We’re lucky to have Michigan man and highly decorated social scientist Robert Groves running the U.S. Census Bureau. Yet, when Groves worked at the Census Bureau on the 1990 count and suggested adjusting the census’s official numbers based on sampling that showed severe undercounts for children and minority groups, he was shut down by the first Bush administration’s Commerce Secretary Robert Mosbacher.
And in a 2009 article in Time magazine, U.S. Rep. Darrell Issa (R–Calif.) suggested that President Barack Obama’s selection of Groves was “troubling” and would open the door to advance “an ulterior political agenda.” From the perspective of scientific accuracy and political justice, that 1990 decision and Issa’s complaints are misguided. Social scientists like Groves have developed statistical tools that allow us to make more accurate inferences about populations than do direct attempts at counting. As currently designed, the census undercounts minority communities, unjustly costing them federal aid.
The problem is that the direct count undertaken by the census is not some ideal. Instead, the direct count is a survey that still has all the problems that any other survey does, as Professor of Political Science and Research Associate Professor for the Center for Political Studies Vincent Hutchings pointed out in the Daily’s Apr. 2, 2009 article detailing Groves’ nomination.
Although it tries to count everyone personally — unlike a national sample used in, for example, the Gallup poll measuring Obama’s approval rating — the census still falls prey to all the potential pitfalls of any survey research.
Here’s how: The first and most important key for any survey’s accuracy is that each member of the population you are trying to survey must have an equal chance of being selected for the survey. Because it is trying to count everyone directly, in principle the U.S. Census questionnaire satisfies this requirement — every resident has a 100-percent chance of being selected. But in reality, the count fails to include every resident. This failure would not be a massive problem if members of every demographic group were equally likely to respond to the survey, but the truth is some groups are less likely to be counted than others. That creates non-random error, which leads to biased results.
College students, for example, are often confused by census instructions and do not respond to the survey, thinking that they will be counted with their parents in their hometowns. The Daily reported on Jan. 31 that Ann Arbor’s overall response rate to the 2000 census survey was 76 percent, but one primarily student-intensive neighborhood had a response rate of 38 percent.
Of course, investigators then follow up the survey in an attempt to count everyone, scouring the country — with special focus on neighborhoods with low response rates. But despite their best efforts, they miss people and they miss them in a non-random fashion. Again, taking the college student example, the Daily reported that follow-up investigations to low-response areas by census takers that start in late April tend to miss college students, many of whom go home for summer vacation starting in May.
Though missing college students is bad, the problem is much broader. After the census is completed, follow-up researchers randomly select samples of residents and ask them if they weren’t counted, were counted once or counted multiple times. The results of those follow-up surveys show that some groups tend to be undercounted much more severely than others. For example after the 1990 count, the Census Bureau estimated that African Americans were undercounted by a rate of 4.4 percent and Hispanics by roughly 5 percent. In contrast, whites were undercounted by only 0.9 percent. Children under 18 were undercounted by 3.2 percent – double the overall rate. Minority children were undercounted, most significantly among African Americans, by rates up to 7.1 percent.
These discrepancies have serious political and economic consequences. Since the census affects apportionment and redistricting of congressional seats, states with large numbers of minorities might miss out on representatives. More important is the lack of federal money flowing to communities that need it. Communities of minority children tend to be clustered together in major cities, which then don’t get their fair share of government funding for education and other social programs.
Of course, when social scientists like Groves have pointed this out in the past, they have been accused of being biased and political opportunists for Democrats, who tend to be supported by undercounted minority groups. In this case, though, the truth is that the only bias is in the original direct count — and against the correctly funded demographic groups which are undercounted through no fault of their own.
So, if post-count surveys indicate that census data need adjusting, Republicans ought to shut their yaps and let the professionals make the adjustments.
Patrick O’Mahen can be reached at pomahen@umich.edu.