NHANES variable map — the reproduction’s own crosswalk

Generated by R/07_variable_map.R from the files in data/raw/. Do not edit by hand. Cycle availability, labels and missing-value conventions below were read out of the .xpt files, not copied from documentation.

Every variable this reproduction touches, and for each one the four things students most often get wrong:

1. Where it lives — which component file, and therefore which download
2. What its raw codes mean — and whether a 77-looking value is a code or a measurement
3. Whether it exists in every cycle — several do not
4. What we did to it — the line in our pipeline that reads or transforms it

This page is a method, not just an answer. When you meet an unfamiliar NHANES variable in your own paper, these are the questions to ask of it, in this order. The loop runs: paper sentence → operational concept → NHANES variable → component → that cycle’s codebook → raw .xpt → your implementation → cohort consequence.

Three different kinds of link, which are easy to conflate:

Cycle letters: E 2007-08 · F 2009-10 · G 2011-12 · H 2013-14 · I 2015-16 · J 2017-18. The file name carries the suffix (BMX_H.xpt); the variable name inside usually does not (BMXBMI in every cycle).

Join key

Eligibility

Survey design

⚠ WTSAF2YR — A design variable living in a laboratory file. Build the survey design on the full frame, then subset() - never on the subsample alone.

Case definition (FLI)

Cardiometabolic 1 (adiposity)

Cardiometabolic 2 (glycaemia)

Cardiometabolic 3 (blood pressure)

⚠ BPXDI1 — A diastolic reading of 0 is a measurement. NHANES records a failure to obtain a reading as NA, separately. Do not na_if(., 0).

Cardiometabolic 4 (triglycerides)

Cardiometabolic 5 (HDL)

Exposure

⚠ RIDRETH3 — This is what identifies Asian participants for the paper’s lower BMI threshold - so the rule cannot be applied in 2007-2010 at all.

Alcohol exclusion

⚠ ALQ130 — CDC’s SAS Label reads “Avg # alcohol drinks/day - past 12 mos”. CDC’s English Text for the same variable reads “During the past 12 months, on those days that {you} drank alcoholic beverages, on the average, how many drinks did {you} have?” — that is drinks per drinking day. Applying a per-day threshold to it directly over-excludes ~2,000 people here; average intake needs frequency x quantity. The short label is not the question. Read the English Text, every time.

⚠ ALQ121 — Its reserved codes are 77 (Refused) and 99 (Don’t know), not the 777/999 used by ALQ130 — a rule hard-coded to c(777, 999) misses them. And 7 and 9 are ordinary answers here: 7 = “Once a month”, 9 = “3 to 6 times in the last year”. CDC’s own Analytic Notes state “ALQ120Q and ALQ120U became ALQ121” in 2017-18, so this is an instrument change, not a missing column.

Covariate

Not every variable exists in every cycle

Verified against the files, not assumed. bind_rows() will happily fill an absent column with NA for a third of your sample, and that is a changing instrument, not Which cycle’s codebook are you reading?** For these variables that is not a rhetorical question, so every applicable cycle is linked separately.

Reserved codes are not a property of NHANES. They are a property of the variable.

The single most dangerous piece of half-learned NHANES advice is “strip 777 and 999”. Applied to a laboratory variable it deletes real measurements. Below, every sentinel-looking value this reproduction’s variables actually contain, classified by whether it sits outside the variable’s own legitimate range.

What this table does and does not license. For the variables used in this reproduction, the laboratory and examination measurements shown above use NA for missing, while several questionnaire variables use variable-specific reserved codes. Do not generalise either pattern to another variable — including another variable in the same component, or the same variable in another cycle. Check its codebook.

A 77 in LBXTR is a triglyceride of 77 mg/dL. A 77 in ALQ121 is a refusal. Same digits, opposite meaning. The point of this table is not a replacement rule to memorise; it is that there is no such rule, and the codebook decides.

The outcome does not live in NHANES

Downloaded by R/01_download.R from https://ftp.cdc.gov/pub/Health_Statistics/NCHS/datalinkage/linked_mortality. Follow-up ends 31 December 2019, per the NCHS 2019 linkage methodology document — not inferred from the filename.

Regenerate with Rscript R/07_variable_map.R. Codebook links use the current CDC path .../Public/<year>/DataFiles/<COMP>_<suffix>.htm#<VARIABLE>; the older wwwn.cdc.gov/Nchs/Nhanes/<cycle>/ form no longer resolves.