One of the worst problems with conventional return to education estimates is that they ignore drop-outs. That's like a bank ignoring defaults when it calculates its return on loans. According to a recent experiment, a lot of parents ignore drop-out risk, too. From Andrew Kelly and Mark Schneider:
We asked a representative sample of one thousand parents of
high-school-age children in five states to choose between two public
colleges in their state based on their own judgments and information we
provided to them. Respondents were randomly assigned to a treatment or
control group. Treated respondents received the same set of basic facts
as the control group as well as information about each school's
six-year graduation rate...
we found that providing graduation-rate information increased the
probability that parents would choose the institution with the higher
graduation rate by about 15 percentage points. [emphasis mine]
Could giving minority students more choices make them worse
off?It could if they are
unrealistically optimistic about their probability of success, leading them to
choose an opportunity beyond their capabilities.Self-serving bias might also incline each
student to assume that he was admitted on his own merits: "If I were being
admitted because of affirmative action, I should be worried.But unlike many other students, I was accepted on my merits."
Consistent with this worry, Kelly and Schneider find that less-educated parents are much more influenced by the absence of information about the graduation rate. Parents with no college were 23 percentage-points more likely to choose the more selective school when they weren't told the graduation rate. The information effect for parents with college degrees was only 7 percentage points.
If you wonder, "How much does this have to do with real-world college selection?," Kelly and Schneider have a pretty good answer:
[B]ecause respondents were asked about
real public colleges and universities in their region and received true
information about school characteristics, the experiments discussed
below have considerable external validity.