NATIONAL CENTER FOR HEALTH STATISTICS 10 Series 2, Number 207
weights, as they incorporate the sampling probabilities of both the Sample Adult
and the Sample Child and are adjusted for pair-level nonresponse. Using Sample
Adult weights or Sample Child weights for pair-level statistical analysis may lead
to biased results. Appendix II compares mean estimates using the pair weights
and the Sample Adult weights under a simplified scenario. Factors found to be
related to the differences in the mean estimates using the two sampling weights
included the distribution of the outcome of interest, the number of children across
households, and the sampling weights of Sample Adults.
Three logistic regression models were
applied to the 2019 NHIS dyadic data,
which use the dyadic-level or the
individual-level measurement as the
response variables, respectively. Other
statistical models, such as structural
equation modeling (22) and multilevel
modeling (23,24), may also be applied
to the NHIS parent–child data. In
practice, different estimation methods
can be used for different research
goals; and more research is needed
to explore how to use the pair data
from NHIS (2019 and forward). Design-
based variance estimation was used
for the repeated measurement model
in this report, which incorporates the
survey design features (strata, PSU,
and sampling weights) for variance
estimation and is expected to yield
conservative variance estimates.
However, it does not reflect the nested
data structure of parent–child pairs
within a household. To control for the
additional parent–child correlation,
alternative statistical methods can be
used, for example, random or mixed-
effect models, which may incorporate
the correlation of the Sample Adult and
the Sample Child within a household. In
addition, resampling methods such as
Jackknife and Bootstrap methods may
also be used for variance estimation of
the dyadic data.
Although traditional household surveys
usually focus on the household-
level and the individual-level
measurements, dyadic data in national
household surveys are not uncommon.
The National Survey of Drug Use and
Health, conducted by the Substance
Abuse and Mental Health Services
Administration, collects detailed
information on tobacco, alcohol, and
drug use, as well as mental health-
related issues in the United States (25).
Zero, one, or two people are selected
within a household, and the sampling
weights for the selected pairs have
been developed. NHIS selects a Sample
Adult and a Sample Child (when
applicable) independently within a
family or a household. Because the
sampling weights for the selected pairs
are the inverse of the pairs’ selection
probabilities, the adult–child pair
Table E. Odds ratio and 95% confidence interval estimates of the logistic
regression model in Example 4 predicting the child’s health status given
selected characteristics with results for mother–child pairs: National
Health Interview Survey, 2019
Characteristic,
(variable name), and category
Mother–Child pairs
Odds ratio 95% confidence interval
Child’s age (AGEP_C)
� � � � � � � � � � � � � � � � � � � � � � 0�97 0�92 1�01
Census region (REGION)
Northeast
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 1�29 0�57 2�89
Midwest � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 0�69 0�30 1�59
South � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 0�58 0�32 1�04
West � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � Ref … …
2013 NCHS
Urban–Rural Classification (URBRRL)
Large central metropolitan
1
� � � � � � � � � � � � � � � � � 0�89 0�44 1�81
Large fringe metropolitan
2
� � � � � � � � � � � � � � � � � � 1�18 0�56 2�49
Medium or small metropolitan
3
� � � � � � � � � � � � � � 1�02 0�52 1�99
Nonmetropolitan
4
� � � � � � � � � � � � � � � � � � � � � � � � � Ref … …
Child’s sex (SEX_C)
Male
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 0�46 0�28 0�75
Female � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � Ref … …
Child’s race and ethnicity (HISPALLP_R_C )
Black, non-Hispanic
� � � � � � � � � � � � � � � � � � � � � � � Ref … …
White, non-Hispanic
� � � � � � � � � � � � � � � � � � � � � � � 1�19 0�54 2�64
Other, non-Hispanic
5
� � � � � � � � � � � � � � � � � � � � � � 1�76 0�65 4�79
Hispanic
6
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 0�63 0�29 1�38
Mother’s education (EDUC_R_A)
High school or less
� � � � � � � � � � � � � � � � � � � � � � � 0�47 0�25 0�87
Some college (including
associate’s degree) � � � � � � � � � � � � � � � � � � � � � � � 0�62 0�33 1�16
Bachelor’s degree and above � � � � � � � � � � � � � � � � Ref … …
Mother’s health status (HEALTH_A)
Excellent, very good, or good
� � � � � � � � � � � � � � � 4�07 2�31 7�17
Fair or poor � � � � � � � � � � � � � � � � � � � � � � � � � � � � � Ref … …
… Category not applicable.
1
Counties in metropolitan statistical areas (MSAs) of 1 million or more population that contain the
entire population of the largest principal city of the MSA, have their entire population contained in the
largest principal city of the MSA, or contain at least 250,000 inhabitants of any principal city of the
MSA.
2
Counties in MSAs of 1 million or more population that did not qualify as large central metropolitan
counties.
3
Counties in MSAs of populations of 250,000 to 999,999 and counties in MSAs of populations less
than 250,000.
4
Counties in micropolitan statistical areas and nonmetropolitan counties that did not qualify as
micropolitan.
5
Includes other non-Hispanic people not shown separately due to smaller groups not being statistically
reliable.
6
People of Hispanic origin may be of any race.
NOTE: Ref is the reference group.
SOURCE: National Center for Health Statistics, 2019 National Health Interview Survey.