Guidelines for using Racial and Ethnic Groups in Data Analyses
Updated: July 2003
Purpose
Background and Context
Definition of Minimum Categories
Guidelines
Implementation Schedule
Data Collection
Data Tabulation
Data Presentation
Additional Information
Special Issues
Estimates of "More than One Race" in Washington State
Bridging Ratios for Trend Analysis
Bridging Methods
Sample Use of Statistical Methods
Recommended Method
Additional Information
Converting Multiple to Single Race: Washington State Department of Health Discussion Paper
Guidelines
For Using Racial and Ethnic Groups in Data Analyses (Word Document)
<<<
Previous <<< >>>
Next >>>
Data from the 2000 Census, which used the 1997
standard for collecting race/ethnicity data, are now
available for Washington State. The data show that
3.6% of Washingtonians report having two or more
races. The percent reporting more than one race
depends on the race group. Over 40% of NHOPIs and
AIANs report more than one race; approximately 20%
of Blacks, Asians and people of Other races report
more than one race; less than 5% of Whites report
more than one race. The most common 'multiracial'
combinations are AIAN/White, Asian/White,
Other/White, and Black/White. Of those reporting
more than one race, 92% reported two races, 7%
reported three races, and 1% reported four or more
races.
Data collected under the 1997 standard will not
generally be comparable to data collected under the
1977 standard. In addition, not all programs will
begin using the new standard at the same time. Thus,
when tabulating data using the new standard, some
researchers may also need to tabulate data that show
what responses would have been under the old
standard. The comparison between the old and the new
standard provides a bridging ratio. This
ratio can be used in two ways: 1) to compare current
to past data for trend analyses; and 2) to provide
consistent numerators and denominators before all
data are available in the new format.
Note that agencies may not need a bridging
estimate if they can tolerate a 'break' in their
data series or if comparison to another series
provides users with enough information about the
change. However, if no bridging estimates are
provided agencies should be sure to document the
change in the data series. Depending on the nature
of the analysis, appropriate means for conveying the
change might include producing separate tables or
graphs, clearly demarcated breaks in trend lines,
footnotes, technical notes, and explanations within
the text.
- Bridging estimates only need to be made for
persons who select more than one race. It is
assumed that those who select a single race
would have selected the same race under both
standards.
- Recommended method: Ask a follow-up question
to those reporting more than one racial
identity, asking them to select the one they
feel best describes them. Use these data to
calculate a bridging ratio. This method
was used by the National Center for Health
Statistics based on questions from the
National Health Interview Survey (NHIS). For
information on the NHIS go to http://www.cdc.gov/NCHS/nhis.htm.
- If it is not possible to calculate a
bridging ratio that is directly applicable to
a specific data collection system, then use a bridging
ratio available from another source that
most closely approximates the population under
consideration (e.g., state or county survey).
- If neither of these methods is possible,
there are statistical methods for assigning an
individual to a category:
- whole assignment: assign an individual’s
responses to one race group using either
- the race with the smallest single-race
count (smallest group);
- the race other than white with the highest
single-race count (largest group other than
white);
- the race with the largest single-race count
(largest group); or
- the race with which other
individuals having the same
combination of races identify most
strongly, using data collected in
the NHIS (plurality).
- fractional assignment: assign an individual’s responses
as fractions of multiple categories using either
- equal fractions for each group (equal fractions); or
- the fractional proportion of individuals
identifying most strongly with each race, based on NHIS data (NHIS fractions).
- all inclusive:
assign an individual’s responses to each race reported. In this case,
the sum of the categories totals more than 100% and may need to be
adjusted for comparison with percent distributions under the old
standard.
- Bridging methods using NHIS data will
currently only work if just two races are
given. Data were not retained in the NHIS for
more than two races. Some bridging methods use
NHIS data for two races and equal allocation
for three or more races.
- · Another possible bridging method is hot
deck imputation: assign the race of the
'nearest neighbor' who has similar demographic
characteristics and has reported a single race
that is the same as one of the races reported
by the multiracial individual.
The following scenario shows how a single race
would be assigned using each of the bridging methods. Assume a person
reports being both Black and White. For a Black/White combination in the
NHIS, 70% of those expressing a clear preference identified most closely
with Black and 30% with White.
NCHS, under a collaborative agreement with the US
Census Bureau, has developed bridged single-race
population counts that will be used to calculate birth
and death rates for 2000 and later. The method used to
develop these data is a variant of the Fractional
Assignment Method using NHIS fractions. The Washington
State Department of Health has adopted this method as
its standard.
The NCHS method uses a regression model on four years
of NHIS data (1997 - 2000) to estimate the probability
of selecting each primary race possible for a multiple
race group. For example, two probabilities were
generated for the AIAN/White group: the probability of
selecting AIAN as the primary race and the probability
of selecting White as the primary race. These
probabilities were generated by county, single year of
age, sex, and Hispanic origin group. They are referred
to as the NHIS bridging proportions. To use this method,
apply these bridging proportions to the multiple race
population to get single race estimates. Then add these
single race estimates to the corresponding counts of
people who reported only a single race to get counts for
each race group. In some cases it may be necessary to
adjust the counts so that they add up to a particular
total.
We recommend using fractional assignment based on
these NHIS fractions. If the NHIS fractions are not
available, we recommend fractional assignment for those
reporting two races as follows:
- AIAN/White: assign ¼ to AIAN and ¾ to White;
- Black/White: assign 7/10 to Black and 3/10 to
White;
- NHOPI or Asian/White: collapse the NHOPI and Asian
groups to form an Asian and Pacific Islander group
(API) and assign ½ to API and ½ to White
- Assign equal fractions for people reporting more
than two races.
For more detailed analysis see the discussion paper Converting Multiple to Single Race below.
Note that geographic characteristics can affect the
choice of a bridging method. Knowledge of local
populations and how they are likely to respond (i.e.,
which single race they would be likely to choose) may
dictate a different bridging method.
Regardless of the method used, any report or table
that presents bridged data or rates calculated from
bridged data should note which bridging method was used.
- For documentation on the NCHS bridging method and
bridged population estimates see www.cdc.gov/nchs/nvss.htm.
- For a more detailed description and evaluation of
the various bridging methods see the
"Provisional Guidance on the Implementation of
the 1997 Standards for Federal Data on Race and
Ethnicity," (see webpage <http://www.census.gov/population/www/socdemo/race.html>)
|
