There is no way to measure exactly how much racial groups changed between 1990 and 2000 because of new categories in the census. The fact that respondents can choose more than one race creates a whole new playing field. The best hope is to identify clear trends, then convey them in a way that doesn't mislead readers into thinking that it's an exact science.
Here are a few simple techniques for looking at change:
Compare the percentage of 'minorities.' Look at how many people list themselves as something other than non-Hispanic white to get a sense of whether the community has gotten more or less diverse. Let readers compare the apples and oranges for themselves.
Look at how many people list themselves as something other than non-Hispanic white to get a sense of whether the community has gotten more or less diverse.
Put multiracial respondents in their own category. Then, run pie charts with the new 2000 categories and the old 1990 categories side by side. Readers will see that things have changed, but they will also get a sense of the general trends.
When studying a single race, look at the range of potential change. Say you want to look for changes in the American Indian population alone. Look at both the number of people who described themselves as being American Indian "alone" and the larger group of those who describe themselves as being partly or wholly of American Indian heritage. Compare that to the 1990 population. If both comparisons show an increase or a decrease, you have a clear trend. If not, it's impossible to conclude what's really happening.
Look at Hispanics as if they were a race. Build 1990 and 2000 pie charts showing Hispanics displayed as a race, alongside non-Hispanics in each race (non-Hispanic whites, non-Hispanic blacks, non-Hispanic Asians, etc.). This works well in areas like the West where most Hispanics are white. If there are many Hispanic blacks, it will result in an undercount of blacks.
Pitfalls to avoid:
Sometimes it will look like the population of a race has decreased when you compare the number of people who describe themselves as being of that race alone to 1990 figures. But, the population will appear to have increased when you compare those who chose that race alone or in combination with another race. This is a statistical tie. Because of the sometimes huge margin of error brought in by the new mixed-race categories, it's impossible to tell what's really going on.
Census tract boundary changes
When comparing 1990 census tracts to 2000, make sure the Census Bureau hasn't quietly rejiggered the tract boundaries. Usually, when the Census Bureau splits or combines tracts, it changes the tract number, so it's easy to tell there has been a change. But, we found many cases where the bureau shaved off a tiny sliver of a tract with no announcement. It often seemed to be a sliver that had a lot of the population in it, making it easy to draw some pretty wild conclusions about unexplained population declines.
Erin McCormick can be reached by e-mail at emccormick@sfchronicle.com