Skip Navigation
Managing an Identity Crisis: Forum Guide to Implementing New Federal Race and Ethnicity Categories
NFES 2008-802
October 2008

Appendix C: The Good, the Bad, and the Problematic: Alternative Bridging Methodologies

NHIS Regression Method

In addition to these nine methods considered by OMB, there is a potentially more accurate bridging technique called the NHIS Regression method.  This method essentially addresses the question, "What characteristics make a multirace person likely to choose one of their component races as dominant over the other(s)?"

The NHIS Regression method, which is an extension of the NHIS Fractions method (2b), was developed at the NCHS12, 13 and is used by NCHS for its Vital Statistics program.  By taking into account primary race data and several geographic and demographic variables available on the NHIS or Census 2000, including information about the respondent's county of residence, such as region, level of urbanization, percent of county's residents who reported more than one race, and the age, sex, and Hispanic origin of the respondent, this approach has been found to result in bridging estimates that more accurately match the preferences of the multirace populations in question.  Using this method, regression models were developed for each of the major multirace groups and a "composite" model was developed for the smaller multirace groups. The regression coefficients obtained from fitting the models to the 1997–2000 NHIS were used to derive the probabilities of multirace respondents selecting each possible single race as their primary race.  For example, for an AIAN/White respondent, the probabilities of selecting AIAN as the primary race or White as the primary race were derived for each county, age, sex, and Hispanic-origin combination.

Using this methodology to bridge local multirace populations would basically require someone at the state or local level to apply the NHIS probabilities to their own data.  The use of the NHIS probabilities would be complicated only because they are county-age-sex-Hispanic origin-specific.

  • Race 1/Race 2 — Fraction of individuals to Race 1, another fraction to Race 2 based on probabilities derived from the regressions on NHIS data

For more information on this methodology and its logistics, see Ingram et al. (2003) and the NCHS race bridging web site at http://www.cdc.gov/nchs/about/major/dvs/popbridge/popbridge.htm.

Why not just prorate?  A cautionary note

An alternative method to bridging that may seem at first to be the obvious and logical approach, but may actually be quite problematic, is proration based on racial distributions in the education agency's population.  By this methodology, a district could simply use the relative proportions of the race groups in their populations to generate probabilities for race assignments.  For instance, if a district had 700 White students and 300 Black students, White/Black multirace students would be divided 70/30 into these respective single-race categories. While this method may seem like a reasonable way to go, it will likely produce poor bridging estimates.

Remember that the goal of bridging is to estimate what the racial distribution of a population would have been had individuals been allowed only to select a single race category.  Its purpose is to enable trend analyses with data collected before and after a shift to the 1997 standards.  Since the selection of racial identity is a function of individual preferences, the bridge estimate should seek to approximate those preferences.  The relative sizes of racial groups in the population do not necessarily resemble those preferences.

Using proration to assign multirace individuals to a single race group will likely produce inaccurate estimates because this technique assumes that the relative sizes of the single race groups determines multirace individuals' preferences for identification with those groups.  The preferences of some multirace populations may happen to align with such a distribution, but those populations are not likely to be the norm.  Multirace White/AIAN individuals, which comprise one of the largest multirace groups, present perhaps the most extreme example of the possible disparity between racial preferences and racial population distributions.  The AIAN population is usually quite small compared to the White population, so the use of proration to bridge AIAN/white individuals would result in most being assigned to the White population and a very small  proportion being assigned to the AIAN population.  Such an assignment would be erroneous though because White/AIAN individuals are much more likely to choose AIAN as their single race than they are to choose White.14  While this erroneous assignment generally would have little impact on estimates of the White population, it could result in substantial underestimation of the AIAN population because such a large proportion of individuals of AIAN ancestry identify themselves as multirace.  For such reasons, we discourage the use of proration as a bridging technique.

Why not just base probabilities on changes in racial distributions over time? A second cautionary note

Another alternative method of deriving bridging estimates is to base probabilities on the year-to-year changes in racial distributions in a school or district.  While it may seem like a logical approach, it too may be problematic.  For instance, after the shift to the new collection standards, education agencies will likely see, along with new multirace students and staff, drops in the various single race groups (i.e., if there are 30 Black/White multirace students in a district, there will likely be an associated drop in the Black and White population totals compared to the previous year equal to about 30 students.  Let's say, for instance, that the number of White students drops by about 20 for that grade cohort since the previous year and the number of Black students drops by about 10 students.  The district could theoretically derive a probability for bridging Black/White children based on those relative differences.  However, this technique is problematic for a number of reasons. 

For example, because of the migration of students and staff in and out of the district from year to year, data from consecutive years will not be directly comparable and the differences in the single race groups will probably not add to exactly 30.  Also, people tend to be inconsistent in the way they identify their racial identity.  A person who is White/Black this year may have identified as White in the previous year and as Black the year prior to that.

As the flux of populations and reported identities can render this technique unreliable, we do not endorse the use of this approach in districts with unstable or racially heterogeneous populations.

Primary Race: An Alternative to Bridging?

A potentially useful avenue to pursue that could either eliminate the need to bridge or at least limit the scope of the bridging that is required, is to collect an additional data item from students and staff during the bridge period, called "primary race/ethnicity."  Like the NHIS has done for more than a decade, and similar to what the state of Kansas has done since the 2005–06 school year, education agencies might consider including an additional question on race and ethnicity, which asks multirace respondents to select one race with which they most identify or how their community most commonly recognizes them.  The options for this question should be the race and ethnicity categories under the 1977 standards.  For trend analyses, states can simply use this primary race, thus avoiding the need to use a bridging methodology.  Instead of estimating how the multirace individuals in a population would have identified themselves if limited to the single-race system, this primary race question would ask them directly.

The main problem with this approach is the issue of nonresponse.  For those multirace individuals who select a primary race, the need to bridge is averted.  However, since this item cannot be required by the state, some multirace individuals may refuse to designate a primary race.  For this portion of the multirace population, however small, bridging will still be necessary.  However, with smaller numbers of respondents needing to be bridged, any distortions that result from the bridging method used will be limited.  Additionally, the information on preferred primary race assignment that is obtained from the multirace individuals who do respond could be utilized as locally specific bridging probabilities that could be used to bridge the nonrespondents.

Conclusion

In the search for a bridging methodology, states must consider a number of factors.  The characteristics of local populations as well as the capabilities of school district staff and data systems should all be weighed in the choice of a bridging technique.  Additionally, the merits and characteristics of the individual bridging methodologies must be considered.  States may frame their assessment of these methodologies with a focus on the balance between ease of use and implementation and the quality of the bridging estimate.  In addition to deliberating on these questions, states may also opt for an empirical approach, trying out a number of the methods discussed in this report with data collected under the 1997 standards and comparing the resulting estimates to prior years' data.  Assuming that local race and ethnicity distributions will not change very much from one year to the next, good matches between the racial and ethnic distributions created in the bridge estimates and those of the previous year's population may indicate good bridging and, thus, inform a decision on which method to use.

While the number of multirace individuals in a local population is likely to be small, considering the national rate was only 2.4 percent as of 2000,15 it will tend to be higher among students than teachers, and among younger than older student cohorts. When the number of multirace individuals is small, the distortions created by bridging may be minimal. However, the percentage of the population that reports itself as multiple-race varies considerably from state to state and from county to county. Moreover, some multiple-race groups are more prevalent in some areas than in others, members of some multiple-race groups are more likely to report multiple races than members of other multiple-race groups, and members of some multiple-race groups are more likely to identify most closely with the majority race than others are. States with a large number of multirace students and staff may be more affected by their choice of methodology, and therefore may be more inclined to spend more resources to pursue a more involved method such as NHIS Fractions or NHIS Regression to get a potentially more accurate estimate. In both cases, states may consider collecting "primary race" data as a way of limiting the size of the multirace population that will need to be bridged.

Though bridging is necessary for only a small portion of the population and will only be needed for a few years at most, the decision to bridge and, thereafter, of which method to use can have a great impact on a state's student and staff data.  To ensure that the data are of the highest quality possible, great care should be taken in crossing this bridge.

Top


12 Schenker and Parker (2003).
13 Ingram et al. (2003).
14 National Health Interview Survey.
15 United States Census Bureau (2000).