Dataset Viewer
Auto-converted to Parquet Duplicate
cycle
stringlengths
4
9
nsfg_cycle
stringclasses
8 values
fieldwork
stringlengths
4
19
age_universe
stringclasses
4 values
female_respondents
int32
5.55k
12.3k
male_respondents
int32
4.37k
10.4k
⌀
pregnancies
int32
8.25k
23k
⌀
files
int32
1
6
weight
stringlengths
4
76
stratum
stringclasses
6 values
cluster
stringclasses
7 values
variance_method
stringclasses
7 values
pooling_guidance
stringclasses
5 values
doc_url
stringlengths
45
52
1973
Cycle 1
1973
Ever-married women 15-44 (and never-married mothers)
9,797
null
null
1
PSWT
FAM_PSEUDO
FAM_PAIRNUM
Taylor linearization with pseudo-strata and paired PSUs (Cycle 1 codebook, p. 7)
Separate survey; not designed for pooling with other cycles (different universe: ever-married women in 1973-1976, women 15-44 in 1982-2002).
https://www.cdc.gov/nchs/nsfg/nsfg_cycle1.htm
1976
Cycle 2
1976
Ever-married women 15-44 (and never-married mothers)
8,611
null
23,042
3
WTFINAL
STRAT (cyc2psu)
PSU (cyc2psu)
Taylor linearization; PSU/stratum in the separate cyc2psu file (join on RESPID)
Separate survey; not designed for pooling with other cycles (different universe: ever-married women in 1973-1976, women 15-44 in 1982-2002).
https://www.cdc.gov/nchs/nsfg/nsfg_cycle2.htm
1982
Cycle 3
1982
Women 15-44
7,969
null
13,768
2
FINALWT
PSEUDOST
PSUCODE
Taylor linearization with PSUCODE / PSEUDOST
Separate survey; not designed for pooling with other cycles (different universe: ever-married women in 1973-1976, women 15-44 in 1982-2002).
https://www.cdc.gov/nchs/nsfg/nsfg_cycle3.htm
1988
Cycle 4
1988
Women 15-44
8,450
null
15,759
2
W5 (two implied decimals)
null
null
Balanced repeated replication with 100 replicate weights REPWGT1-REPWGT100 (two implied decimals)
Separate survey; not designed for pooling with other cycles (different universe: ever-married women in 1973-1976, women 15-44 in 1982-2002).
https://www.cdc.gov/nchs/nsfg/nsfg_cycle4.htm
1995
Cycle 5
1995
Women 15-44
10,847
null
21,332
3
POST_WT
COL_STR
PANEL
Taylor linearization, with-replacement design: NEST COL_STR PANEL, WEIGHT POST_WT (Cycle 5 User's Guide, App. 2)
Separate survey; not designed for pooling with other cycles (different universe: ever-married women in 1973-1976, women 15-44 in 1982-2002).
https://www.cdc.gov/nchs/nsfg/nsfg_cycle5.htm
2002
Cycle 6
Jan 2002 - Mar 2003
Women and men 15-44
7,643
4,928
13,593
6
FINALWGT
SEST
SECU_R (female), SECU_P (pregnancy), SECU (male)
Taylor linearization: strata SEST, clusters SECU*, weight as listed
Separate survey; compare estimates across cycles, do not pool for population totals.
https://www.cdc.gov/nchs/nsfg/nsfg_cycle6.htm
2006-2010
Continuous
Jun 2006 - Jun 2010
Women and men 15-44
12,279
10,403
20,492
3
WGTQ1Q16 (4-year); WGTQ1Q8, WGTQ5Q16, WGTQ9Q16, FINALWGT30 (partial periods)
SEST
SECU
Taylor linearization: strata SEST, clusters SECU*, weight as listed
Use WGTQ1Q16 for the full 2006-2010 file. For combinations with 2006-2010 or older see Appendix 2 of the 2013-2015 User's Guide.
https://www.cdc.gov/nchs/nsfg/nsfg_2006_2010_puf.htm
2011-2013
Continuous
Sep 2011 - Sep 2013
Women and men 15-44
5,601
4,815
9,543
3
WGT2011_2013
SEST
SECU
Taylor linearization: strata SEST, clusters SECU*, weight as listed
Use the 2-year weight for comparisons; for pooled 2011-2019 periods use the combined-period weights in 2011_2019_FemaleWgtData / 2011_2019_MaleWgtData (WGT2011_2015, WGT2013_2017, WGT2015_2019, WGT2011_2017, WGT2013_2019, WGT2011_2019), join on CASEID. Respondents 45-49 (2015-2019) have no 4-, 6- or 8-year weight.
https://www.cdc.gov/nchs/nsfg/nsfg_2011_2013_puf.htm
2013-2015
Continuous
Sep 2013 - Sep 2015
Women and men 15-44
5,699
4,506
9,358
3
WGT2013_2015
SEST
SECU
Taylor linearization: strata SEST, clusters SECU*, weight as listed
Use the 2-year weight for comparisons; for pooled 2011-2019 periods use the combined-period weights in 2011_2019_FemaleWgtData / 2011_2019_MaleWgtData (WGT2011_2015, WGT2013_2017, WGT2015_2019, WGT2011_2017, WGT2013_2019, WGT2011_2019), join on CASEID. Respondents 45-49 (2015-2019) have no 4-, 6- or 8-year weight.
https://www.cdc.gov/nchs/nsfg/nsfg_2013_2015_puf.htm
2015-2017
Continuous
Sep 2015 - Sep 2017
Women and men 15-49
5,554
4,540
9,553
3
WGT2015_2017
SEST
SECU
Taylor linearization: strata SEST, clusters SECU*, weight as listed
Use the 2-year weight for comparisons; for pooled 2011-2019 periods use the combined-period weights in 2011_2019_FemaleWgtData / 2011_2019_MaleWgtData (WGT2011_2015, WGT2013_2017, WGT2015_2019, WGT2011_2017, WGT2013_2019, WGT2011_2019), join on CASEID. Respondents 45-49 (2015-2019) have no 4-, 6- or 8-year weight.
https://www.cdc.gov/nchs/nsfg/nsfg_2015_2017_puf.htm
2017-2019
Continuous
Sep 2017 - Sep 2019
Women and men 15-49
6,141
5,206
10,215
3
WGT2017_2019
SEST
SECU
Taylor linearization: strata SEST, clusters SECU*, weight as listed
Use the 2-year weight for comparisons; for pooled 2011-2019 periods use the combined-period weights in 2011_2019_FemaleWgtData / 2011_2019_MaleWgtData (WGT2011_2015, WGT2013_2017, WGT2015_2019, WGT2011_2017, WGT2013_2019, WGT2011_2019), join on CASEID. Respondents 45-49 (2015-2019) have no 4-, 6- or 8-year weight.
https://www.cdc.gov/nchs/nsfg/nsfg_2017_2019_puf.htm
2022-2023
2022-2023
Jan 2022 - Dec 2023
Women and men 15-49
5,586
4,371
8,247
3
WGT2022_2023
VEST
VECL
Taylor linearization: strata VEST, clusters VECL, weight WGT2022_2023
NCHS does not advise combining 2022-2023 with earlier releases (break in fieldwork Oct 2019 - Dec 2021 and multimode design); no combined weights exist (2022-2023 FAQ, Q12).
https://www.cdc.gov/nchs/nsfg/nsfg-2022-2023-puf.htm

NSFG 1973–2023 in Parquet (with variable dictionaries and survey design)

This dataset contains every public-use data file of the U.S. National Survey of Family Growth (NSFG), conducted by the CDC's National Center for Health Statistics (NCHS), from Cycle 1 (1973) through the 2022–2023 release. Each file has been converted from the NCHS fixed-width ASCII or SAS7BDAT original to Parquet, one Parquet file per original file, under the original file name. It also includes:

  • a variable dictionary for each cycle, built from the official NCHS SAS setup programs (column positions, labels, value formats) and, where NCHS publishes machine-readable codebooks (1995, 2002, 2022–2023), the codebook question text, universe and value tables with published frequencies;
  • the 2011–2019 combined-period weight files (4-, 6- and 8-year weights);
  • the 1976 sample-design file, the 1995 fieldwork (sampling-frame) file and the 2002 household and revised insurance supplements;
  • a cycle table with the weight, stratum and cluster variables of every cycle and the NCHS guidance on pooling cycles.

Built for the Bioanalysis R package. It works with any tool that reads Parquet: DuckDB, Arrow, Polars, pandas or R.

Cycles 12 (1973, 1976, 1982, 1988, 1995, 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023)
Files 37 Parquet data files + 26 dictionary files (13 cycle folders × 2) + 2 metadata tables
Interviews 94,177 female respondents, 38,769 male respondents (2002 onward), 154,902 pregnancy records
Variables 49,988 file × variable dictionary rows; 371,138 value-label rows
Values Original public-use codes; nothing recoded, imputed or dropped

Layout

data/<cycle>/<component>/<FILE>.parquet
data/<cycle>/Dictionary/VARIABLES.parquet     one row per file x variable
data/<cycle>/Dictionary/VALUES.parquet        one row per file x variable x code
metadata/FILES.parquet                        every file: rows, columns, grain, join keys, source URL, MD5
metadata/CYCLES.parquet                       every cycle: fieldwork, universe, counts, design variables, pooling guidance
pipeline/                                     the R scripts that rebuilt everything from the NCHS sources

<cycle> is the NSFG release (1973 … 2022-2023, plus 2011-2019 for the combined-period weights) and <component> the NCHS file type; Data by category below lists every file. The 1976 and 1982 originals are single hierarchical files holding respondent and pregnancy records; they are split by record type into <FILE>_FemResp and <FILE>_Preg using the record-type columns given in the NCHS setup programs and codebooks (1976: column 6, 1–4 respondent / 5 pregnancy; 1982: columns 1499–1500, 00 respondent / 01–17 pregnancy).

Data by category

NCHS publishes each NSFG cycle as a small set of files: a female respondent file, a male respondent file (from 2002), a female pregnancy (interval) file with one record per pregnancy, and, for some cycles, weight, design and supplementary files. The folder under each cycle (<component>) keeps that grouping: FemResp, Male, FemPreg, Weights, Design, Household, Supplementary, Fieldwork. All files are tabular (49,988 columns in all); every column is numeric except 38 text columns (CASEID, DEVICE_TYPE, INTVWYEAR, PHASE, QUARTER, SCREENTIME, TEMP_NLAMOUNT).

Category Contents Folders Files Cycles
Female respondent files One record per woman interviewed: household screener items, marriage and cohabitation, pregnancies and births summarised, contraception, infertility, sexual activity, health, recodes, and the weight and design variables. FemResp 12 1973, 1976, 1982, 1988, 1995, 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023
Male respondent files One record per man interviewed (from 2002): fatherhood, partners, contraception, sexual activity, health, recodes, weight and design variables. Male 7 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023
Female pregnancy (interval) files One record per pregnancy reported by a female respondent: outcome, dates, gestation, birth weight, prenatal care, breastfeeding, wantedness; carries the respondent's weight and design variables. FemPreg 11 1976, 1982, 1988, 1995, 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023
Weights and sample design files Combined-period weights for pooling the 2011-2019 releases (4-, 6- and 8-year), and the Cycle 2 (1976) PSU/stratum file. Design, Weights 3 1976, 2011–2019
Supplementary files Cycle 6 (2002) household and revised insurance variables for women and men, and the Cycle 5 (1995) fieldwork file covering the whole sampling frame. Fieldwork, Household, Supplementary 4 1995, 2002

Cycles at a glance (from metadata/CYCLES.parquet):

Cycle NCHS name Fieldwork Universe Women Men Pregnancy records Files
1973 Cycle 1 1973 Ever-married women 15-44 (and never-married mothers) 9,797 – – 1
1976 Cycle 2 1976 Ever-married women 15-44 (and never-married mothers) 8,611 – 23,042 3
1982 Cycle 3 1982 Women 15-44 7,969 – 13,768 2
1988 Cycle 4 1988 Women 15-44 8,450 – 15,759 2
1995 Cycle 5 1995 Women 15-44 10,847 – 21,332 3
2002 Cycle 6 Jan 2002 - Mar 2003 Women and men 15-44 7,643 4,928 13,593 6
2006–2010 Continuous Jun 2006 - Jun 2010 Women and men 15-44 12,279 10,403 20,492 3
2011–2013 Continuous Sep 2011 - Sep 2013 Women and men 15-44 5,601 4,815 9,543 3
2013–2015 Continuous Sep 2013 - Sep 2015 Women and men 15-44 5,699 4,506 9,358 3
2015–2017 Continuous Sep 2015 - Sep 2017 Women and men 15-49 5,554 4,540 9,553 3
2017–2019 Continuous Sep 2017 - Sep 2019 Women and men 15-49 6,141 5,206 10,215 3
2022–2023 2022-2023 Jan 2022 - Dec 2023 Women and men 15-49 5,586 4,371 8,247 3

Plus 2011-2019 (combined-period weights): 2011_2019_FemaleWgtData (22,995 rows), 2011_2019_MaleWgtData (19,067 rows). The 1973 pregnancy file is not distributed by NCHS (ICPSR only).

File-naming convention. Each Parquet file keeps the NCHS data-file name without its extension:

Cycles NCHS pattern Examples
1973, 1976, 1982 <YYYY>NSFGData (one hierarchical file per cycle; split here by record type into _FemResp and _Preg); 1976 design file cyc2psu 1976NSFGData_FemResp, 1982NSFGData_FemResp, 1973NSFGData
1988, 1995 <YYYY>FemRespData, <YYYY>PregData; C5FieldworkData (Cycle 5) 1988FemRespData, 1995FemRespData, 1988PregData
2002 2002FemResp, 2002FemPreg, 2002Male, plus supplements under their NCHS names 2002FemResp, 2002FemPreg, 2002HHvars
2006–2010 2006_2010_<FemResp|FemPreg|Male> 2006_2010_FemResp, 2006_2010_FemPreg, 2006_2010_Male
2011–2013, 2013–2015, 2015–2017, 2017–2019 <YYYY>_<YYYY>_<FemResp|FemPreg|Male>Data 2011_2013_FemRespData, 2013_2015_FemRespData, 2015_2017_FemRespData
2011–2019 2011_2019_<Female|Male>WgtData 2011_2019_FemaleWgtData, 2011_2019_MaleWgtData
2022–2023 NSFG-2022-2023-<FemResp|FemPreg|MaleResp>PUFData NSFG-2022-2023-FemRespPUFData, NSFG-2022-2023-FemPregPUFData, NSFG-2022-2023-MaleRespPUFData

Expand a category to see every file: cycle, NCHS file name, title, rows, columns, grain and join key. Path: data/<cycle>/<component>/<FILE>.parquet.

Female respondent files (FemResp): 12 files, cycles 1973, 1976, 1982, 1988, 1995, 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023
Cycle File Title (NCHS) Rows Columns Grain Join key
1973 1973NSFGData Cycle 1 (1973) Female Respondent File (currently and formerly married women 15-44) 9,797 450 one row per respondent INT_ID
1976 1976NSFGData_FemResp Cycle 2 (1976) Female Respondent File (respondent records, record type 1-4) 8,611 468 one row per respondent RESPID
1982 1982NSFGData_FemResp Cycle 3 (1982) Female Respondent File (respondent records, record type 00) 7,969 741 one row per respondent CASEID
1988 1988FemRespData Cycle 4 (1988) Female Respondent File 8,450 1,437 one row per respondent CASEID
1995 1995FemRespData Cycle 5 (1995) Female Respondent File 10,847 5,754 one row per respondent CASEID
2002 2002FemResp Cycle 6 (2002) Female Respondent File 7,643 3,087 one row per respondent CASEID
2006–2010 2006_2010_FemResp 2006-2010 NSFG Female Respondent File 12,279 3,741 one row per respondent CASEID
2011–2013 2011_2013_FemRespData 2011-2013 NSFG Female Respondent File 5,601 3,094 one row per respondent CASEID
2013–2015 2013_2015_FemRespData 2013-2015 NSFG Female Respondent File 5,699 3,207 one row per respondent CASEID
2015–2017 2015_2017_FemRespData 2015-2017 NSFG Female Respondent File 5,554 3,024 one row per respondent CASEID
2017–2019 2017_2019_FemRespData 2017-2019 NSFG Female Respondent File 6,141 2,609 one row per respondent CASEID
2022–2023 NSFG-2022-2023-FemRespPUFData 2022-2023 NSFG Female Respondent File 5,586 1,912 one row per respondent CaseID
Male respondent files (Male): 7 files, cycles 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023
Cycle File Title (NCHS) Rows Columns Grain Join key
2002 2002Male Cycle 6 (2002) Male Respondent File 4,928 1,993 one row per respondent CASEID
2006–2010 2006_2010_Male 2006-2010 NSFG Male Respondent File 10,403 2,801 one row per respondent CASEID
2011–2013 2011_2013_MaleData 2011-2013 NSFG Male Respondent File 4,815 3,059 one row per respondent CASEID
2013–2015 2013_2015_MaleData 2013-2015 NSFG Male Respondent File 4,506 2,940 one row per respondent CASEID
2015–2017 2015_2017_MaleData 2015-2017 NSFG Male Respondent File 4,540 2,945 one row per respondent CASEID
2017–2019 2017_2019_MaleData 2017-2019 NSFG Male Respondent File 5,206 3,009 one row per respondent CASEID
2022–2023 NSFG-2022-2023-MaleRespPUFData 2022-2023 NSFG Male Respondent File 4,371 1,157 one row per respondent CaseID
Female pregnancy (interval) files (FemPreg): 11 files, cycles 1976, 1982, 1988, 1995, 2002, 2006–2010, 2011–2013, 2013–2015, 2015–2017, 2017–2019, 2022–2023
Cycle File Title (NCHS) Rows Columns Grain Join key
1976 1976NSFGData_Preg Cycle 2 (1976) Pregnancy (Interval) File (interval records, record type 5) 23,042 80 one row per pregnancy (several rows per respondent) ID + BPREC1
1982 1982NSFGData_Preg Cycle 3 (1982) Pregnancy (Interval) File (interval records, record type 01-17) 13,768 174 one row per pregnancy (several rows per respondent) CASEID + BPREC
1988 1988PregData Cycle 4 (1988) Pregnancy (Interval) File 15,759 284 one row per pregnancy (several rows per respondent) SEQNUM + PREG_NUM
1995 1995PregData Cycle 5 (1995) Pregnancy (Interval) File 21,332 245 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2002 2002FemPreg Cycle 6 (2002) Female Pregnancy (Interval) File 13,593 243 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2006–2010 2006_2010_FemPreg 2006-2010 NSFG Female Pregnancy (Interval) File 20,492 287 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2011–2013 2011_2013_FemPregData 2011-2013 NSFG Female Pregnancy (Interval) File 9,543 278 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2013–2015 2013_2015_FemPregData 2013-2015 NSFG Female Pregnancy (Interval) File 9,358 278 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2015–2017 2015_2017_FemPregData 2015-2017 NSFG Female Pregnancy (Interval) File 9,553 248 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2017–2019 2017_2019_FemPregData 2017-2019 NSFG Female Pregnancy (Interval) File 10,215 171 one row per pregnancy (several rows per respondent) CASEID + PREGORDR
2022–2023 NSFG-2022-2023-FemPregPUFData 2022-2023 NSFG Female Pregnancy (Interval) File 8,247 111 one row per pregnancy (several rows per respondent) CaseID + PREGORDR
Weights and sample design files (Design, Weights): 3 files, cycles 1976, 2011–2019
Cycle File Title (NCHS) Rows Columns Grain Join key
1976 cyc2psu Cycle 2 (1976) Sample Design File (PSU and stratum for variance estimation) 8,611 3 one row per respondent RESPID
2011–2019 2011_2019_FemaleWgtData 2011-2019 Combined Files: Female Respondent Weights (4-, 6- and 8-year) 22,995 7 one row per respondent CASEID
2011–2019 2011_2019_MaleWgtData 2011-2019 Combined Files: Male Respondent Weights (4-, 6- and 8-year) 19,067 7 one row per respondent CASEID
Supplementary files (Fieldwork, Household, Supplementary): 4 files, cycles 1995, 2002
Cycle File Title (NCHS) Rows Columns Grain Join key
1995 C5FieldworkData Cycle 5 (1995) Fieldwork File (complete sampling frame: respondents and nonrespondents) 14,000 128 one row per respondent CASEID
2002 2002HHvars Cycle 6 (2002) Household Variables (female and male respondents) 12,571 10 one row per respondent CASEID
2002 hhpartypnew Cycle 6 (2002) Household Parent Type, 2nd version (HHPARTYP2) 12,571 3 one row per respondent CASEID
2002 c6_curr_ins Cycle 6 (2002) Revised Current Health Insurance Recode (CURR_INS) 12,571 3 one row per respondent CASEID

Hugging Face subsets and splits

On Hugging Face a subset (also called a config) is one named table: it is the drop-down list above the Dataset Viewer and the second argument of load_dataset("jamesjiadazhan/NSFG", "<subset>"). A split is a part of a subset, chosen with split=. In this dataset a split is all (the whole subset), one survey cycle (cycle_2017_2019 is the 2017–2019 release; the dictionaries also carry the cycle in their cycle column, so splits can be stacked), or, for the 2011–2019 weights, female / male.

Subset What it holds Splits
files Catalogue of the 37 data files: cycle, folder, title, rows, columns, grain, join keys, path, source URL, MD5. all
cycles One row per cycle (12): fieldwork, universe, counts, weight / stratum / cluster variables, variance method, pooling guidance. all
dictionary_variables Variable dictionary: one row per file × variable. 13, one per cycle (cycle_1973 … cycle_2022_2023); column cycle
dictionary_values Value dictionary: one row per file × variable × code. 13, one per cycle (cycle_1973 … cycle_2022_2023); column cycle
weights_2011_2019 2011_2019_FemaleWgtData and 2011_2019_MaleWgtData (22,995 / 19,067 rows, 7 columns): 2011-2019 Combined Files: Female Respondent Weights (4-, 6- and 8-year); 2011-2019 Combined Files: Male Respondent Weights (4-, 6- and 8-year). female, male
pregnancy_2022_2023 NSFG-2022-2023-FemPregPUFData (8,247 rows, 111 columns): 2022-2023 NSFG Female Pregnancy (Interval) File. all
pregnancy_2017_2019 2017_2019_FemPregData (10,215 rows, 171 columns): 2017-2019 NSFG Female Pregnancy (Interval) File. all

33 of the 37 data files are not subsets; the Dataset Viewer does not show them, but the Files and versions tab and every reader below do. There are two reasons:

  • Too wide for the viewer. 16 respondent files have 1,157–5,754 columns, and the Hugging Face viewer and datasets subsets stop at 1,000 columns: 1988FemRespData (1,437), 1995FemRespData (5,754), 2002FemResp (3,087), 2002Male (1,993), 2006_2010_FemResp (3,741), 2006_2010_Male (2,801), 2011_2013_FemRespData (3,094), 2011_2013_MaleData (3,059), 2013_2015_FemRespData (3,207), 2013_2015_MaleData (2,940), 2015_2017_FemRespData (3,024), 2015_2017_MaleData (2,945), 2017_2019_FemRespData (2,609), 2017_2019_MaleData (3,009), NSFG-2022-2023-FemRespPUFData (1,912), NSFG-2022-2023-MaleRespPUFData (1,157). For this reason the earlier subsets female_2017_2019, male_2017_2019, female_2022_2023 and male_2022_2023 were removed (the viewer could not display them); the files themselves are unchanged and are read by path as shown below.
  • Columns differ between cycles. A subset needs the same columns in all its files, but each NSFG cycle has its own questionnaire and variable list, so files of different cycles cannot share a subset. Only the two most recent pregnancy files are offered as one-file subsets; the other 17 files (3–741 columns) are read by path: 1973NSFGData, cyc2psu, 1976NSFGData_Preg, 1976NSFGData_FemResp, 1982NSFGData_Preg, 1982NSFGData_FemResp, 1988PregData, 1995PregData, C5FieldworkData, 2002FemPreg, 2002HHvars, hhpartypnew, c6_curr_ins, 2006_2010_FemPreg, 2011_2013_FemPregData, 2013_2015_FemPregData, 2015_2017_FemPregData.

Read any file by its path (listed in the files subset and in Data by category), for example with DuckDB or Arrow, selecting only the columns you need:

import duckdb
duckdb.sql("SELECT CASEID, AGER, WGT2017_2019, SECU, SEST FROM 'hf://datasets/jamesjiadazhan/NSFG/data/2017-2019/FemResp/2017_2019_FemRespData.parquet'").df()

Data dictionary

Each cycle folder has its own dictionary (the 2011–2019 weights have theirs under data/2011-2019/Dictionary/). Labels and value codes come from the official NCHS SAS setup programs (1973–2019) or SAS7BDAT labels (2022–2023); question text, universe and published frequencies come from the NCHS codebooks where they are machine-readable.

File Subset One row per Main columns
data/<cycle>/Dictionary/VARIABLES.parquet dictionary_variables file × variable file, file_label, variable, position, start/end (columns in the fixed-width original), type, label, sas_label, sas_format, question_text, universe (who was asked), variable_type, notes, n_nonmissing, in_codebook, label_source, codebook_files, doc_url
data/<cycle>/Dictionary/VALUES.parquet dictionary_values file × variable × code file, variable, format, code (as written by NCHS, including ranges such as 1-3 and ./Blank for missing), description, count (published frequency, codebooks only), source

Coverage by cycle:

Cycle Files Variables With question text / universe Variables with value labels Value rows Labels from Values from
1973 1 450 0 383 2,743 SAS setup program SAS setup PROC FORMAT
1976 3 551 0 495 9,400 SAS setup program, NCHS documentation, SAS7BDAT label SAS setup PROC FORMAT
1982 2 915 0 902 9,479 SAS setup program SAS setup PROC FORMAT
1988 2 1,721 0 1,489 18,762 SAS setup program SAS setup PROC FORMAT
1995 3 6,127 6,119 6,071 70,722 SAS setup program, codebook codebook, SAS setup PROC FORMAT
2002 6 5,339 5,323 5,316 46,141 SAS setup program, NCHS documentation, SAS7BDAT label codebook, SAS setup PROC FORMAT
2006–2010 3 6,829 0 6,755 59,041 SAS setup program SAS setup PROC FORMAT
2011–2013 3 6,431 0 5,847 23,322 SAS setup program SAS setup PROC FORMAT
2013–2015 3 6,425 0 6,251 38,889 SAS setup program SAS setup PROC FORMAT
2015–2017 3 6,217 0 6,163 40,788 SAS setup program SAS setup PROC FORMAT
2017–2019 3 5,789 0 5,633 37,295 SAS setup program SAS setup PROC FORMAT
2022–2023 3 3,180 1,264 2,025 14,556 SAS7BDAT label, codebook codebook, codebook of another file of this cycle (same SAS format)
2011–2019 2 14 0 0 0 SAS setup program –

The setup programs' "missing value recode" statements are not applied to the data: codes such as 98 (refused) and 99 (don't know) stay as published, so check VALUES before computing means. The 2022–2023 female respondent codebook was not available as text, so that file has labels but its value labels are borrowed from the male and pregnancy codebooks where the SAS format is the same (source says so) and it has no counts.

Example: file = '2017_2019_FemRespData', variable = 'CONSTAT1' (Current contraceptive status (1st priority code) (RECODE)) has 35 codes, e.g. 1 = Female sterilization, 2 = Male sterilization, 3 = Norplant or Implanon implant.

import duckdb
base = 'hf://datasets/jamesjiadazhan/NSFG/data/2017-2019/Dictionary'
duckdb.sql(f"SELECT code, description, source FROM '{base}/VALUES.parquet' "
           "WHERE file = '2017_2019_FemRespData' AND variable = 'CONSTAT1'").show()

Official NCHS documentation: each cycle's page (codebooks, user's guide, questionnaires) is in doc_url of the dictionary, of files and of cycles, for example 2022–2023 and 2017–2019; codebook_files names the codebook PDFs that were parsed.

Joining

  • Within a cycle, files join on the respondent ID. Its name changes across cycles; see metadata/FILES.parquet (id_variable, grain, join_keys, joins_to, ids_matched):
    • 1973 INT_ID; 1976 RESPID (respondent, cyc2psu) = ID (pregnancy); 1982 CASEID; 1988 CASEID (respondent) = SEQNUM (pregnancy); 1995–2019 CASEID; 2022–2023 CaseID.
    • Pregnancy files have one row per pregnancy: key = ID + pregnancy order (PREGORDR from 1995; BPREC1 1976, BPREC 1982, PREG_NUM 1988).
    • In 2002 CASEID is text in the respondent and pregnancy files ($ in the NCHS setup) but numeric in 2002HHvars, hhpartypnew and c6_curr_ins; cast before joining. The 2002 household and insurance files hold women and men together (12,571 rows, R_SEX).
    • The 1995 fieldwork file covers the whole sampling frame (14,000 women); 10,847 of its CASEIDs are respondents.
  • Female and male files never share IDs within a cycle; they are separate samples and are not linked (a man and a woman in the file are not couples).
  • Across cycles, IDs are not person identifiers: the same number can occur in different cycles (1973–1988 numbers reappear in later cycles) and nobody is followed over time. The exception is 2011–2019: IDs are unique over the four releases, so 2011_2019_FemaleWgtData / 2011_2019_MaleWgtData join to the stacked 2011–2019 respondent files on CASEID.
  • Variables differ by cycle (questionnaires changed); never assume that a name means the same across cycles without checking each cycle's dictionary.

Quick start

DuckDB (Python, R or CLI). No download is needed:

import duckdb
con = duckdb.connect()
base = "hf://datasets/jamesjiadazhan/NSFG/data"
con.sql(f"""
    SELECT f.CASEID, f.AGER, f.CONSTAT1, f.SEST, f.SECU, f.WGT2017_2019,
           count(p.PREGORDR) AS n_preg_records
    FROM read_parquet('{base}/2017-2019/FemResp/2017_2019_FemRespData.parquet') f
    LEFT JOIN read_parquet('{base}/2017-2019/FemPreg/2017_2019_FemPregData.parquet') p USING (CASEID)
    GROUP BY ALL
""").df()

Dictionary lookup:

con.sql(f"""
    SELECT file, variable, label, question_text, universe
    FROM read_parquet('{base}/2002/Dictionary/VARIABLES.parquet')
    WHERE lower(label) LIKE '%contracept%'
""").show()
con.sql(f"""
    SELECT code, description, count, source FROM read_parquet('{base}/2017-2019/Dictionary/VALUES.parquet')
    WHERE file = '2017_2019_FemRespData' AND variable = 'CONSTAT1'
""").show()

R (arrow + survey):

library(arrow); library(survey)
base <- "https://huggingface.co/datasets/jamesjiadazhan/NSFG/resolve/main/data"
f <- read_parquet(file.path(base, "2017-2019/FemResp/2017_2019_FemRespData.parquet"),
                  col_select = c("CASEID", "AGER", "CONSTAT1", "SEST", "SECU", "WGT2017_2019"))
des <- svydesign(ids = ~SECU, strata = ~SEST, weights = ~WGT2017_2019, nest = TRUE, data = f)
svymean(~I(CONSTAT1 %in% 6), des)

Hugging Face datasets (only the subsets listed in Hugging Face subsets and splits; the wide respondent files are read by path as above):

from datasets import load_dataset
dic = load_dataset("jamesjiadazhan/NSFG", "dictionary_variables", split="cycle_2022_2023")

Survey design: required for every NSFG estimate

The NSFG is a multistage area probability sample with oversampling (for example of Black and Hispanic persons and teenagers). Every estimate, including percentages, means, regressions and survival analyses, must use the weight and the design variables of the cycle (metadata/CYCLES.parquet):

Cycle Weight Stratum Cluster (PSU) Notes
1973 PSWT FAM_PSEUDO FAM_PAIRNUM pseudo-strata with paired PSUs
1976 WTFINAL STRAT PSU stratum and PSU are in cyc2psu (join on RESPID)
1982 FINALWT PSEUDOST PSUCODE NONRWT is for methodological work only
1988 W5 – – balanced repeated replication with REPWGT1–REPWGT100; W5 and the replicate weights carry two implied decimals (divide by 100)
1995 POST_WT COL_STR PANEL with-replacement design (Cycle 5 User's Guide, App. 2)
2002 FINALWGT SEST SECU_R (female), SECU_P (pregnancy), SECU (male)
2006–2010 WGTQ1Q16 SEST SECU WGTQ1Q8, WGTQ5Q16, WGTQ9Q16, FINALWGT30 for sub-periods
2011–2013 … 2017–2019 WGT2011_2013, WGT2013_2015, WGT2015_2017, WGT2017_2019 SEST SECU
2022–2023 WGT2022_2023 VEST VECL multimode (in-person and web)

In R: survey::svydesign(ids = ~SECU, strata = ~SEST, weights = ~WGT2017_2019, nest = TRUE, data = ...) (Taylor linearization, as in the NCHS SAS PROC SURVEYFREQ / Stata svy examples). Pregnancy files carry the respondent's weight and design variables; analyses of pregnancies use them as they are.

Pooling cycles (NCHS guidance).

  • 2011–2019: use each 2-year weight to compare periods. To pool releases, use the combined-period weights in 2011_2019_FemaleWgtData / 2011_2019_MaleWgtData (WGT2011_2015, WGT2013_2017, WGT2015_2019, WGT2011_2017, WGT2013_2019, WGT2011_2019), joined on CASEID. Using the 2-year weights for pooled population totals "will result in inflated, inaccurate population size estimates". Respondents aged 45–49 (2015–2019) have no 4-, 6- or 8-year weight. Source: Combining data across NSFG file releases from 2011–2019.
  • Combinations with 2006–2010 or older: Appendix 2 of the 2013–2015 User's Guide.
  • 2022–2023: "NCHS does not advise that users combine data for 2022-2023 with prior NSFG data releases for purposes of making populations estimates for a wider span of years", and urges caution when comparing 2022–2023 with earlier estimates (break in fieldwork from October 2019 to December 2021, multimode design, lower response rates). No combined weights exist. Source: 2022–2023 NSFG FAQ, Q12.
  • 1973–2002 are separate surveys with different universes (1973 and 1976: ever-married women 15–44 and never-married mothers; 1982–1995: women 15–44; 2002–2013: women and men 15–44; 2015 onward: 15–49). Compare them; do not pool them for population totals.

Sources and processing

Part Source Processing
1973–2019 data ftp.cdc.gov/…/Datasets/NSFG fixed-width .dat/.asc Fields cut at the column positions of the NCHS SAS setup program (sas/); blanks → missing; numeric fields parsed as numbers (as SAS column input does), $ fields kept as text; no recoding
2022–2023 data ftp.cdc.gov/…/NCHS/NSFG SAS7BDAT haven::read_sas; labels and formats moved to the dictionary
1976 PSU, 2002 supplements cyc2psu, hhpartypnew, c6_curr_ins SAS7BDAT as above (the ASCII duplicates HHPARTYPNEWASC.DAT, 2002curr_ins.dat are not converted again)
1995 fieldwork C5FieldworkData.asc no setup program exists; layout taken from the fieldwork codebook (C5FieldworkCodebk.pdf)
2002 household variables 2002HHvars.dat column layout from Dataset_Documentation/NSFG/HouseholdVariables.sas
Dictionaries setup programs + codebook PDFs (pdftotext) see above

Setup-program quirks are kept as published: in 1982 F21_1CM/F22_1CM and DIS1MAR/AGEREMAR, and in 1988 F44O/F44P, share columns in the NCHS layout. Labels that state implied decimals (1988 REPWGT*; 1982 age at first marriage "to two decimal places") are not rescaled.

Validation (logs/validation.csv in the pipeline, 120 checks):

  • Rows of all 37 files equal the NCHS user's guides / codebooks, and record length and number of variables equal the 2006–2019 user's-guide tables, except two pregnancy files where the distributed file is smaller than the guide: 2006–2010 (20,492 vs 20,497) and 2011–2013 (9,543 vs 9,547). In both, the rows equal the sum of the respondent recode PREGNUM, so the files are internally consistent.
  • Pregnancy rows equal ΣPREGNUM on the female file for 1982, 1995–2015 and 2022–2023 (1988: 15,759 vs 15,760; 2017–2019: 10,215 vs 10,217).
  • 116,408 published codebook frequencies (1995, 2002, 2022–2023) were compared with the data: all 2002, 1995 fieldwork and 2022–2023 pregnancy counts match; overall 99.97 % match (the 34 differences are 1995 and 2022–2023 male codebook entries whose value lists the text parser merged with a neighbouring variable).
  • Every pregnancy, design, household and weight-file ID is found in its respondent file.
  • Published estimate reproduced exactly: NSFG 2017–2019 Variance Estimation Example 1 (women 15–49 currently using the pill, by age): sum of weights 72,671,926, 18 strata, 72 clusters, and all six percentages (19.5112 … 6.5215) and Taylor-series SEs (1.8670 … 1.0029) to four decimals.

Coverage and gaps. Every file on the NCHS public FTP folders is included. Not included: the 1973 pregnancy file and SPSS/Stata versions of 1973–1995 (NCHS refers users to the ICPSR Integrated Fertility Survey Series); restricted-use files (ACASI details, contextual and geographic data; NCHS Research Data Center only). The 2022–2023 female respondent codebook (PDF, over 10 MB) could not be retrieved automatically, so that file has labels but no question text or published counts; placing 2022-2023-NSFG-FemRespPUFCodebook.pdf in raw/docs/ and rerunning 04_dictionary.R --force adds them.

Licence and citation

  • NSFG data are produced by NCHS, a U.S. Government agency; they are in the public domain. Cite: National Center for Health Statistics. National Survey of Family Growth (cycle), public-use data files and documentation. Hyattsville, MD: Centers for Disease Control and Prevention. https://www.cdc.gov/nchs/nsfg/
  • This compilation (pipeline, dictionaries, layout) is released under CC0-1.0.

Data users must follow the NCHS data use restrictions: use the data for statistical reporting and analysis only, make no attempt to learn the identity of any person, and do not link these data with individually identifiable data from other datasets.

Downloads last month
172