding, not race at all. Those five years are therefore named but carry no position and import as all-`NA`; keeping the name preserves a single column set across 2003-2019. Published narcan versions through 0.5.3 declared 489-490 for 2003-2011 and so returned state codes under `racer40` for 2003-2007 – discard that column for those years. Note the NCHS 2003-latest restricted layout dates the field to 2018, which the data contradicts in the other direction.