The online version of the article doesn't include the following chart of ten-county population by race, attributed to the American Community Survey of the U. S. Census Bureau, which caught my interest:
|Black or African American||1115683||1167451||4.64%|
|American Indian and Alaska Native||8817||8848||0.35%|
|Native Hawaiian and Other Pacific Islander||1262||1768||40.10%|
|Hispanic or Latino||319101||371502||16.42%|
The point of the table, of course, is to illustrate that the Asian market has been growing in this area faster than the population as a whole, and that therefore it is logical that banks would be reaching out to this community.
Now, when I read this, something seemed suspicious. The population in every category except the neglible "American Indian and Alaska Native" one went up by at least 2.76%. But the total population went up by only 1.41%. Something is wrong here.
If you then add up the first six entries in each column, you get 5,582,372 and 5,836,718, respectively. For some strange reason, the sum of the racial numbers is 85,318 less than the total population in 2002, but 89,365 more than the total population in 2005. The obvious explanations are the following:
- "Hispanic or Latino" is being treated differently than the other items here, which is actually fairly common in discussions of race in America; basically if your ancestors come from somewhere Spanish-speaking, then you're somehow outside of the usual "race" categories.
- A lot more people have suddenly started to identify as multiracial between 2002 and 2005. A lot. Three percent of the total population. I'm aware that a lot more people have started to identify as multiracial recently instead of feeling like they have to pick one identity or the other, but it doesn't seem like it could happen that fast.
Usually on tables like this, one sees some note saying that the various pieces of population don't add up to the total population, due to certain minor categories not being listed, or due to people being included in more than one category. But there is no such note.
I had actually been anticipating something like Simpson's paradox, but it doesn't seem to be anything that complicated.
By the way, the ten-county area here is Bucks, Delaware, Chester, Montgomery, and Philadelphia counties in Pennsylvania; Camden, Burlington, Gloucester, and Atlantic counties in New Jersey; and New Castle county in Delaware. This strikes me as being slightly "off"; I would replace Atlantic with either Mercer or Salem. I suspect this doesn't make a huge difference in the data; however, it might be an example of "selective reporting", where one draws the boundaries in a nonstandard way to make some sort of point. The census bureau itself uses a nine-county area (the first eight above, plus Salem County, New Jersey); I'd say the most common definition of the Philadelphia area is as the first eight counties above.