Datasets:

hesscl
/

quackrfss

license: mit
language:
  - en
tags:
  - brfss
  - health
  - survey
  - public-health
  - epidemiology
  - cdc
  - united-states
  - duckdb
pretty_name: BRFSS 1990–2024
size_categories:
  - 10M<n<100M
task_categories:
  - tabular-classification
  - tabular-regression

🦆 BRFSS 1990–2024

Behavioral Risk Factor Surveillance System (BRFSS) survey microdata for all 35 years of publicly available data (1990–2024), converted from CDC SAS Transport (XPT) format to Parquet. ~10.1 million respondents.

Source pipeline: hesscl/quackrfss

🚀 Quick start

No account, no download, no build. Just DuckDB:

import duckdb
con = duckdb.connect()

# Single year
con.sql("""
    SELECT GENHLTH_lbl, COUNT(*) AS n
    FROM read_parquet('hf://datasets/hesscl/quackrfss/data/BRFSS_2024.parquet')
    GROUP BY 1 ORDER BY 2 DESC
""").show()

# Trend across all years
con.sql("""
    SELECT
        YEAR,
        ROUND(100.0 * COUNT(*) FILTER (WHERE GENHLTH_lbl IN ('Fair', 'Poor'))
              / COUNT(*), 1) AS pct_fair_poor
    FROM read_parquet('hf://datasets/hesscl/quackrfss/data/BRFSS_*.parquet')
    WHERE GENHLTH_lbl IS NOT NULL
    GROUP BY 1 ORDER BY 1
""").show()

# Load a year into pandas
df = con.sql("""
    SELECT * FROM read_parquet('hf://datasets/hesscl/quackrfss/data/BRFSS_2024.parquet')
""").df()

📦 Dataset structure

Files

One Parquet file per year: data/BRFSS_{year}.parquet (1990–2024).

Years	Files	Approx. rows	Weight variable
2011–2024	`BRFSS_2011.parquet` … `BRFSS_2024.parquet`	400k–510k/year	`_LLCPWT`
1990–2010	`BRFSS_1990.parquet` … `BRFSS_2010.parquet`	80k–450k/year	`_FINALWT`

Columns

Every file includes:

YEAR (int16) — survey year
Raw numeric variables (float32) — original CDC-coded values (e.g. GENHLTH = 3)
*_lbl companion columns (dict<int8, string>) — human-readable label for each categorical variable (e.g. GENHLTH_lbl = 'Good'). Dictionary-encoded for compact storage.

Variable sets differ across years (BRFSS adds and drops questions). Columns absent in a given year simply aren't present in that year's file.

Key variables

Variable	Description
`GENHLTH` / `GENHLTH_lbl`	General health (Excellent → Poor)
`_STATE` / `_STATE_lbl`	State FIPS code
`_LLCPWT`	Final survey weight (2011–2024)
`_FINALWT`	Final survey weight (1990–2010)
`SEX` / `SEX_lbl`	Sex of respondent
`AGE` / `_AGEG5YR_lbl`	Age / age group
`SMOKE100`	Ever smoked 100+ cigarettes
`DIABETE3` / `DIABETE4`	Ever told have diabetes
`BPHIGH4`	Ever told blood pressure high

⚠️ Comparability notes

2011 methodology change: BRFSS introduced combined landline + cellphone sampling in 2011 and a new weighting methodology (_LLCPWT). Pre- and post-2011 data are not directly comparable without adjustment.
2020: COVID-19 forced telephone-only collection and reduced response rates.
1999: No value-label columns (*_lbl) — the source SAS file for this year contains no parseable value mappings.
Variable drift: Questions are added and dropped year to year. Always check which years a variable appears in before running cross-year analyses.

🔬 Source data

BRFSS data is collected annually by state health departments in collaboration with CDC. Raw XPT files are published at: https://www.cdc.gov/brfss/annual_data/annual_data.htm

This dataset was built using quackrfss, which downloads the XPT files, parses value labels from SAS format and sasout files, and converts to Parquet with *_lbl companion columns.

📄 License

BRFSS data is produced by the US Centers for Disease Control and Prevention and is in the public domain as a work of the US federal government. Pipeline code is MIT licensed.

📝 Citation

If you use this dataset, please cite the CDC BRFSS program:

Centers for Disease Control and Prevention (CDC). Behavioral Risk Factor Surveillance System Survey Data. Atlanta, Georgia: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, 1990–2024.