Dataset Viewer
Auto-converted to Parquet Duplicate
year
int32
2k
2k
component
stringclasses
5 values
file
stringclasses
20 values
variable
stringlengths
2
8
code
stringclasses
166 values
description
stringclasses
773 values
count_data
float64
0
103k
βŒ€
value_source
stringclasses
2 values
sas_format
stringclasses
359 values
count_published
float64
0
103k
βŒ€
value_order
int32
1
19
1,997
Core
FAMILYXX
RECTYPE
60
Family
40,623
SAS format
FAP001X
40,623
1
1,997
Core
FAMILYXX
SRVY_YR
1997
1997
40,623
SAS format
FAP002X
40,623
1
1,997
Core
FAMILYXX
FMX
01" - "06
Family number 1 thru 6
0
SAS format
$FAP004X
null
1
1,997
Core
FAMILYXX
FMX
01-06
Family number 1 thru 6
null
codebook
null
40,623
2
1,997
Core
FAMILYXX
INTV_QRT
1
Quarter 1
9,489
SAS format
FAP005X
9,489
1
1,997
Core
FAMILYXX
INTV_QRT
2
Quarter 2
10,329
SAS format
FAP005X
10,329
2
1,997
Core
FAMILYXX
INTV_QRT
3
Quarter 3
10,528
SAS format
FAP005X
10,528
3
1,997
Core
FAMILYXX
INTV_QRT
4
Quarter 4
10,277
SAS format
FAP005X
10,277
4
1,997
Core
FAMILYXX
ASSIGNWK
01
Week 1
2,330
SAS format
FAP006X
2,330
1
1,997
Core
FAMILYXX
ASSIGNWK
02
Week 2
3,235
SAS format
FAP006X
3,235
2
1,997
Core
FAMILYXX
ASSIGNWK
03
Week 3
3,256
SAS format
FAP006X
3,256
3
1,997
Core
FAMILYXX
ASSIGNWK
04
Week 4
3,241
SAS format
FAP006X
3,241
4
1,997
Core
FAMILYXX
ASSIGNWK
05
Week 5
3,257
SAS format
FAP006X
3,257
5
1,997
Core
FAMILYXX
ASSIGNWK
06
Week 6
3,132
SAS format
FAP006X
3,132
6
1,997
Core
FAMILYXX
ASSIGNWK
07
Week 7
3,175
SAS format
FAP006X
3,175
7
1,997
Core
FAMILYXX
ASSIGNWK
08
Week 8
3,122
SAS format
FAP006X
3,122
8
1,997
Core
FAMILYXX
ASSIGNWK
09
Week 9
3,325
SAS format
FAP006X
3,325
9
1,997
Core
FAMILYXX
ASSIGNWK
10
Week 10
3,122
SAS format
FAP006X
3,122
10
1,997
Core
FAMILYXX
ASSIGNWK
11
Week 11
3,196
SAS format
FAP006X
3,196
11
1,997
Core
FAMILYXX
ASSIGNWK
12
Week 12
3,090
SAS format
FAP006X
3,090
12
1,997
Core
FAMILYXX
ASSIGNWK
13
Week 13
3,142
SAS format
FAP006X
3,142
13
1,997
Core
FAMILYXX
LIVQRT
01
House, apartment, flat, condo
37,870
SAS format
FAP007X
37,870
1
1,997
Core
FAMILYXX
LIVQRT
02
HU in nontransient hotel, motel
21
SAS format
FAP007X
21
2
1,997
Core
FAMILYXX
LIVQRT
03
HU-permanent in transient hotel, motel
3
SAS format
FAP007X
3
3
1,997
Core
FAMILYXX
LIVQRT
04
HU in rooming house
18
SAS format
FAP007X
18
4
1,997
Core
FAMILYXX
LIVQRT
05
Mobile home/trailer w/no permanent rooms added
1,931
SAS format
FAP007X
1,931
5
1,997
Core
FAMILYXX
LIVQRT
06
Mobile home/trailer w/ 1+ permanent rooms adde
319
SAS format
FAP007X
319
6
1,997
Core
FAMILYXX
LIVQRT
07
HU not specified above
21
SAS format
FAP007X
21
7
1,997
Core
FAMILYXX
LIVQRT
08
Quarters not HU in room or board house
49
SAS format
FAP007X
49
8
1,997
Core
FAMILYXX
LIVQRT
09
Unit not permanent-transient hotel, motel
1
SAS format
FAP007X
1
9
1,997
Core
FAMILYXX
LIVQRT
10
Unoccupied site for mobile home/trailer/tent
0
SAS format
FAP007X
0
10
1,997
Core
FAMILYXX
LIVQRT
11
Student quarters in college dormitory
217
SAS format
FAP007X
217
11
1,997
Core
FAMILYXX
LIVQRT
12
Group quarter unit not specified above
161
SAS format
FAP007X
161
12
1,997
Core
FAMILYXX
LIVQRT
98
Not ascertained
12
SAS format
FAP007X
12
13
1,997
Core
FAMILYXX
TELENUM
1
Yes (gave phone number)
36,712
SAS format
FAP008X
36,712
1
1,997
Core
FAMILYXX
TELENUM
2
No (no phone service)
1,837
SAS format
FAP008X
1,837
2
1,997
Core
FAMILYXX
TELENUM
7
Refused
1,886
SAS format
FAP008X
1,886
3
1,997
Core
FAMILYXX
TELENUM
8
Not ascertained
1
SAS format
FAP008X
1
4
1,997
Core
FAMILYXX
TELENUM
9
Don't know
187
SAS format
FAP008X
187
5
1,997
Core
FAMILYXX
INSIDE
1
Yes
38,405
SAS format
FAP009X
38,405
1
1,997
Core
FAMILYXX
INSIDE
2
No
340
SAS format
FAP009X
340
2
1,997
Core
FAMILYXX
INSIDE
7
Refused
25
SAS format
FAP009X
25
3
1,997
Core
FAMILYXX
INSIDE
8
Not ascertained
1
SAS format
FAP009X
1
4
1,997
Core
FAMILYXX
INSIDE
9
Don't know
15
SAS format
FAP009X
15
5
1,997
Core
FAMILYXX
INSIDE
.
<Blank: Not in universe>
1,837
SAS format
FAP009X
null
6
1,997
Core
FAMILYXX
INSIDE
<Blank: Not in universe>
null
codebook
null
1,837
7
1,997
Core
FAMILYXX
LNG_INTV
1
English
35,704
SAS format
FAP011X
35,704
1
1,997
Core
FAMILYXX
LNG_INTV
2
Spanish
1,631
SAS format
FAP011X
1,631
2
1,997
Core
FAMILYXX
LNG_INTV
3
English and Spanish
831
SAS format
FAP011X
831
3
1,997
Core
FAMILYXX
LNG_INTV
4
Other
118
SAS format
FAP011X
118
4
1,997
Core
FAMILYXX
LNG_INTV
8
Not ascertained
2,339
SAS format
FAP011X
2,339
5
1,997
Core
FAMILYXX
REGION
1
Northeast
7,813
SAS format
FAP012X
7,813
1
1,997
Core
FAMILYXX
REGION
2
Midwest
9,160
SAS format
FAP012X
9,160
2
1,997
Core
FAMILYXX
REGION
3
South
14,573
SAS format
FAP012X
14,573
3
1,997
Core
FAMILYXX
REGION
4
West
9,077
SAS format
FAP012X
9,077
4
1,997
Core
FAMILYXX
MSASIZEP
1
5,000,000 or more
4,863
SAS format
FAP013X
4,863
1
1,997
Core
FAMILYXX
MSASIZEP
2
2,500,000 - 4,999,999
4,937
SAS format
FAP013X
4,937
2
1,997
Core
FAMILYXX
MSASIZEP
3
1,000,000 - 2,499,999
10,177
SAS format
FAP013X
10,177
3
1,997
Core
FAMILYXX
MSASIZEP
4
500,000 - 999,999
4,657
SAS format
FAP013X
4,657
4
1,997
Core
FAMILYXX
MSASIZEP
5
250,000 - 499,999
4,769
SAS format
FAP013X
4,769
5
1,997
Core
FAMILYXX
MSASIZEP
6
Under 250,000
3,515
SAS format
FAP013X
3,515
6
1,997
Core
FAMILYXX
MSASIZEP
7
Non-MSA
7,705
SAS format
FAP013X
7,705
7
1,997
Core
FAMILYXX
FCHLMYN
1
Yes
153
SAS format
FAP009X
153
1
1,997
Core
FAMILYXX
FCHLMYN
2
No
6,183
SAS format
FAP009X
6,183
2
1,997
Core
FAMILYXX
FCHLMYN
7
Refused
4
SAS format
FAP009X
4
3
1,997
Core
FAMILYXX
FCHLMYN
8
Not ascertained
11
SAS format
FAP009X
11
4
1,997
Core
FAMILYXX
FCHLMYN
9
Don't know
3
SAS format
FAP009X
3
5
1,997
Core
FAMILYXX
FCHLMYN
.
<Blank: Not in universe>
34,269
SAS format
FAP009X
null
6
1,997
Core
FAMILYXX
FCHLMYN
<Blank: Not in universe>
null
codebook
null
34,269
7
1,997
Core
FAMILYXX
FCHLMCT
00 - 30
00-30 children
6,354
SAS format
FAP018X
null
1
1,997
Core
FAMILYXX
FCHLMCT
.
<Blank: Not in universe>
34,269
SAS format
FAP018X
null
2
1,997
Core
FAMILYXX
FCHLMCT
00-30
Number of children
null
codebook
null
6,354
3
1,997
Core
FAMILYXX
FCHLMCT
<Blank: Not in universe>
null
codebook
null
34,269
4
1,997
Core
FAMILYXX
FSPEDYN
1
Yes
1,302
SAS format
FAP009X
1,302
1
1,997
Core
FAMILYXX
FSPEDYN
2
No
14,031
SAS format
FAP009X
14,031
2
1,997
Core
FAMILYXX
FSPEDYN
7
Refused
9
SAS format
FAP009X
9
3
1,997
Core
FAMILYXX
FSPEDYN
8
Not ascertained
33
SAS format
FAP009X
33
4
1,997
Core
FAMILYXX
FSPEDYN
9
Don't know
17
SAS format
FAP009X
17
5
1,997
Core
FAMILYXX
FSPEDYN
.
<Blank: Not in universe>
25,231
SAS format
FAP009X
null
6
1,997
Core
FAMILYXX
FSPEDYN
<Blank: Not in universe>
null
codebook
null
25,231
7
1,997
Core
FAMILYXX
FSPEDCT
00 - 30
00-30 children
15,392
SAS format
FAP018X
null
1
1,997
Core
FAMILYXX
FSPEDCT
.
<Blank: Not in universe>
25,231
SAS format
FAP018X
null
2
1,997
Core
FAMILYXX
FSPEDCT
00-30
Number of children
null
codebook
null
15,392
3
1,997
Core
FAMILYXX
FSPEDCT
<Blank: Not in universe>
null
codebook
null
25,231
4
1,997
Core
FAMILYXX
FLAADLYN
1
Yes
1,372
SAS format
FAP021X
1,372
1
1,997
Core
FAMILYXX
FLAADLYN
2
No
39,227
SAS format
FAP021X
39,227
2
1,997
Core
FAMILYXX
FLAADLYN
7
Refused
15
SAS format
FAP021X
15
3
1,997
Core
FAMILYXX
FLAADLYN
8
Not ascertained
0
SAS format
FAP021X
0
4
1,997
Core
FAMILYXX
FLAADLYN
9
Don't know
9
SAS format
FAP021X
9
5
1,997
Core
FAMILYXX
FLAADLCT
00 - 30
00-30 family members
40,623
SAS format
FAP022X
null
1
1,997
Core
FAMILYXX
FLAADLCT
00-30
Number of family members
null
codebook
null
40,623
2
1,997
Core
FAMILYXX
FLIADLYN
1
Yes
2,779
SAS format
FAP021X
2,779
1
1,997
Core
FAMILYXX
FLIADLYN
2
No
37,814
SAS format
FAP021X
37,814
2
1,997
Core
FAMILYXX
FLIADLYN
7
Refused
15
SAS format
FAP021X
15
3
1,997
Core
FAMILYXX
FLIADLYN
8
Not ascertained
0
SAS format
FAP021X
0
4
1,997
Core
FAMILYXX
FLIADLYN
9
Don't know
15
SAS format
FAP021X
15
5
1,997
Core
FAMILYXX
FLIADLCT
00 - 30
00-30 family members
40,623
SAS format
FAP022X
null
1
1,997
Core
FAMILYXX
FLIADLCT
00-30
Number of family members
null
codebook
null
40,623
2
1,997
Core
FAMILYXX
FWKLIMYN
1
Yes
7,892
SAS format
FAP009X
7,892
1
1,997
Core
FAMILYXX
FWKLIMYN
2
No
32,341
SAS format
FAP009X
32,341
2
End of preview. Expand in Data Studio

NHIS 1997–2025 in Parquet (with dictionaries and linked mortality)

This dataset contains the public-use data files of the U.S. National Health Interview Survey (NHIS) of the National Center for Health Statistics (NCHS), survey years 1997 through 2025, converted to Parquet: one Parquet file per original NCHS data file, under the NCHS file name. It also includes:

  • a variable dictionary for each survey year, built from the NCHS SAS input programs (column positions, labels, value formats) and the NCHS codebooks (question text, universe, published frequencies), with the observed count of every coded value;
  • the NCHS public-use linked mortality files for NHIS 1997–2018 (follow-up through 31 December 2019);
  • the imputed income files, paradata, supplements, injury/poisoning episode files, revised and parent–child pair weights, and the NCHS dummy records for singleton PSUs.

Note on 2016: NCHS revised the 2016 weights in 2017. The 2016 ASCII files used here already carry the revised weights (they equal the separately published *_REVWTS files, which are included as published).

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

Years 29 (1997–2018 original design, 2019–2025 redesigned questionnaire)
Files 457 Parquet data files plus 2 Γ— 29 dictionary files
Rows 28.2 million (0.37 GB of Parquet)
Persons (1997–2018 PERSONSX) 2,061,980
Sample adults (1997–2025) 878,760
Sample children (1997–2025) 313,708
Values Original public-use codes; nothing recoded, imputed or dropped

Layout

data/<year>/<component>/<FILE>.parquet
data/<year>/Dictionary/VARIABLES.parquet     one row per file x variable
data/<year>/Dictionary/VALUES.parquet        one row per file x variable x code
metadata/FILES.parquet                       every file: rows, columns, grain, join keys, design variables, source URL, MD5
metadata/YEARS.parquet                       every year: design period, sample sizes, strata/PSU/weights, pooling guidance
metadata/VALIDATION.parquet                  every validation check: expected, observed, source
pipeline/                                    the R scripts that rebuilt everything from the NCHS sources

<year> is the survey year (1997 … 2025), <component> the folder of the file's category and <FILE> the upper-cased NCHS file name; Data by category below lists every file.

Data by category

NCHS publishes NHIS as separate files for each record level and topic. Until 2018 the interview was hierarchical (household β†’ family β†’ person, with one sample adult and one sample child per family); the 2019 redesign replaced this with one sample adult and one sample child per household. The table follows that organisation. File type is the NCHS file name without year codes; the folder is the <component> in data/<year>/<component>/<FILE>.parquet.

Category Contents Folder File types Files Years
Household file (1997–2018) One record per interviewed household: interview status, housing, household weight. Core 1 22 1997–2018
Family file (1997–2018) One record per family: family structure, income, health insurance and limitations summed over members, family weight. Core 1 22 1997–2018
Person file (1997–2018) Every member of every family: demographics, health status, limitations, insurance, access and use of care, person weight and design variables. Core 1 22 1997–2018
Sample adult file (1997–2018) One randomly selected adult per family: conditions, health behaviours (smoking, alcohol, activity), access, sample adult weight. Core 1 22 1997–2018
Sample child file (1997–2018) One randomly selected child per family: conditions, development, access, sample child weight. Core 1 22 1997–2018
2019+ redesigned sample adult and sample child files Since 2019 one sample adult and one sample child per household; household and family content is folded in. Includes the 2019–2020 longitudinal and 2020 partial-sample weight files. Core, Longitudinal 4 16 2019–2025
Supplements and topical files Periodic supplements (immunization, prevention, complementary medicine, cancer control, disability, quality of life), birth weight and a correction file. Supplement 16 48 1997–2017
Injury and poisoning episode files (1997–2017) One record per injury or poisoning episode, and the verbatim (text) descriptions. Injury 4 38 1997–2017
Imputed income files Multiply imputed family income and personal earnings: five implicate files a year (1997–2018), ten stacked implicates (2019+). ImputedIncome 4 184 1997–2025
Paradata (2006+) Interview process data: contact attempts, mode, interview length, response outcome. Paradata 1 20 2006–2025
Design and weight files Parent–child pair weights (2019+), revised 2016 weights, the 2010 sample adult Γ— disability test weight, and NCHS dummy records for singleton PSUs. Design 4 19 2003, 2005, 2007, 2010, 2016, 2019–2025
Linked mortality (NCHS, 1997–2018 survey years) Public-use linked mortality files: vital status, quarter/year and leading cause of death, follow-up through 31 December 2019. Mortality 1 22 1997–2018

File naming.

  • 1997–2018 file names carry no year (PERSONSX, SAMADULT, INJPOIEP): the year is the folder.
  • 2019+ names end with the two-digit year (ADULT19, CHILDINC24, PARADATA25); pair weights end with the four-digit year (FINAL_PAIR_WEIGHT2025).
  • Imputed income files end with the implicate number 1–5 (INCMIMP3; NCHS named them INCMAP1–5 in 2003–2004).
  • NCHS dummy records for singleton PSUs start with the year (2007SAMCHILD); linked mortality files are NHIS_<year>_MORT_2019_PUBLIC (follow-up through 2019).
  • Names are upper-cased NCHS names; - became _ (PERSON_REVWTS).

Expand a category for every file type: the official NCHS title of its most recent release (from the NCHS codebook, frequency or summary report; none when NCHS publishes no document for that file), a plain description, the years it exists, its file names, and its grain and join keys (both checked on the data).

Household file (1997–2018): 1 file types, 22 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
HOUSEHLD Household Public Use File Household file 1997–2018 HOUSEHLD one row per HHX HHX
Family file (1997–2018): 1 file types, 22 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
FAMILYXX Family Public Use File Family file 1997–2018 FAMILYXX one row per HHX + FMX HHX + FMX
Person file (1997–2018): 1 file types, 22 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
PERSONSX Person Public Use File Person file 1997–2018 PERSONSX one row per HHX + FMX + PX; one row per HHX + FMX + FPX HHX + FMX + PX / HHX + FMX + FPX
Sample adult file (1997–2018): 1 file types, 22 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
SAMADULT Sample Adult Public Use File Sample Adult file 1997–2018 SAMADULT one row per HHX + FMX HHX + FMX + PX / HHX + FMX + FPX
Sample child file (1997–2018): 1 file types, 22 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
SAMCHILD Sample Child Public Use File Sample Child file 1997–2018 SAMCHILD one row per HHX + FMX HHX + FMX + PX / HHX + FMX + FPX
2019+ redesigned sample adult and sample child files: 4 file types, 16 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
ADULT Sample Adult File Sample Adult file (2019+ redesign) 2019–2025 ADULT19 … ADULT25 one row per HHX HHX
CHILD Sample Child File Sample Child file (2019+ redesign) 2019–2025 CHILD19 … CHILD25 one row per HHX HHX
ADULTLONG Adultlong file 2019-2020 longitudinal Sample Adult IDs and weight (WTSA_L) 2020 ADULTLONG20 one row per HHX_2019 HHX_2019 + HHX_2020
ADULTPART Adultpart file 2020 partial-sample Sample Adult IDs and weight for pooled 2019-2020 (WTSA_P) 2020 ADULTPART20 one row per HHX_2020 HHX_2020
Supplements and topical files: 16 file types, 48 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
IMMUNIZE CHILD IMMUNIZATION Immunization file (children) 1997–2003 IMMUNIZE one row per HHX + FMX + PX; one row per HHX + FMX HHX + FMX + PX + IMRESPNO / HHX + FMX + PX
PREGSMKG PREGNANCY AND SMOKING Pregnancy and smoking supplement 1998 PREGSMKG one row per HHX + FMX + PX HHX + FMX + PX
PREVADLT SAMPLE ADULT PREVENTION Sample Adult prevention supplement 1998 PREVADLT one row per HHX + FMX HHX + FMX + PX
PREVCHLD SAMPLE CHILD PREVENTION Sample Child prevention supplement 1998 PREVCHLD one row per HHX + FMX HHX + FMX + PX
ALTHEALT Adult Complementary and Alternative Medicine Public Use File Adult complementary and alternative medicine supplement 2002, 2007, 2012 ALTHEALT one row per HHX + FMX HHX + FMX + PX / HHX + FMX + FPX
ALTVERBT none Adult alternative medicine verbatim file 2002 ALTVERBT one row per HHX + FMX HHX + FMX + PX
CHILDCAM Child Complementary and Alternative Medicine Public Use File Child complementary and alternative medicine supplement 2007, 2012 CHILDCAM one row per HHX + FMX HHX + FMX + FPX
CANCERXX Sample Adult Cancer Public Use File Cancer control supplement 2005, 2010, 2015 CANCERXX one row per HHX + FMX HHX + FMX + FPX
DISBTEST Disability Questions Tests 2010 Public Use File Disability questions tests 2009–2010 DISBTEST one row per SRVY_YR + HHX + FMX + FPX; one row per HHX + FMX + FPX SRVY_YR + HHX + FMX + FPX / HHX + FMX + FPX
QUALLIFE Quality of Life Public Use File Quality of life supplement 2010 QUALLIFE one row per HHX + FMX HHX + FMX + FPX
ADLTDISB Adult Disability Questions Test Public Use File Adult disability supplement 2011–2012 ADLTDISB one row per HHX + FMX HHX + FMX + FPX
CHLDDISB Child Disability Questions Test Public Use File Child disability supplement 2011–2012 CHLDDISB one row per HHX + FMX HHX + FMX + FPX
FMLYDISB Family Disability Questions Public Use File Family disability questions 2011–2017 FMLYDISB one row per HHX + FMX + FPX HHX + FMX + FPX
FUNCDISB Functioning and Disability Public Use File Functioning and disability supplement 2011–2017 FUNCDISB one row per HHX + FMX HHX + FMX + FPX
SC_BWT none Sample Child birth weight file 2004–2010 SC_BWT04 … SC_BWT10 one row per HHX + FMX HHX + FMX + FPX
RATCAT none Corrected family poverty ratio (RAT_CAT3) for 2007 2007 RATCAT07 one row per HHX + FMX HHX + FMX
Injury and poisoning episode files (1997–2017): 4 file types, 38 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
INJEPSOD INJURY EPISODE Injury episode file 1997–1999 INJEPSOD one row per HHX + FMX + PX + INJEPNO HHX + FMX + PX + INJEPNO
POIEPSOD POISON EPISODE Poisoning episode file 1997–1999 POIEPSOD one row per HHX + FMX + PX + POIEPNO HHX + FMX + PX + POIEPNO
INJPOIEP Episode Public Use File Injury/poisoning episode file 2000–2017 INJPOIEP one row per HHX + FMX + PX + IPEPNO; one row per HHX + FMX + FPX + IPEPNO HHX + FMX + PX + IPEPNO / HHX + FMX + FPX + IPEPNO
INJVERBT Episode Verbatim Public Use File Injury/poisoning verbatim file 1997–2010 INJVERBT one row per HHX + FMX + PX + INJEPNO; one row per HHX + FMX + PX + IPEPNO; one row per HHX + FMX + FPX + IPEPNO HHX + FMX + PX + INJEPNO / HHX + FMX + PX + IPEPNO / HHX + FMX + FPX + IPEPNO
Imputed income files: 4 file types, 184 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
INCMIMP IMPUTED FAMILY INCOME / PERSONAL EARNINGS FILE Imputed family income / personal earnings (one implicate) 1997–2018 INCMIMP1 … INCMIMP5 one row per HHX + FMX + PX; one row per HHX + FMX + FPX HHX + FMX + PX + IMPNUM / HHX + FMX + FPX + IMPNUM
INCIMPS Public Use Supplemental Imputed Income Supplemental imputed income, top-coded (one implicate) 1997–2008 INCIMPS1 … INCIMPS5 one row per HHX + FMX + FPX HHX + FMX + FPX + IMPNUM
ADULTINC Sample Adult Imputed Income File Sample Adult imputed income (10 implicates, stacked) 2019–2025 ADULTINC19 … ADULTINC25 one row per HHX + IMPNUM; one row per HHX + IMPNUM_A HHX + IMPNUM / HHX + IMPNUM_A
CHILDINC Sample Child Imputed Income File Sample Child imputed income (10 implicates, stacked) 2019–2025 CHILDINC19 … CHILDINC25 one row per HHX + IMPNUM; one row per HHX + IMPNUM_C HHX + IMPNUM / HHX + IMPNUM_C
Paradata (2006+): 1 file types, 20 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
PARADATA Paradata File Paradata (interview process) file 2006–2025 PARADATA … PARADATA25 one row per HHX + FMX; one row per HHX HHX + FMX / HHX
Design and weight files: 4 file types, 19 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
FINAL_PAIR_WEIGHT PUBLIC-USE PAIR WEIGHTS DATA FROM THE NATIONAL HEALTH INTERVIEW SURVEY Parent-child pair weights 2019–2025 FINAL_PAIR_WEIGHT2019 … FINAL_PAIR_WEIGHT2025 one row per HHX HHX
REVWTS none Revised 2016 weights 2016 FAMILY_REVWTS … SAMCHILD_REVWTS one row per HHX + FMX; one row per HHX + FMX + FPX; one row per HHX HHX + FMX / HHX + FMX + FPX / HHX
SAD_WGTS none Weights for Sample Adult merged with the 2010 disability tests 2010 SAD_WGTS one row per HHX + FMX HHX + FMX + FPX
SINGLETON none Supplemental dummy records for singleton PSUs (layout of SAMCHILD file) 2003, 2005, 2007 2003SAMCHILD … 2007SAMCHILD one row per HHX HHX + FMX + PX / HHX + FMX + FPX
Linked mortality (NCHS, 1997–2018 survey years): 1 file types, 22 files
File type Official NCHS title (latest) Description Years File names Grain Join keys
MORT PUBLIC-USE LINKED MORTALITY FILE Public-use linked mortality file (NDI follow-up through 31 December 2019) 1997–2018 NHIS_1997_MORT_2019_PUBLIC … NHIS_2018_MORT_2019_PUBLIC one row per PUBLICID PUBLICID

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/NHIS", "<subset>"). A split is a part of a subset, chosen with split=. In this dataset a split is either all (the whole table) or one survey year (year_2019 is the 2019 NHIS). The Year column tells which year a row belongs to when splits are stacked.

Subset What it holds Splits Year column
files Catalogue of every file: year, component, file, official title, description, rows, columns, grain, join keys, design variables, path, source URL, MD5. all year
years One row per survey year: questionnaire era, sample design period, file counts, sample sizes, strata / PSU / weight variables, pooling guidance. all year
dictionary_variables Variable dictionary: one row per file Γ— variable (position, type, label, question text, universe). 29, one per survey year (year_1997 … year_2025) year
dictionary_values Value dictionary: one row per file Γ— variable Γ— code (description, observed and published counts). 29, one per survey year (year_1997 … year_2025) year
linked_mortality The NCHS public-use linked mortality file of each survey year (same 10 columns every year). 22, one per survey year (year_1997 … year_2018) first 4 characters of PUBLICID
pair_weights Parent–child pair weights of each year (HHX, final_pair_weight). 7, one per survey year (year_2019 … year_2025) none (one split per year; add it when stacking)

Subsets cover only tables whose columns are the same in every year. The 457 NCHS data files (listed in the files subset and in Data by category) are not subsets: their variables change from year to year (the 2019 redesign changed nearly all of them), and 5 files have more than 1,000 columns (largest: ALTHEALT 2007, 3,027 columns), more than the viewer shows. Read them by path, for example data/2019/Core/ADULT19.parquet or data/2010/Core/SAMADULT.parquet (see Quick start). metadata/VALIDATION.parquet is also read by path. The Dataset Viewer shows only the subsets; the Files and versions tab shows every file.

Data dictionary

Each survey year has its own dictionary covering every file of that year. It is built from the NCHS SAS input programs (positions, labels, value formats), the NCHS codebooks (question text, universe, published frequencies) and the data (counts):

File Subset One row per Main columns
data/<year>/Dictionary/VARIABLES.parquet dictionary_variables file Γ— variable file, variable, label, question_text, universe (the condition that defines who was asked) and universe_description, type, start/end/implied_decimals (ASCII layout), n_nonmissing, sas_format, question_id, notes, codebook (source document), label_source
data/<year>/Dictionary/VALUES.parquet dictionary_values file Γ— variable Γ— code code, description, count_data (count in the Parquet file), count_published (NCHS codebook frequency, 1997–2003 and 2019+), value_source, value_order

Look a variable up by year, file and variable: for example in data/2019/Dictionary/VARIABLES.parquet, file = 'ADULT19', variable = 'SMKCIGST_A' is Cigarette smoking status, and VALUES.parquet lists its codes (1 = every day smoker … 9 = unknown) with their counts. Quick start shows a search by label. The official NCHS document of each file (codebook or layout PDF) is linked in the doc_url column of the files subset and named in the codebook column of the dictionary; the NCHS SAS program used for each file is in the program column of files.

Joining

Within a year, 1997–2018. Records are identified hierarchically:

  • household: HHX;
  • family: HHX + FMX;
  • person: HHX + FMX + FPX (the person number is called PX in 1997–2003 files and in the 1997–2003 imputed income files; rename it to FPX to stack years);
  • episode files add the episode number: INJEPNO / POIEPNO (1997–1999), IPEPNO (2000+);
  • IMMUNIZE adds IMRESPNO; the imputed income files hold one implicate each (IMPNUM);
  • the 2009 DISBTEST file spans the 2008 and 2009 interviews: its key includes SRVY_YR.

So PERSONSX, SAMADULT, SAMCHILD, the person-level supplements and INCMIMPn join on HHX + FMX + FPX (or PX); FAMILYXX joins on HHX + FMX; HOUSEHLD on HHX. HHX is only unique within a year: always join within a year (or include SRVY_YR in the key when stacking).

Within a year, 2019+. There is one sample adult and at most one sample child per household, so every file joins on HHX. ADULTINCyy / CHILDINCyy have ten rows per HHX (one per implicate: IMPNUM in 2019–2020, IMPNUM_A / IMPNUM_C from 2021; from 2021 the ADULTyy / CHILDyy files also carry an IMPNUM_A / IMPNUM_C column, see the NCHS imputation documentation). FINAL_PAIR_WEIGHTyyyy joins on HHX. ADULTPART20 and ADULTLONG20 use HHX_2020 (= HHX of ADULT20) and ADULTLONG20 also carries HHX_2019 (= HHX of ADULT19): this is the only cross-year identity in the public files.

Linked mortality. The mortality files identify people by PUBLICID, a 14-character string that NCHS builds as SRVY_YR (4) + HHX (6) + FMX (2) + FPX (2) of the 1997–2018 person files. Values are kept as published, so join with substr(PUBLICID,1,4) = SRVY_YR, substr(PUBLICID,5,6) = HHX, substr(PUBLICID,11,2) = FMX, substr(PUBLICID,13,2) = FPX (or PX for 1997–2003). Mortality follow-up covers adults 18+ (ELIGSTAT = 1 eligible); use WGT_NEW (person) or SA_WGT_NEW (sample adult), the weights NCHS adjusted for ineligible records. The 1997–2014 files list every person of PERSONSX; the 2015–2018 files list sample adults and sample children only.

metadata/FILES.parquet gives the grain (checked on the data: one row per key, or several) and join_keys of every file.

Quick start

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

import duckdb
con = duckdb.connect()
base = "hf://datasets/jamesjiadazhan/NHIS/data"
con.sql(f"""
    SELECT a.HHX, a.AGEP_A, a.SEX_A, a.SMKCIGST_A, a.WTFA_A, a.PSTRAT, a.PPSU, i.FAMINCTC_A
    FROM read_parquet('{base}/2019/Core/ADULT19.parquet') a
    LEFT JOIN (SELECT * FROM read_parquet('{base}/2019/ImputedIncome/ADULTINC19.parquet')
               WHERE IMPNUM = 1) i USING (HHX)
""").df()

1997–2018 person + sample adult + mortality:

con.sql(f"""
    SELECT s.HHX, s.FMX, s.FPX, s.AGE_P, s.SEX, s.WTFA_SA, s.STRAT_P, s.PSU_P,
           m.ELIGSTAT, m.MORTSTAT, m.UCOD_LEADING, m.DODYEAR, m.SA_WGT_NEW
    FROM read_parquet('{base}/2010/Core/SAMADULT.parquet') s
    LEFT JOIN read_parquet('{base}/2010/Mortality/NHIS_2010_MORT_2019_PUBLIC.parquet') m
      ON m.PUBLICID = CAST(s.SRVY_YR AS VARCHAR) || s.HHX || s.FMX || s.FPX
""").df()

Dictionary lookup:

con.sql(f"""
    SELECT file, variable, label, question_text, universe
    FROM read_parquet('{base}/2019/Dictionary/VARIABLES.parquet')
    WHERE lower(label) LIKE '%cigarette%'
""").show()
con.sql(f"""
    SELECT code, description, count_data, count_published
    FROM read_parquet('{base}/2019/Dictionary/VALUES.parquet')
    WHERE file = 'ADULT19' AND variable = 'SMKCIGST_A'
""").show()

R (arrow):

library(arrow)
base <- "https://huggingface.co/datasets/jamesjiadazhan/NHIS/resolve/main/data"
adult <- read_parquet(file.path(base, "2019/Core/ADULT19.parquet"))

Hugging Face datasets (subsets and splits are explained in Hugging Face subsets and splits):

from datasets import load_dataset
mort = load_dataset("jamesjiadazhan/NHIS", "linked_mortality", split="year_2010")

Survey design: required for every NHIS estimate

NHIS is a multistage area probability sample. Every estimate must use the weights and the public-use design variables (Taylor linearization, PSUs treated as sampled with replacement, as NCHS documents). metadata/YEARS.parquet lists them for every year:

Years Strata PSU Weights
1997–2005 STRATUM PSU WTFA_HH (household), WTFA_FAM (family), WTFA (person), WTFA_SA (sample adult), WTFA_SC (sample child), supplement-specific weights
2006–2015 STRAT_P PSU_P as above
2016–2018 PSTRAT PPSU as above
2019–2025 PSTRAT PPSU WTFA_A (sample adult), WTFA_C (sample child); 2019 also WTIA_A/WTIA_C; 2020 WTSA_P, WTSA_L; final_pair_weight (parent–child pairs)
  • In R: survey::svydesign(ids = ~PPSU, strata = ~PSTRAT, weights = ~WTFA_A, nest = TRUE, data = adult).
  • Use the weight of the file the variables come from (person, sample adult, sample child, supplement); the imputed income files carry no weights, join them to the file whose weight you use and combine the implicates with Rubin's rules.
  • Pooling years (NCHS variance estimation guidance, see sources): divide each weight by the number of years pooled.
    • Years within one sample design period (1997–2005, 2006–2015, 2016–2018) are pooled as one sample with that period's design variables; do not treat the years as independent.
    • Across design periods (2005/2006, 2015/2016), treat the periods as independent and create a stratum variable with distinct values per period (for example, add 1000 to the 2016+ PSTRAT), because the codes are not related across periods.
    • 2019 and later: NCHS states that public-use files from 2019 on, including 2025 (first year of the 2025 sample redesign), can be pooled with PSTRAT/PPSU as they are. For pooled 2019–2020 sample adult estimates use WTSA_P (ADULTPART20) for the 2020 records.
    • The 2019 questionnaire, file structure and weighting changed: pooling 1997–2018 with 2019+ data needs harmonised variables and care; NCHS gives no public-use recipe for it.
  • The 2025 files are smaller: NCHS reports budget-driven sample reductions in the first year of the 2025 design and a data-collection pause from 1 October to 17 November 2025.
  • Singleton PSUs: 2003, 2005 and 2007 sample child (and 2007 child CAM) files contain PSUs with a single respondent. NCHS published dummy records (Design/2003SAMCHILD etc., weight 1) that can be appended for software that cannot handle them; or use options(survey.lonely.psu = "adjust").

Sources and processing

Part Source Processing
1997–2018 data NCHS FTP Datasets/NHIS/<year> fixed-width ASCII Read with the column positions of the NCHS SAS input program (Program_Code/NHIS/<year>); $ variables as text, others as numbers
2019–2025 data NCHS CSV files (*csv.zip) Variable order and types from the NCHS SAS program; the CSV is used because the ASCII copy rounds weights to its field width (e.g. WTFA_A 13177.008 vs 13177.01)
Imputed income 1997–2004 Datasets/NHIS/<year>_imputed_income No SAS program on the FTP: layout of the 2005 INCMIMP.sas, which is identical to the NCHS dataset documentation datdocYY.pdf (person number PX through 2003)
Pair weights NCHS SAS data sets (final_pair_weightYYYY.sas7bdat) haven::read_sas
2016 revised weights NCHS CSV (2016/RWCSV) IDs as text, weights as numbers
Linked mortality NCHS public-use linked mortality files NHIS layout of the NCHS SAS_ReadInProgramAllSurveys.sas
Dictionaries NCHS SAS programs + codebooks See below

Reading rules (exactly what the NCHS SAS programs do): blank numeric fields and . are missing; implied decimals of the INPUT statement (WTFA 22-29 .3) are applied only when the field has no decimal point; blank character fields are null; leading and trailing blanks of character fields are removed. Nothing is recoded, top-coded, imputed or dropped. The SAS programs are parsed as text; nothing downloaded is executed.

Dictionaries (columns and paths: see Data dictionary above):

  • labels, positions and value labels come from the NCHS SAS programs (label_source);
  • question text, universe and notes come from the NCHS codebooks: the 1997–2003 codebooks (<file>.pdf), the 2004–2018 layout documents (<file>_layout.pdf) and the 2019+ codebooks (<File>-codebook.pdf), parsed from their text (codebook names the document);
  • count_published is the unweighted frequency printed in the 1997–2003 and 2019+ codebooks; count_data is the count in the Parquet file (ranges such as 01-25 are summed; open ranges such as LOW-HIGH have none). value_source = "codebook" marks categories that appear only in the codebook (for example <Blank: Not in universe>).

Dictionary coverage. 64,117 file Γ— variable rows: 64,095 have a label, 53,473 are matched to a codebook entry (45,543 with question text, 52,491 with a universe). 361,719 value rows, of which 123,643 carry a published frequency (1997–2003 and 2019–2025). Files without a codebook (labels and value formats from the SAS program only): imputed income files, linked mortality, revised weights, pair weights (labels of the SAS data set), birth weight SC_BWT, RATCAT07, SAD_WGTS and the 2006 paradata. 2004–2018 layout documents carry no frequencies, so count_published is empty for those years (count_data is always filled).

Validation (pipeline/05_validate.R; every check with its source is in metadata/VALIDATION.parquet):

  • 825 checks, all passed.
  • Row counts: 223 files equal the record count printed by NCHS (1997–2003 codebook headers, RECTYPE frequencies of the 2004–2018 frequency reports and 2019+ codebooks); the 2019–2024 Sample Adult / Sample Child files equal the interviewed counts of Table 2 of the 2024 Survey Description (2020: ADULTPART20 = 21,153; ADULT20 also holds the 10,415 re-interviewed 2019 adults).
  • Published frequencies: all 123,643 published unweighted frequencies equal the counts in the Parquet files (87 files).
  • Layout: every fixed-width record fits the SAS layout and no numeric field failed to parse (406 files).
  • Second NCHS copy: 71 files of 2015–2025 agree cell by cell with the other format NCHS publishes (1997–2018 Parquet vs the CSV copy; 2019+ Parquet vs the ASCII copy, within the ASCII rounding). Two NCHS differences are explained, not errors: the 2015–2018 CSV copies store BMI without its implied decimals (2358 for 23.58), and the 2016 Sample Child CSV copy has the original weights while the ASCII file (used here) has the revised 2016 weights (all 2016 ASCII core files equal the *_REVWTS files). In 2021 PARADATA21, 11 character flags are . in the ASCII copy and blank in the CSV.
  • Linked mortality: every PUBLICID of the 22 files matches one person of the same year's PERSONSX. 1997–2014 files cover all persons; 2015–2018 files cover sample adults and sample children only, as NCHS released them.
  • Published estimates, reproduced with survey (Taylor series, PSU nested in strata, lonely PSU "adjust"): current cigarette smoking among adults, 2005 20.90% (published 20.9%, MMWR 2006;55:1145), 2010 19.35% (19.3%, MMWR 2011;60:1207), 2018 13.73% (13.7%, MMWR 2019;68:1013), 2019 13.98% (14.0%, MMWR 2020;69:1736).

Coverage. Every data file in the NCHS FTP folders Datasets/NHIS/1997 … 2025 and the <year>_imputed_income and 2006_paradata folders, and the NHIS linked mortality files for 1997–2018. Not included: survey years 1963–1996 (different file structures), CSV copies of the same files, the restricted-use files of the NCHS Research Data Center, and IPUMS NHIS (its terms do not allow redistribution).

Licence and citation

  • NHIS and linked mortality: NCHS public-use data, U.S. Government works in the public domain ("…is in the public domain, and may be freely used or reproduced without obtaining copyright permission", CDC). Cite: National Center for Health Statistics. National Health Interview Survey, . Public-use data file and documentation. Hyattsville, MD: NCHS, Centers for Disease Control and Prevention.
  • This compilation (pipeline, dictionaries, layout) is released under CC0-1.0.

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

Downloads last month
136