About the Modified Race Estimates (MARS)

Since 1980, the Census Bureau has produced estimates of age, race, and sex from the published census to correct shortcomings in the data. The 1990 file was called the Modified Age Race Sex (MARS) file. It corrected problems in age 0 and 1 because of an error in the way the age question was asked. The wording of the questions caused a 24 percent under-response in age 0.

Modifications were also made for “age heaping.” This is the tendency of people to round their age to the nearest five-year interval. Also, people avoid or “rush” certain ages. People may tend to avoid age 40 and rush 21 and 65.

The race question in the 1990 census recognized five race groups: White; Black or African American; American Indian, Eskimo, or Aleut; Asian and Pacific Islander; and Some other race. “Some other race” is a catchall for those who didn’t identify with the other four categories. The people who reported “Some other race” turned out to be predominately Hispanic. For the census, “Hispanic origin” is an ethnicity, not a race category. People of Hispanic origin may be of any race. The 1990 MARS file redistributed the “some other race” back into the remaining four race groups.

The U.S. Office of Management and Budget in 1997 revised standards for collection, tabulation, and presentation of federal data on race and ethnicity. The revised OMB standards identify five racial categories: White; Black or African American; American Indian and Alaska Native; Asian; and Native Hawaiian and Other Pacific Islander. In addition, respondents were allowed, for the first time, to identify more than one race. As a result, the Census 2000 reports a person’s race either alone or in combination with one or more races. In addition, the census kept the “Some other race” category. As in 1990, the chief group who responded “Some other race” were Hispanic. (For a complete discussion of 2000 race, see “Race and Ethnicity in Alaska” in the October 2001 Alaska Economic Trends.)

The 2000 MARS file has only modified the race of each age/sex group to adjust for the “Some other race” response. No adjustments have been made to age or sex to date.

Unfortunately, the 2000 Census had significant processing errors for forms from rural Alaska. The basic census form used in door-to-door enumeration allowed for only five household members to respond (six on the mailout form). If the household was larger, the persons were listed by name on the back and a supplementary form was used for the additional people. In the processing, the private data capture contractor separated the supplementary forms and the connecting information was lost. This meant that the age of children less than 18 years old had to be imputed for a substantial number of children based on the age distribution of similar households by a method statisticians refer to as a “Hot Deck.” The method assumed that people listed their children on the census form in random order, rather than sequentially. In fact, most people tended to list their children in age order.

As a result, a large number of children whose age had to be imputed were young children rather than a normal distribution by age under 18 years of age. The result was that for parts of rural Alaska with large households, the census reported too many children ages 10-17 and too few children 0-9. Ten census areas had errors of at least 6 percent. In some areas, the problem was severe. In Wade Hampton, we estimate that 16.0 percent of children had misreported ages, and Bethel had almost 15 percent. To date, the Census Bureau has not attempted to correct this problem because modifying Alaska’s data would lead to cumulative changes at the state and national level. It is also possible that the problem most clearly visible in Alaska’s data may exist with large families elsewhere in the United States.

USERS OF 2000 CENSUS DATA FOR CHILDREN BY AGE GROUP SHOULD BE AWARE THAT THE DATA FOR RURAL ALASKA AS REPORTED IN CENSUS TABLES MAY BE INACCURATE. Data for children under 18 as a whole should be accurate. Children were not missed by the census, but their reported age is inaccurate.

We have attempted to correct the age of children under 18 in the rural parts of ten boroughs and census areas utilizing age distributions from the Alaska Permanent Fund as a model. The corrected areas are: Wade Hampton Census Area, Bethel Census Area, Dillingham Census Area, Nome Census Area, North Slope Borough, Valdez Cordova Census Area, Northwest Arctic Borough, Yukon-Koyukuk Census Area, Southeast Fairbanks Census Area, and Lake and Peninsula Borough. Since the large families in these areas are predominately Alaska Native, the correction was applied only to the American Indian and Alaska Native race group. There also appears to be a small erroneous imputation of sex as well in ages 0, 1, and 2 for Alaska in the 2000 Census. Because of time constraints, we have not attempted to correct this error in this set of estimates.