Misshapen Statistics on Racial Quotas
Sowell, Thomas, The American Spectator
Racial preferences in college admissions are good for minorities and society at large, argue two former Ivy League presidents. But their numbers don't add up.
Back in 1936, a leading magazine of the time, the Literary Digest, made a prediction about the presidential election of that year, based on a poll with a very large sample-more than 2 million people. Their poll indicated that Gov. Alfred Landon of Kansas would defeat President Franklin D. Roosevelt's bid for re-election. When FDR carried 46 of the 48 states at that time, this not only contributed to the discrediting and demise of the Literary Digest, it taught a lesson about the use of statistics. The size of a sample is by no means as important as its representativeness.
That lesson seems to have been lost amid a chorus of media and academic approval of a recent study of affirmative action in college admissions by ex-college presidents William G. Bowen (Princeton) and Derek Bok (Harvard). Their book, The Shape of the River (Princeton University Press, 472 pages, $24.95), has a sample size of more than go,ooo college students, but the sample is highly unrepresentative. Moreover, the authors' internal breakdown of this sample makes comparisons among colleges that are like comparing apples with oranges. The widespread approval of the Bowen and Bok study may have much less to do with its statistical methods than with its politically correct conclusions.
The central message of their book is that racial preferences and quotas in college admissions-"race-sensitive admissions" is their euphemism-have been beneficial and have not created the problems claimed by critics of affirmative action. But that conclusion rests on their statistics-and on their ignoring other statistics which indicate otherwise.
That the Bowen-Bok sample used is unrepresentative of college students-black and white-is easily demonstrated. In this sample, the graduation rate of black students is 75 percent. In other samples collected by other individuals and organizations, black graduation rates are less than 5 percent. Indeed, black students' graduation rates are only about half of the graduation rates of white students, many of whom also drop out. At Berkeley, for example, the black graduation rate has been 30 percent-but Berkeley is not part of the Bowen-Bok sample.
The authors of The Shape of the River admit that they are studying students in an atypical group of elite colleges and universities, rather than a representative national sample of either students or colleges. Yet Bowen and Bok repeatedly make sweeping declarations about national policy, just as if they had not admitted the serious limitations of their data.
For example, they make much of an apparent refutation of critics who have blamed racial preferences for high black failure rates at colleges with which these students have been mismatched academically. This apparent refutation is based on comparisons of the graduation rates among black students attending colleges and universities that Bowen and Bok break down into three groups, according to how high the average SAT scores are at the institutions in these groups. Then the authors triumphantly show that black students graduate at a higher rate on the campuses with the highest SAT scores. But there is less here than meets the eye.
First of all, these are not comparisons among individual colleges and universities. Bryn Mawr, Duke, Princeton, Rice, Stanford, Swarthmore, Williams, and Yale are all lumped together as the top group with higher SATs, while Denison, Miami of Ohio, Michigan (Ann Arbor), North Carolina (Chapel Hill), Penn State, and Tulane are lumped together as the bottom group with the lowest SATs within the elite BowenBok sample. Unfortunately, these two sets of colleges differ not only in SAT scores but also-and dramatically-in size.
The median number of undergraduates on campuses in the top group is today less than 3,000, while the median number of undergraduates on campuses in the bottom group is more than 13,000. …