User Note - Sampling Weights
Use of Sampling Weights with IPUMS NHIS
The National Health Interview Survey (NHIS) is a complex, multistage probability sample that incorporates stratification, clustering, and oversampling of some subpopulations (e.g., Black, Hispanic, and Asian) in some years. Because of the complex sampling design of the NHIS, users of IPUMS NHIS data must make use of sampling weights to produce representative estimates. While appropriate use of sampling weights will produce correct point estimates, statistical techniques that account for the complex sample design are also necessary to produce correct standard errors and statistical tests, so analysts are advised to review the user note on VARIANCE ESTIMATION as well.
Sampling weights are constructed so that each person can be inflated or expanded to represent the total population of the United States. The approach for creating NHIS sampling weights changed significantly in 2019; large declines in response rates, from approximately 90% to 60% or less, required the introduction of more sophisticated adjustment techniques to better correct for nonresponse. Below we summarize the process employed by NCHS staff to create the NHIS sampling weights.
First, each sampled household was assigned a "base weight," reflecting their probability of selection. The sum of the base weights for households approximates the total size of the number of households in the United States. Next, sampling weights for households, sample adults, and sample children were adjusted for nonresponse using the following procedures:
STEP 1. NCHS staff fit multilevel models predicting response for households using a set of contextual variables drawn from paradata involving information about contact history and neighborhood characteristics collected by interviewers; demographic data from the tract-level Census Planning Database; and medical population data from the county-level Area Health Resources File. Based on the multilevel models, predicted response propensities were generated and divided into quintiles. The base weights were then multiplied by a ratio adjustment factor equal to 1/(median response propensity for the quintile), where the nonresponse adjustment was capped at 2.5 to prohibit the variances from becoming too large.
STEP 2. The adjusted household base weights were used as the starting point to generate the sample adult and sample child weights. For each sample adult or sample child, the nonresponse-adjusted household base weight was multiplied by the inverse of the individual's probability of selection within the household. For example, if a household contained three adults and one child, the adjusted household base weight would be multiplied by 3 for the sample adult weight and by 1 for the sample child weight.
STEP 3. Multilevel models predicting response using contextual variables were estimated for sample adults and sample children, and the process of dividing the predicted response propensities and calculation of the corresponding ratio adjustment factors was repeated to generate the interim sample adult and sample child weights. The set of covariates included in the multilevel models used to predict response can differ across the different units (household, sample adult, and sample child) and from year to year.
Last, the adjusted sample adult and sample child weights were raked, that is, proportionally inflated through an iterative process within categories of demographic variables one at a time until marginals approximately matched population control totals from U.S. Census Bureau population projections and American Community Survey one-year estimates for age, sex, race and ethnicity, educational attainment, Census division, and Metropolitian Statistical Area. The demographic variables included in each year's raking process differ for sample adults and sample children, and will be re-considered and potentially updated each year. While household weights (provided on the paradata file) were adjusted for nonresponse, they were not raked to population controls and should therefore only be used for certain methods of variance estimation that require interim weights. They should not be used to produce population estimates of the total number of US households. For more detailed information about the procedures employed by NCHS to adjust NHIS sampling weights for nonresponse, please refer to their 2020 nonresponse adjustment report.
Before 2019, sampling weights were constructed so that each unit (survey respondent, family, or household) could be inflated or expanded to represent other individuals, families, or households in the United States. There were four general components to the NHIS sampling weights. First, the sampling weight represents the inverse probability of unit selection into the sample. The probability of selection is the cross-product of probabilities at each stage of sampling. (See the user note on SAMPLE DESIGN for further details.) Second, the probability of selection is then adjusted for household non-response. These first two steps determine the household weight. For person-level weights, the third component is referred to as a "first stage ratio adjustment." This is used to correct potential bias due to sample under-coverage, by applying a ratio adjustment to each weight based on a race/MSA-residence classification. The fourth component is a post-stratification adjustment for age, race, and sex using quarterly Census Bureau population control totals.
|First stage ratio adjustment|
|1969-1974||6 color-residence classes|
|1975-1984||12 color-residence classes within region|
|1985-1994||16 race-residence-region classes|
|1995-2005||24 race-ethnicity-residence-region classes|
|2006-2015||32 race-ethnicity-residence-region classes|
|2016-2018||No first-stage ratio adjustment|
|Second stage (post-stratification) adjustment|
|1969-1994||60 age/sex/race categories|
|1995-2005||88 age/race/ethnicity/sex categories|
|2006-2018||100 age/race/ethnicity/sex categories|
Sample Person Weight
SAMPWEIGHT is an IPUMS-constructed variable that harmonizes the Final Annual Sample Adult and Sample Child Weights in the original NHIS public use files for 1997 forward. SAMPWEIGHT also contains the sampling weight for a subset of the pre-1997 survey supplements that followed a sampling scheme in which sample persons (one randomly selected person per household, often restricted to either persons 18+ or persons < 18), rather than all persons, were selected for certain survey supplements. SAMPWEIGHT represents the inverse probability of selection into a sample adult/child supplement, adjusted for non-response with additional post-stratification (1997-2018) or raking (2019-forward) adjustments using the Census Bureau's population control totals.
PERWEIGHT is an IPUMS-constructed variable that harmonizes the Final Annual Weight in the original NHIS public use files for 2018 and earlier samples. This weight should be used for analyses at the person level, for variables in which information was collected on all persons. PERWEIGHT represents the inverse probability of selection into the sample, adjusted for non-response with post-stratification adjustments for age, race/ethnicity, and sex using the Census Bureau's population control totals. For each year, the sum of these weights is equal to that year’s civilian, non-institutionalized U.S. population.
FWEIGHT is an IPUMS-constructed variable that harmonizes the Final Annual Family Weight in the original NHIS public use files for 2018 and earlier samples. Because no Census control totals for the number of civilian, non-institutionalized families exist, this weight is equal to the final person weight of the family member with the smallest post-stratification adjustment. For analyses using the family as the unit of analysis (e.g., how many families could not afford to eat balanced meals in the past 30 days?), researchers should use the family weight, FWEIGHT.
HHWEIGHT is an IPUMS-constructed variable that harmonizes the Final Annual Household Weight in the original NHIS public use files for 2018 and earlier samples. For analyses using the household as the unit of analysis (e.g., how many households contained a person who needed help with activities of daily living?), researchers should use the household weight, HHWEIGHT. Beginning in 1997, vacant housing units and households that could not be interviewed due to resident absence or refusal to participate have a value of zero for HHWEIGHT.
Longitudinal Sample Weight
LONGWEIGHT applies to the persons included in the 2020 longitudinal sample. The longitudinal sample includes those sample adults who previously responded to the 2019 NHIS and were re-contacted to complete the 2020 NHIS. According to the 2020 Survey Description, LONGWEIGHT should be used to evaluate individual-level changes among the same adults before and during the COVID-19 pandemic. Please see the user note on COVID-related changes to the NHIS for more information on the longitudinal sample.
Partial Sample WeightPARTWEIGHT applies to the persons included in the 2020 partial sample, the group of sample adults and sample children included in the original 2020 sample. The 2020 Survey Description advises that PARTWEIGHT should be used with the 2020 data when pooling 2019 and 2020 data to increase sample size; SAMPWEIGHT should be used for the 2019 data. Otherwise, SAMPWEIGHT should be used to produce official estimates for 2020 and to compare estimates between 2019 and 2020. The partial sample does not include sample adults from the longitudinal sample.
MORTWT is an IPUMS-recode variable that represents the NCHS-created sample weights that "account for ineligible status due to insufficient identifying information for linkage" in the original public use NHIS Linked Mortality files. MORTWT should be used when analyzing mortality variables in conjunction with variables originally included in the NHIS person files. To analyze mortality variables in conjunction with variables originally included in the NHIS sample adult files, researchers should instead use MORTWTSA. Linked public use mortality variables are available only for NHIS respondents who were at least 18 years old at the time of the survey.
The SUPPXWT series (i.e., SUPP1WT, SUPP2WT, SUPP3WT) are IPUMS-constructed variables that harmonize the Final Annual Weight in selected supplements of the original NHIS public use files. For analyses using variables that are located in different supplements across the years, researchers should review the variable description for the appropriate sampling weights for each year. In some cases, researchers will need to create a new sampling weight by combining different weights from different years.
The CONDWTX series (i.e., CONDWT1, CONDWT2, CONDWT3, CONDWT4, CONDWT5, and CONDWT6) and PARALWT and DIABWT are IPUMS-constructed variables that harmonize chronic condition prevalence factors for person-level variables constructed from the 1978 to 1996 condition records. To analyze variables constructed from the many-to-one condition records, researchers should review the variable description to determine the appropriate condition weight to use with analyses.
Adjusting Sampling Weights When Pooling Multiple Years of Data
The sampling weights in the IPUMS NHIS represent annual inflation factors. In other words, for each individual, the person weight reflects the number of people that individual survey respondent represents in the total U.S. non-institutionalized population for a given year. Thus, if the analyst chooses to use multiple years of data, the sampling weight needs to be adjusted. For example, imagine that an analyst wants to use data from 1990-1999, pooling 10 years of data. The sampling weights need to be adjusted so that the total sample will represent the U.S. population (on average) for the 10-year period. The simplest adjustment method is to simply divide weight by the number of years of data pooled (i.e., divide PERWEIGHT by 10 in this example). Other, more sophisticated methods of adjustment are available, if the analyst is so inclined. However, it is not clear that these methods perform substantially better.
TAKE NOTE: Special Considerations when Pooling Data
1. Change to Sampling Weight Methodology Implemented in 2019. The process of generating sampling weights changed sharply from the approach employed in 2018 and earlier years. Because of this marked change, which was accompanied by a major redesign of the NHIS questionnaire and data collection approach, it is not possible to know whether any changes detected between 2019 and earlier years are due to changes in the sampling weights, the questionnaire or data collection redesign, or reflect actual change in the phenomena under study. Results of a test conducted in 2018-19 by NCHS indicate that differences in prevalence estimates between pre-2019 and 2019 forward years of data are likely influenced by the 2019 redesign. Based on the results of the Bridge Test, IPUMS NHIS recommends that users do not compare the trends in the pre-2019 with the trends in the 2019-forward data. NCHS has signaled that they plan to release additional evaluation results as more 2019-forward data become available. We will update our guidance based on the findings of any such evaluations.
2. Extra adjustments needed when pooling 2019 and 2020 samples. To improve adjustment of the 2020 sampling weights for nonresponse, NCHS re-contacted selected 2019 NHIS sample adults to complete the 2020 NHIS interview between August and December of 2020. This longitudinal sample, also known as the 2020 followback sample, is comprised of 10,415 sample adults and can be analyzed as a one-time longitudinal panel with observations, spaced one year apart, that take place before and during the COVID-19 pandemic. Because both the 2019 and the 2020 samples contain these 10,415 sample adults, however, special measures must be taken when combining the 2019 and 2020 samples for pooled analyses. NCHS advises adjusting the sample and the sampling weight when combining the 2019 and 2020 samples. First, drop any records with zero values on the partial sample weight for 2020 (PARTWEIGHT, so that only the 2019 observations of longitudinal sample members will be represented in the pooled sample. Second, use PARTWEIGHT rather than SAMPWEIGHT for the 2020 sample. Make any other adjustments for analyses of pooled data as described above. For more information about COVID-19 impacts on NHIS data collection, please see our user note.
Combining Sampling Weights When a Variable is Located in Different Files across Years
In some cases, a variable of interest may be located in different original NHIS files with different sampling schemes across the years. For example, the IPUMS NHIS variable PAPEVER indicates whether a women ever had a Pap test. For the years, 1982, 1992 and 2002, the variable comes from three different files: 1982 Preventive Care supplement, 1992 Cancer Control supplement, and 2002 Sample Adult section. Accordingly, the sampling weights for each individual variable are PERWEIGHT, SUPP2WT, and SAMPWEIGHT, respectively. For analysis, these weights will need to be combined in a new variable. Researchers should generate a new weight, perhaps called PAPWEIGHT, such that PAPWEIGHT = PERWEIGHT if year = 1982; PAPWEIGHT = SUPP2WT if year = 1992; and PAPWEIGHT = SAMPWEIGHT if year = 2002.
For additional information on the construction of weights within each of the NHIS redesigns, users can access original NCHS documentation through links provided below.
National Center for Health Statistics. (1975). Health Interview Survey Procedure 1957-1974. Vital Health Stat, 1(11).
National Center for Health Statistics. (1985). The National Health Interview Survey Design, 1973-84, and Procedures, 1975-83. Vital Health Stat, 1(18).
National Center for Health Statistics. (1989). Design and Estimation for the National Health Interview Survey, 1985-94. Vital Health Stat, 2(110).
National Center for Health Statistics. (2000). Design and Estimation for the National Health Interview Survey, 1995-2004. Vital Health Stat, 2(130).
National Center for Health Statistics. (2010). National Health Interview Survey (1986-2004) Linked Mortality Files. Analytic Guidelines
Updated Mortality, 1986-2009
National Center for Health Statistics. Office of Analysis and Epidemiology. Analytic Guidelines for NCHS 2011 Linked Mortality Files, August, 2013. Hyattsville, Maryland.
National Center for Health Statistics. (2014). Design and Estimation for the National Health Interview Survey, 2006-2015. Vital Health Stat, 2(165).
National Center for Health Statistics. (2017). Survey Description, National Health Interview Survey, 2016. Hyattsville, MD.
Bramlett MD, Dahlhammer JM, Bose J, and Blumberg SJ. New procedures for nonresponse adjustments to the 2019 National Health Interview Survey sampling weights. Published September, 2020.