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 >>>

Converting Multiple to Single Race

Washington State Department of Health discussion paper developed December 2002 and updated April 2003

Paper prepared by Juliet VanEenwyk
Data prepared by Catherine O'Connor

For more information contact Juliet VanEenwyk (juliet.vaneenwyk@doh.wa.gov).

Background

The 2000 US census allowed people to report more than one race. Prior US censuses and several key Washington data systems that are used to develop information on health status allow for the reporting of one race only, or if they allow more than one race, they collect this information differently from how it was collected in the 2000 US census. As a result, reporting of more than one race is substantially lower in these data systems than it is in the 2000 US census. Until health and related data systems capture race in a manner similar to that of the US census, developing rates of health events by race is problematic for rates that use data from the Washington data systems in the numerator of the rate and data from the census in the denominator. There are several potential methods, referred to as bridging methods, for adjusting race as reported in the census so that it is compatible with race as collected by Washington's data systems.

The Washington State Department of Health (DOH) recommends using the National Center for Health Statistics (NCHS) multiple to single race population counts for 2000 US Census data and the NCHS bridging algorithm for bridging other years or sources of data. Before the NCHS bridged population counts were available, DOH worked with Public Health Seattle & King County (PHSKC), Department of Social and Health Services (DSHS), and the Office of Financial Management to determine how to bridge multiple to single race in Washington. The following discussion explains our methods, results, and recommendations.

Methods

DOH, PHSKC and DSHS considered several methods of using the race data collected in the 2000 US Census.

Whole allocation to all or no race groups

This method involves providing a range of rates, representing the full range of rates that would be developed if we knew how people reporting multiple races would have reported their race if asked for report one race only. The range is developed by assuming that everyone who reported multiple races would report the same race if they were required to choose only one race. Thus, the maximum number of people of a particular race, say race A, is the number who reported race A only and the number who reported race A in addition to other races. The minimum number of people of race A is the number who reported race A only.

Thus for the maximum number of people in any race group, each race group is assigned one person for anyone who reported that race as a single race or in combination with other races. For example, both blacks and whites would get a whole person for everyone who reported that they were black and white. This method overestimates the number of people in any race one group, because people reporting more than one race are counted multiple times. Since overestimating the number of people in the denominator of a rate will underestimate the rate, this approach provides rates that are the low end of the range of possible rates.

For the minimum number of people in any race group, each group is assigned only those people who reported one race and people who reported two or more races are not counted. Since this method does not include everyone, it underestimates the number of people in any race group. When the number of people in the denominator of a rate is underestimated, the rate will be overestimated. Thus, this approach represents the upper end of the range of possible rates.

Fractional allocation to race groups

This method attempts to quantify rates more precisely by counting each person exactly once. Rather than assuming that everyone who reported more than one race would choose the same race if required to select one race only, this approach attempts to quantify how many people would choose each race if required to choose only one. We considered two methods for fractional allocation. Both methods involve assigning fractions of people to every race group reported such that each person is counted exactly once.

One method assigned equal fractions to each race group. For example, both black and white race groups received half a person for everyone who reported that they were black and white. The second method used fractional assignment based on surveys that have asked people who report more than one race which race they would select if asked to pick one race only. We were able to assess reports from the 2000 Washington State Survey of Adolescent Health Behavior (WSSAHB) and the 2000 and 2001 Behavioral Risk Factor Surveillance System (BRFSS). Additionally, we had limited information available from the National Health Interview Survey.

The Washington State Population Survey also allowed respondents to select more than one race and then asked which race they would select if they could choose only one. However, the 1998 data were difficult to interpret and later years of data were not available to us. Thus, data from this survey could not be used to inform fractional assignment.

Results

Number of people reporting more than one race

Table 1 shows reporting of race in Washington from the 2000 US census, as compiled in the 2000 MARS (Modified Age, Race, and Sex) file and with people reporting Asian and/or Native Hawaiian or Other Pacific Islander (API) combined into one group following the former federal Office of Management and Budget (OMB) classifications. The MARS file imputes race for the approximately 4% of Washingtonians who indicated "other" race in the census. Historically, the majority of people in Washington who report their race as "other" are of Hispanic origin.

Approximately 3.6% of Washington residents reported they were more than one race in the 2000 US census. On the MARS file and when Asians and Native Hawaiians and other Pacific Islanders are treated as one race, approximately 2.6% of Washington residents are more than one race. Of these, the largest groups are white and API, white and American Indian or Alaska Native (AI/AN), and white and black. These groups each represent about 0.5% to 1% of the total population. All other multiple race groups each represent less than 0.1% of the total population. In determining fractional allocation, the allocation of the three largest groups will have the largest affect on the overall number of people in any single race group.

Table 2 shows the proportion of people in Washington with more than one race by age group. The percent of people with more than one race decreases with older age.

The percent of people with more than one race also varies by county. For example, more than 10% of children under 10 in Pierce County were reported as having more than one race.

Information for fractional allocation with unequal fractions

Results of how people who reported multiple races classify themselves when asked to select one race only are presented below for the three largest multiple race groups. Information on other race groups is unreliable due to small numbers. This information was used to determine fractional allocation. Table 3 provides detail on the results summarized below.

White and API

• Although this is the largest multiple race group in the 2000 US census, only 55 respondents to the 2000 and 2001 BRFSS reported they were white and API. Of these approximately half selected white and half selected API when asked to pick a single race.
• On the 2000 WSSAHB, 290 students said they were white and API. Of these approximately 40% selected API and 60% selected white when asked to choose one race.

White and AI/AN

• Of 160 BRFSS respondents who reported they were white and AI/AN, approximately 20% said they were AI/AN and 70% said they were white when selecting a single race, with the remainder choosing "other" or "no preference." If those choosing other or no preference are omitted, approximately 25% reported AI/AN and 75% reported white.
• Of 451 WSSAHB respondents who said they were white and AI/AN, approximately 25% reported AI/AN and 75% reported white as a single race.

White and Black

• Only 10 respondents reported both black and white on the BRFSS. These are too few respondents to assess how they might classify themselves if asked to report one race only. However, the results for these 10 respondents are not inconsistent with other data.
• On the WSSAHB, 111 respondents reported they were black and white. Of these, approximately 70% selected black and 30% selected white when asked to choose one race only.
• On the 1993 - 1995 National Health Interview Survey, of those who said they were both black and white and reported a single race preference, approximately 70% preferred black and 30% preferred white.¹

Table 3. Percent¹ and 95% confidence interval of for how people choosing multiple races classifying themselves if asked to select one race only.

	API2	AI/AN3	Black	White
White and API
BRFSS4 (n=55)	43% ± 14%			49% ± 14%
WSSAHB5 (n=290)	39% ±   8%			57% ±   8%
White and AI/AN
BRFSS (n=160)		22% ± 7%		72% ± 7%
WSSAHB (n=451)		26% ± 5%		73% ± 5%
White and Black
BRFSS (n=10)			58% ± 31%	26% ± 28%
WSSAHB (n=111)			70% ± 15%	29% ± 16%
NHIS			70%	30%

¹ Percents do not always add to 100% because some people do not answer the single race question or select a race other than that selected when allowed to choose more than one race.
² Asian and Pacific Islander
³ American Indian and Alaska Native
⁴ Behavioral Risk Factor Surveillance System (2000, 2001)
⁵ Washington State Survey of Adolescent Health Behavior (2000)

Based on these findings, the Assessment Operations Group (AOG) at DOH recommended the following fractional assignment:

• Assign ¼ of each person selecting AI/AN and white to AI/AN and ¾ to white. Both the BRFSS and the WSSAHB are consistent with this apportionment.
• Assign .7 of each person selecting black and white to black and .3 to white. The WSSAHB and the NHIS are consistent with this allocation and the BRFSS, while unreliable, is at least not inconsistent with this approach.
• Assign ½ of each person reporting API and white to API and ½ to white. This allocation is based on the fact that data from the BRFSS and the WSSAHB are not entirely consistent and thus there is no clear pattern of how people reporting both API and white classify themselves if asked for a single race; that 50% is within the confidence intervals for three of the four percentages presented above; and that in the absence of countervailing information, equal allocation may be the best method of allocation.
• Allocate people in all other multiple race groups using equal allocation since we have no data on which to base unequal fractional assignment.

^[1] Based on table published in ‘Provisional Guidance on the Implementation of the 1997 Standards for Federal Data on Race and Ethnicity,’ Office of Management and Budget, December 15, 2000.

<<< Previous <<<   >>> Next >>>