How We Know What Isn't So (8 page)

The problem of hidden or absent data has also clouded many people’s thoughts about the effectiveness of the Scholastic Aptitude Test (SAT) in predicting students’ success in college. The SAT has been criticized as a poor predictor of college success because the relationship between SAT scores and college GPAs among students in most universities is quite modest (the correlation coefficient is generally about 0.2). Note, however, that students with very different SAT scores tend not to enroll in the same schools: Those with high scores attend the most prestigious institutions and those with lower scores attend less renowned schools. Thus, students with very different SAT scores are never assessed together, and so the correlation between SAT scores and GPAs within a university cannot tell us how students with very different SAT scores would perform in the same environment. All that the modest correlation between SAT’s and GPA’s can tell us, then, is that the SAT does not make very fine discriminations—someone who gets, say, a 610 cannot be counted on to get a higher GPA than someone who receives a 570.

But perhaps the test can make gross discriminations with greater accuracy. Maybe someone who gets a 610 can indeed be counted on to get better grades than someone who receives a 410. If so, we would expect there to be a much higher correlation between SAT’s and GPAs among a group of students with a wide range of SAT scores. There is. There are a few schools that have an open enrollment (and thus do not admit only the best prospects), but that, due to their location, still attract many students with very high SATs. Among these more heterogeneous students, correlations as high as .6 to .7 have been observed.
¹⁵
Students with high SAT scores are indeed more likely to do well in college than those with low SAT scores. By looking only at the restricted data from schools with the usual patterns of enrollment, many people have failed to recognize the true effectiveness of the SAT (in terms of making gross discriminations between applicants). The problem of absent data has thus served to misdirect much of the discussion of whether to use the SAT in admissions decisions.
*

Moving away from the domain of policy evaluation, it should be clear that the problem of hidden or absent data also affects the kinds of inferences we draw and the beliefs we have about everyday social life. Oftentimes, the lifestyles we lead, the roles we play, and the positions we occupy in a social network deny us access to important classes of information and thus distort our view of the world. At large research universities where there is less informal contact between students and faculty than one would like, professors learn early on that unless they are careful, it is easy to be exposed mainly to the alibis and complaints of the most difficult students and rarely see the more successful and more pleasant students who make teaching so gratifying. Similarly, the experience of therapists who treat alcoholics appears to predispose many of them to expect the worst from even the most temperate drinking. We can occasionally overcome our limited exposure to relevant data, but doing so is difficult: It requires that we not only recognize the existence of a class of information to which we have not been exposed, but that we accurately characterize what that information is like.

There is a special version of the hidden data problem that arises whenever our expectations lead us to act in ways that fundamentally change the world that we observe. When this happens, we often accept what we observe at face value, with little consideration of how things might have been different if we had acted differently. Sociologist Robert Merton used the term “self-fulfilling prophecy” to describe this phenomenon, and he gave the example of how a false rumor of a bank’s insolvency can generate a panic that creates the very insolvency that was initially feared.
¹⁶

There are several aspects of self-fulfilling prophecies that warrant further analysis and discussion. First, because self-fulfilling prophecies have received so much attention, there is some danger that their impact can be exaggerated. Not all prophecies are self-fulfilling. As psychologist Robyn Dawes has noted, some can even be self-negating, as when a reckless driver claims that “nothing bad can happen to me.”
¹⁷
For a prophecy to be self-fulfilling, there must be some mechanism that translates the expectation into confirmatory action.

An informative example of the limits of self-fulfilling prophecies is provided by an experiment in which the participants played numerous rounds of a standard “prisoner’s dilemma” game.
*
After hearing the rules of the game, the participants were asked to articulate their opinion of the proper orientation toward the game. Some (“cooperators”) stated that they thought the point of the game was to cooperate with one’s partner in order to maximize their joint outcomes. Others (“competitors”) said that they thought the purpose was to compete strategically with one’s partner in order to maximize one’s own individual outcomes.

The cooperators and competitors were not equally successful in having their views of the game confirmed. If a cooperator was paired with another cooperator, they quickly began making mutually beneficial, cooperative moves. When paired with a competitor, the cooperator was forced into more competitive actions in order to avoid consistent losses. Competitive players, in contrast, always ended up in a cut-throat game: When paired with another competitor, the game quickly settled into an internecine struggle; when paired with a cooperator, their own actions forced the potential cooperator to become competitive out of self-defense.
¹⁸
Thus, because competitive behavior creates more of a demand for the other person to respond in kind than does cooperation, a competitive person’s belief that the world is full of selfish opportunists will almost always be confirmed, whereas the less gloomy orientation of cooperative individuals will not. Sadly, negative prophecies are often more readily fulfilled.

Another, often-neglected point about the limits of self-fulfilling prophecies is that they usually serve to exaggerate a belief that contains a kernel of truth, rather than create one that is completely erroneous. Rumors of insolvency generally plague banks that are in fact having difficulty. Suppositions that a student might be exceptionally gifted are generally made about students who do in fact have superior intellectual talent. This point often goes unnoticed because of the logic behind the experiments that have examined self-fulfilling prophecies: To show that a teacher’s expectations can influence students’ achievement, for example, it is imperative that the teacher be given different expectations about students who are in fact equal in achievement. Any subsequent differences in performance can then be confidently attributed to the teacher’s expectations. In the real world, however, expectations are not generated randomly, but by cues from the environment. Thus, self-fulfilling prophecies generally turn little effects into big effects, rather than create effects from scratch.

A final point to be made about self-fulfilling prophecies is that there are really two kinds—true self-fulfilling prophecies and seemingly-fulfilled prophecies. True self-fulfilling prophecies are like those already discussed in which a person’s expectation elicits the very behavior that was originally anticipated. Behaving in an unfriendly and defensive manner because you think someone is hostile will generally produce the very hostility that was originally feared. Seemingly-fulfilled prophecies, on the other hand, refer to expectations that alter another person’s world, or limit another’s responses, in such a way that it is difficult or impossible for the expectations to be
disconfirmed
. Thus, the expectancy is confirmed, not by the target person actively conforming to some expectancy, but by the target having little opportunity to disconfirm it. If someone thinks that I am unfriendly, for example, I might have little chance to correct that misconception because he or she may steer clear of me. The absence of friendliness on my part could then be construed as unfriendliness. When little-league baseball players are thought to be incompetent, they are only allowed to play where the ball is rarely hit (for little leaguers, in right field), and thus they have few opportunities to overcome their unfortunate reputation. The continued absence of any positive contributions can then easily be mistaken for an absence of talent rather than an absence of opportunity.

This type of expectancy effect is obviously a special case of the hidden data problem described above. A perceiver’s expectation can cause him or her to behave in such a way that certain behaviors by the target person cannot be observed, making what is observed a biased and misleading indicator of what that person is like. The employers, college admissions officers, and grant review panelists discussed earlier are all potential victims of seemingly-fulfilled prophecies: Their own actions guarantee that they will rarely receive a challenge to their negative assessments of job applicants, potential students, and research proposals. The research on people’s hypothesis-testing strategies that was discussed earlier also provides a good example of a seemingly-fulfilled prophecy: By asking people they suspected to be extroverts what they do to liven things up at a party, one compels them to talk about their most sociable leanings and thus is prevented from observing much in the way of introversion.

The existence of seemingly-fulfilled prophecies implies that negative first impressions should generally be more stable (i.e., less subject to change) than positive first impressions. If we find another person unpleasant initially, we try to avoid that person as much as possible, and he or she will have a difficult time disabusing us of our negative assessment. If we like another person, on the other hand, we seek out his or her company and thereby give him or her ample opportunity to ruin our hopes and expectations.
²⁰
This can sound rather grim, but it does have a positive flip-side: It suggests that our negative assessments of other people are less likely than our positive assessments to be correct, and we should give our foes another chance.

From the perspective of trying to understand questionable and erroneous beliefs, it should be clear that the impact of self-fulfilling prophecies is similar to that of the confirmatory search strategies and hidden data problem described earlier. All of these processes serve to provide us with incomplete and unrepresentative samples of information from which we draw conclusions and evaluate beliefs. Unless we recognize these sources of systematic distortion and make sufficient adjustments for them, we will surely end up believing some things that just aren’t so.

*
Interestingly, it has been shown that people can do much better at this task—i.e., they are more likely to turn over the correct cards, and
only
the correct cards—if it is embedded in just the right substantive context. For instance, suppose you are trying to test the rule, “everyone who drinks alcohol is over 21 years old.” In front of you are 4 cards with a person’s age on one side and what he or she is drinking on the other. The four cards are “drinking beer,” “drinking Coke,” “25 years old,” and “16 years old.” Which would you examine? Most people correctly turn over the cards “drinking beer” and “16 years old,” and do not show a preoccupation with potentially confirmatory information by turning over the “25 years old” card. [See P. W. Cheng & K. J. Holyoak (1985) Pragmatic reasoning schemas.
Cognitive Psychology, 17
, 391-416; P. W. Cheng & K. J. Holyoak (1989) On the natural selection of reasoning theories.
Cognition, 33
, 285-313; L. Cosmides (1989) The logic of social exchange: Has natural selection shaped how humans reason? Studies with the Wason selection task.
Cognition, 31
, 187-276.]

This improved performance is not obtained by embedding the task in just
any
context, but mainly in those that invoke the idea of “permission”—e.g., a person must be over 21 years old to be permitted to drink alcohol. The fact that there are
some
domains in which people are not preoccupied with confirmations, of course, does not undermine the general finding that people often test their hypotheses by seeking out potentially confirmatory information.

*
Note that this discussion addresses only one of the criticisms of the SAT exam—that it supposedly does not adequately predict college GPA. As a proper analysis of the problem makes clear, this criticism is misguided. However, there are other criticisms of the SAT exam that this analysis does not address, such as whether the test is culturally biased and whether college GPA is really the most desireable measure of college performance.

*
The Prisoner’s Dilemma is the most widely researched experimental game used to study conflict and “social dilemmas.” In the original version, two partners have committed a crime and are interrogated separately by the district attorney (A. Rapoport & A. Chammah [1965]
Prisoner’s dilemma
. Ann Arbor: University of Michigan Press). The DA has only enough evidence to convict the two suspects of a lesser offense, so he offers each a chance to confess privately in order to “get the goods” on the other. If one suspect confesses and the other does not, the one who confesses will be granted immunity and the one who does not will receive a harsh sentence of, say, 10 years. If neither confesses, they each receive the penalty for the lesser offense—say, 1 year. If both confess, they each receive a moderate penalty of 5 years.

The participants in a Prisoner’s Dilemma experiment must decide whether they would confess (and thus “defect” from or compete with their partner) or not (and thus cooperate with their partner). Note that it is always better for a person to defect, regardless of what his or her partner does (doing so gives the player 5 years rather than 10 if his or her partner confesses, and 0 years rather than 1 if his or her partner does not confess). However, if both players confess, their fate (5 years each) is clearly much worse than if both do not confess (1 year each).