Equal Opportunity Framing
Of course, framing would be less of an issue if we were not such cognitive misers. Perhaps frame-free politics, where issues could be decided on their real merits, is too much to ask for in the near term. But personally autonomous decision making that is free of arbitrary framing effects is not too much to ask for. The mental operations that are necessary if we are to avoid framing instability are not difficult to acquire.
Framing effects are the source of much dysrationalia, because, interestingly, the tendency to respond passively to the frame as given is largely independent of intelligence. Here a brief word about research methodology is needed. Framing experiments—and most other experiments on rational thinking—can be run either between subjects or within subjects. For the disease framing problem discussed previously, for example, in a between-subjects design one group of subjects would be presented with the gain version (“200 will be saved”) and a different group of subjects would be presented with the loss version (“400 will die”). Random assignment of subjects to conditions ensures that the two groups are roughly equivalent and that the response patterns obtained from them are comparable. In a within-subjects design, each subject responds to both versions of the problem. Usually the two versions are separated in time so that the relation between the problems is not completely transparent. Of course, in within-subjects experiments the two versions are counterbalanced—one half of the subjects receive the gain version first and the other half of the subjects receive the loss version first.
Not surprisingly, between-subjects experiments show large framing effects because this design contains no cue that there is an issue of consistency at stake. Interestingly, however, the magnitude of the framing effect is not at all related to intelligence in this design.10 So when given no cue that they should be consistent, the higher-IQ people in the research sample are just as likely to be framed by irrelevant context as the lower-IQ people.11 The results with within-subjects designs are slightly different. Framing effects still occur in these designs, although they are not as large as those obtained in between-subjects designs. Also, in within-subjects designs, there is a statistically significant association between the magnitude of the framing effect and intelligence—higher-IQ individuals are slightly less likely to show a framing effect.
In short, subjects of higher intelligence are somewhat less likely to show irrational framing effects when cued (by the appearance of both problem versions) that an issue of consistency is at stake; but they are no more likely to avoid framing without such cues. We need to stop here and ponder one implication of the between/within findings regarding intelligence. I need to draw out the implications of the findings by speaking of them in a more colloquial way. The point is that, increasingly, cognitive science is coming to a shocking conclusion—a conclusion so important in its ramifications that it deserves to be set apart:
Intelligent people perform better only when you tell them what to do!
I am referring here specifically to the domain of rational thought and action. If you tell intelligent people what a rational requirement is—if you inform them about a particular stricture of rational thought (avoid intransitivity, avoid framing, do not be overconfident in your own knowledge, etc.)—and then give them a task that requires following the stricture, higher-IQ individuals will adhere to the stricture better than individuals of lower intelligence. However, if you give people tasks without warning them that a particular rational principle is involved—if they have to notice themselves that an issue of rationality is involved—individuals of higher intelligence do little better than their counterparts of lower intelligence.
It is true that there is a statistically significant relationship between intelligence and framing avoidance in within-subjects designs, but it is quite modest, leaving plenty of room for dysrationalia in this domain. It is likewise with some of the characteristics of the cognitive miser discussed in the last chapter—attribute substitution, vividness effects, failures of disjunctive reasoning. None of these characteristics are strongly correlated with intelligence.12 All of the characteristics discussed in this and the previous chapter are critical for achieving rationality of thought and action, yet none of these characteristics are assessed on intelligence tests. If they were, some people would be deemed more intelligent and some people less intelligent than they are now. Why? Because of the empirical evidence that I just mentioned—these processing characteristics show little relation to intelligence. That certainly holds true as well for one of the most defining features of the cognitive miser, to be discussed in the next chapter—myside processing.
EIGHT
Myside Processing: Heads I Win—Tails I Win Too!
If it’s at all feasible then your brain will interpret the question in a way that suits you best.
—Cordelia Fine, A Mind of Its Own, 2006
In a recent study my colleague Richard West and I presented one group of subjects with the following thought problem:
According to a comprehensive study by the U.S. Department of Transportation, a particular German car is 8 times more likely than a typical family car to kill occupants of another car in a crash. The U.S. Department of Transportation is considering recommending a ban on the sale of this German car.
Subjects then answered the following two questions on a scale indicating their level of agreement or disagreement: (1) Do you think that the United States should ban the sale of this car? (2) Do you think that this car should be allowed on U.S. streets, just like other cars? We found that there was considerable support for banning the car—78.4 percent of the sample thought that the German car should be banned and 73.7 percent thought that it should not be allowed on the streets like other cars.
The statistics on dangerousness of the car in the example happen to be real statistics, but they are not the statistics for a German car. They are actually the statistics for the Ford Explorer, which happens to be a very dangerous vehicle indeed, for the passengers of other cars.1 In the scenario just presented, subjects were evaluating the social policy of allowing a dangerous German vehicle on American streets. A second group of subjects in our study evaluated the reverse—the policy of allowing a dangerous American vehicle on German streets. This group of subjects received the following scenario:
According to a comprehensive study by the U.S. Department of Transportation, Ford Explorers are 8 times more likely than a typical family car to kill occupants of another car in a crash. The Department of Transportation in Germany is considering recommending a ban on the sale of the Ford Explorer in Germany. Do you think that Germany should ban the sale of the Ford Explorer? Do you think that the Ford Explorer should be allowed on German streets, just like other cars?
Subjects responded on the same scale, and when they did we found that 51.4 percent thought that the Ford Explorer should be banned and 39.2 percent thought that it should not be allowed on the German streets like other cars. Statistical tests confirmed that these percentages were significantly lower than the proportion of subjects who thought a similar German vehicle should be banned in the United States.
Our study illustrates what has been termed in the literature a myside bias. That is, people tend to evaluate a situation in terms of their own perspective. They judge evidence, they make moral judgments, and they evaluate others from a standpoint that is biased toward their own situation. In this case, they saw the dangerous vehicle as much more deserving of banning if it were a German vehicle in America than if it were an American vehicle in Germany.
Myside bias is a ubiquitous phenomenon, and it has been revealed in a variety of ingenious psychological studies. Drew Westen and colleagues have used an interesting task to study myside processing in contradiction detection.2 Subjects were asked to read materials which revealed a contradiction between a person’s words and actions. Some of the materials concerned political figures. For example, subjects read a statement by George W. Bush about Ken Lay, the CEO of Enron. The statement was made when Bush
was a candidate in 2000: “First of all, Ken Lay is a supporter of mine. I love the man. I got to know Ken Lay years ago, and he has given generously to my campaign. When I’m President, I plan to run the government like a CEO runs a country. Ken Lay and Enron are a model of how I’ll do that.” Subjects where then presented with a fact about Bush’s (then current) actions with respect to Lay. The fact was: “Mr. Bush now avoids any mention of Ken Lay and is critical of Enron when asked.” Subjects were then asked to consider whether the statement and action were inconsistent with each other on a 1 to 4 scale running from 1 (strongly disagree that the action and statement are inconsistent) to 4 (strongly agree that the action and statement are inconsistent).
There were other similar items about different political figures. For example, subjects were told: “During the 1996 campaign, John Kerry told a Boston Globe reporter that the Social Security system should be overhauled. He said Congress should consider raising the retirement age and means-testing benefits. ‘I know it’s going to be unpopular,’ he said. ‘But we have a generational responsibility to fix this problem.’” Subjects were then presented with a fact about Kerry’s actions that contradicted his statement: “This year, on Meet the Press, Kerry pledged that he will never tax or cut benefits to seniors or raise the age for eligibility for Social Security.” Subjects then answered on the same scale reporting whether or not they believed the action and earlier statement were inconsistent.
Myside bias in this contradiction detection paradigm was massive. Subjects’ political beliefs influenced whether they were able to detect the contradictions. For example, for Bush contradictions like the example given, self-identified Democrats gave a mean rating of roughly 3.79 (strongly agree that the statement and action are inconsistent). In contrast, self-identified Republicans gave Bush contradictions a mean rating of roughly 2.16 (disagree that the statement and action are inconsistent). Conversely, for Kerry contradictions like the example given, self-identified Republicans gave a mean rating of roughly 3.55 (strongly agree that the statement and action are inconsistent). In contrast, self-identified Democrats gave Kerry contradictions a mean rating of roughly 2.60 (neutral on whether the statement and action are inconsistent). In short, people could see the contradictions of the other party’s candidate but not the contradictions of their own.
People not only evaluate arguments in a biased manner, they generate arguments in a biased manner as well. My colleagues Maggie Toplak, Robyn Macpherson, and I had subjects explore arguments both for and against various public policy propositions. When the subjects were instructed to be balanced and unbiased, or when they did not have an extremely strong prior opinion on the issue (for example, “People should be allowed to sell their internal organs”), they generated arguments for both sides of the issue that were roughly equal in quality and quantity. But when subjects (university students) had a strong opinion on the issue (for instance, “Tuition should be raised to cover the full cost of a university education”), even when they were given explicit instructions to be unbiased in their reasoning, they generated many more arguments on their side of the issue than for the opposite position.
Myside processing undermines our ability to evaluate evidence as well as generate it. In several studies, Paul Klaczynski and colleagues presented subjects with flawed hypothetical experiments that led to conclusions that were either consistent or inconsistent with prior positions and opinions.3 The subjects ranged from young adults to elderly individuals. Subjects were then asked to critique the flaws in the experiments (which were most often badly flawed). Robust myside bias effects were observed—subjects found many more flaws when the experiment’s conclusions were inconsistent with their prior opinions than when the experiment’s conclusions were consistent with their prior opinions and beliefs.
We have known for some time that it is cognitively demanding to process information from the perspective of another person.4 It is thus not surprising that people are reluctant to engage in it, and that myside processing is a fundamental property of the cognitive miser. Nonetheless, we sometimes underestimate the costs of myside processing and/or fail to recognize it as the source of much irrational thought and action. Finally, as we shall see, intelligence is no inoculation against the perils of myside processing.
Overconfidence: On Thinking We Know What We Don’t
We will begin this section with a little test. For each of the following items, provide a low and high guess such that you are 90 percent sure the correct answer falls between the two. Write down your answers:
1. I am 90 percent confident that Martin Luther King’s age at the time of his death was somewhere between ___ years and ___ years.
2. I am 90 percent confident that the number of books in the Old Testament is between ___ books and ___ books.
3. I am 90 percent confident that the year in which Wolfgang Amadeus Mozart was born was between the year ____ and the year ____.
4. I am 90 percent confident that the gestation period (in days) of an Asian elephant is between ____ days and ____ days.
5. I am 90 percent confident that the deepest known point in the oceans is between ______ feet and ______ feet.
These questions relate to another important aspect of cognition in which people are myside processors. That domain of cognition concerns how people monitor confidence in their own beliefs. Psychologists have done numerous studies using the so-called knowledge calibration paradigm.5 In this paradigm, a large set of probability judgments of knowledge confidence are made. Of course, a single probability judgment by itself is impossible to evaluate. How would I know if you were correct in saying there is a 95 percent chance that your nephew will be married in a year? However, a large set of such judgments can be evaluated because, collectively, the set must adhere to certain statistical criteria.
For example, if the weather forecaster says there is a 90 percent chance of rain tomorrow and it is sunny and hot, there may be nothing wrong with that particular judgment. The weather forecaster might have processed all the information that was available and processed it correctly. It just happened to unexpectedly rain on that particular day. However, if you found out that on half of the days the weatherperson said there was a 90 percent chance of rain it did not rain, then you would be justified in seriously questioning the accuracy of weather reports from this outlet. You expect it to rain on 90 percent of the days that the weatherperson says have a 90 percent chance of rain. You accept that the weatherperson does not know on which 10 percent of the days it will not rain (otherwise she would have said she was 100 percent certain), but overall you expect that if, across the years, the weatherperson has predicted “90 percent chance of rain” on 50 different days, that on about 45 of them it will have rained.
The assessment of how well people calibrate their knowledge proceeds in exactly the same way as we evaluate the weatherperson. People answer multiple choice or true/false questions and, for each item, provide a confidence judgment indicating their subjective probability that their answer is correct. Epistemic rationality is apparent only when one-to-one calibration is achieved—that the set of items assigned a subjective probability of .70 should be answered correctly 70 percent of the time, that the set of items assigned a subjective probability of .80 should be answered correctly 80 percent of the time, etc. This is what is meant by good knowledge calibration. If such close calibration is not achieved, then a person is not epistemically rational because his or her beliefs do not map on to the world in an important way. Such epistemic miscalibration will make it impossible to choose the best actions to take.
The standard finding across a wide variety of knowledge calibration experiments has been one of overconfidence. Subjective probability estimates are consistently higher than the obtained percentage correct. So, for example, people tend to get 88 percent of the items correct on the set of items on which they say they are 100 percent sure that they were correct. When people say they are 90 percent sure they are correct, they actually get about 75 percent of the
items correct, and so on. Often, people will say they are 70 to 80 percent certain when in fact their performance is at chance—50 percent in a true/false paradigm.
The overconfidence effect in knowledge calibration is thought to derive at least in part from our tendency to fix on the first answer that comes to mind, to then assume “ownership” of that answer, and to cut mental costs by then privileging that answer as “our own” in subsequent thinking. Subjects make the first-occurring answer a focal hypothesis (akin to a myside bias) and then concentrate attention on the focal hypothesis, thereby leading to inattention to alternative, or nonfocal, answers. In short: “one reason for inappropriately high confidence is failure to think of reasons why one might be wrong” (Baron, 2000, p. 132). The evidence retrieved for each of the alternatives forms the basis for the confidence judgments, but the subject remains unaware that the recruitment of evidence was biased—that evidence was recruited only for the favored alternative. As a result, subjects end up with too much confidence in their answers.
What Intelligence Tests Miss Page 13