These lecture notes go over much of the material covered in Lecture Outline #11, but I have deliberately omitted some topics. As always, I won't ask any short answer questions that I didn't cover in these lecture notes.

Chapter 10 in Myers reviews the research on thinking, including material on errors and biases in reasoning. Errors and biases in reasoning is the focus of this lecture. Before we start, we need to go over some definitions.

The historical roots of psychological theories and research on thinking and reasoning are, as in most areas of psychology, to be found in philosophy. Philosophers have been interested in logic and reasoning for centuries, and that interest continues today. Epistemology is the branch of philosophy that focuses on how people acquire knowledge about the world. There are two major kinds of theories of reasoning: Descriptive theories vs. prescriptive (also known as normative) theories. Descriptive theories are those that describe what people actually do when they make decisions and judgments. In contrast, prescriptive (or normative) theories are those that describe how people should make decisions and judgments.

Since we are interested here in studying errors and biases in decisions and judgments, we must address the following question: How can we demonstrate that a particular inference or judgment is faulty?

Well--the answer is that we compare it to a normative standard--that is, a set of standards that we believe are truthful and capture the ideal way to make judgments. And if the judgment under consideration deviates from those standards, then we have greater confidence that the judgment may be flawed.

Expected Utility Theory (or EUT) is a simple example of a normative theory. EUT prescribes a way of determining what the best decision is (that is, a normative standard) under certain circumstances. The following are some very simple examples:

Suppose you are given a choice between two games:

Game A: You have a 20% chance of winning $10, and an 80% chance of winning nothing

Game B: You have a 10% chance of winning $50, and a 90% chance of winning nothing

OK--think about which game you would want to play. Game A is more appealing than Game B because you have a greater chance of winning (20% vs. 10%), but Game B is more appealing than Game A because if you win, you will win more money ($50 vs. $10). Thus, there are two separable dimensions of choice here: the likelihood (or probability) of an outcome (let's call it p, so that in Game A, p=20% and in Game B, p=10%) and the value (let's also call this "utility", although there are differences between value and utility, which I don't want to discuss here) of that outcome (let's call it v, so that in Game A, v=$10 and in Game B, v=$50). We clearly have to take into account both dimensions in making our decision, but how should we do this? How should these two dimensions be combined? According to EUT, we simply multiply them together for each game and then compare them:

Expected utility = (p) (v)

Game A: Expected utility = 20% x $10 = $2

Game B: Expected utility = 10% x $50 = $5

In other words, on average, our expected utility (that is, our expected winnings per game) will be $2 for Game A and $5 for Game B. So if you play Game A 100 times, on average you will win $200 at the end of the 100 games, and if you play Game B 100 times, on average you will win $500. And because the expected utility for Game B is greater than the expected utility for Game A, you should choose to play Game B. In other words, the normative decision is to go for Game B. And that's exactly what people tend to do when they are given choices like these, in which the probabilities are fairly similar. In other words, people are making normative decisions: they are indeed reasoning in accordance with the normative model--they're doing what they should be doing.

But that's not always the case. Consider a choice between these two games:

Game C: You have a 1% chance of winning $1,000 and a 99% chance of winning nothing

Game D: You have a 100% chance of winning $5 (in other words, I will just give you $5).

Let's work out the expected utilities for these two games:

Game C: Expected utility = 1% x $1,000 = $10

Game D: Expected utility = 100% x $5 = $5

According to expected utility theory, people should choose Game C. But that's not what happens in real life. You get a large proportion of people who choose Game D. Why? Because Game D involves a sure thing. You will win something, guaranteed. Although your expected winnings will be lower, you will win something. This is a very simple demonstration of when people violate the normative standard implied by expected utility theory, a violation that demonstrates that people are generally risk-averse. Game D is a sure thing, whereas Game C is "risky" in that it involves uncertainty: you have a 1% chance of winning, but a 99% chance of losing. People don't like uncertainty, so they will tend to choose the sure thing, even if it ends up costing them money. In short, EUT gives us a way of comparing people's actual decisions and judgments to a normative standard (what people should be doing).

The games that I have presented above are extremely simple because they involve only one dimension (money) and they involve only winning. We can easily expand EUT to incorporate multi-dimensional choices that have costs as well as benefits. Thus, EUT has great applicability for understanding judgments and decisions in everyday life.

Now that you have been introduced to the logic of EUT, on the next page, I present a much more intriguing example of a very disturbing phenomenon in decision making that involves EUT.

Imagine that Canada is preparing for an outbreak of an unusual Asian disease that is expected to kill 600 people. Two alternative programs to combat the disease have been proposed. Scientific estimates of the consequences of the programs are as follows:

If Program A is adopted, 200 people will be saved.
If Program B is adopted, there is a 1/3 probability that 600 people will be saved and a 2/3 probability that none of the 600 people will be saved.

Which program would you favor?

Kahneman and Tversky found that 72% of subjects favored Program A (and 28% chose B)

But now consider the following version of the same problem:

If Program C is adopted, 400 people will die.
If Program D is adopted, there is a 1/3 probability that nobody will die and a 2/3 probability that 600 will die.

Kahneman and Tversky found that 22% of subjects favored Program C (and 78% chose D)

But wait a second--Program A is identical to Program C (out of 600 people, saving 200 people is obviously the same as having 400 people die!) and Program B is identical to Program D. So why should 72% of subjects choose Program A, but only 22% of subjects choose Program C?

This is known as the framing effect. It turns out that Program A and B are framed in terms of gains (lives saved), whereas Programs C and D are framed in terms of losses (lives lost). Even though the numbers are identical, people are risk-averse when thinking about gains and risk-seeking when thinking about losses.[1]

Think about this for a moment: isn't it disturbing that people can make decisions and judgments about exactly the same situation in very different ways, depending on how the situation is framed?

And this framing effect doesn't just occur in hypothetical situations like the above. It's also true of real-life physicians making realistic decisions about whether to operate or use radiation therapy in cancer cases. In one study (McNeil et al., 1982), physicians changed their recommendations dramatically depending on whether the statistics were framed in terms of gains or in terms of losses.

Why is it important to study inferential errors? There are two reasons. To explain the first, let me make an analogy to the study of human vision. Our understanding of vision has been greatly influenced by the study of visual illusions. Through the study of visual illusions, we have learned much about how the visual system processes visual information.

In much the same way, we can learn much about how we process information about the world--how we make judgments and inferences about events or behavior--by studying errors and biases.

The second reason why the study of errors and biases is important is because despite the belief that most of us have that we make perfectly reasonable decisions and judgments that almost always turn out well, it turns out that we are sometimes very overconfident in our judgments--that is, our confidence often far outstrips our actual ability. Indeed in a study by Fong and Klein (1995), we found that people believed themselves to be far superior in their ability to make social judgments than the average person. For example, 2/3 of our subjects believed that they were above the 50th percentile in their ability to make accurate first impressions, and 85% of them believed that they were above the 50th percentile in being a good judge of a person's character and in detecting changes in a person's mood.

It should be clear from this result that there's something funny going on. By definition, it is impossible for 85% of people to be above average. This result suggests strongly that people seem to have a superiority bias in their perceived ability to make social judgments. And this may help explain in part why people continue to believe strongly in the validity of first impressions, especially their own.

The point is that if we believe that we're so terrific in our judgmental ability, we may fail to recognize those instances when we are subject to error and bias. But by being confronted by our own inabilities, through studying the nature of errors and biases in reasoning, this can help us be more vigilant in avoiding common sources of error in our future judgments. Forewarned is forearmed.

OK--suppose we make errors. What is the best explanation for their occurrence--why do we make them? Well, there are two contrasting (although not necessarily contradictory) explanations: the classical/philosophical explanation and the cognitive explanation.

The ancient philosophers--for example, Plato and Aristotle--had a hierarchical conception of the mind. The believed that the mind consisted of three layers. On top was the rational mind, in the middle was the animal instinctive mind, and the lowest level was the vegetative mind. How did errors occur? The answer was simple: they were due to "intrusions from below" that is, the lower, baser instincts intruded upon the rational mind. For example, Aristotle explained cowardice in battle by suggesting that the more animal instinct of self-preservation intruded upon and superseded the more rational mind, which supposedly was intent on fighting for one's homeland regardless of consequences.

You may quibble about whether fighting in the face of likely death is rational, but the principle that errors are due to lower instincts is one that is very common even today. Freud was heavily influenced by the idea of hierarchies. His psychoanalytic theory of the mind suggested that unconscious processes sometimes intruded upon the rational processes of the ego and caused irrational behavior and thoughts. Without necessarily referring to Freud, all of us have at one point or another explained our judgments or behavior by suggesting that our emotions got the better of our logical mind. Indeed, more modern versions of the classical explanation for error suggests that motivational, affective forces are implicated.

There is a very different alternative explanation for human inferential error--the cognitive perspective. According to this perspective, humans are limited in their ability to process information--that is we are limited capacity information processors. At this very moment, for example, there are literally thousands of stimuli that are impinging on you besides my voice: the light in the room you're in, sounds from the outside, the sensation of the chair on your back, and that irresistible urge you have to swallow right now (did it work on you?)

The point is that it is impossible for us to give equal attention to all the stimuli; indeed that would be extraordinarily inefficient. We must therefore simplify the complex world--pare it down so that we can handle it.

How do we simplify the complex world? There are several ways. First, we have selective attention--we selectively attend to important stimuli and tend to ignore what seems to be unimportant. Second, we rely on schemas (also known as schemata). A schema is a generalization, a summary, of information we possess about a given object or person. So if we were to make to think about Stephen Harper, for example, we would refer to our schema of him rather than call to mind every single bit of information about Harper that we possess. And third, when we make judgments or decisions, we utilize simpler heuristics--rather than employing complex mathematical decision rules, we use simpler rules such as the representativeness heuristic, which states that things that are similar to each other on some dimension are probably similar to each other on all dimensions, at least as a first approximation. Another heuristic is the availability heuristic, which suggests that we will tend to utilize information in a decision or judgment to the extent that it is easily called to mind, that is, available. I will have more to say about heuristics later.

In the cognitive perspective, the errors that we sometimes make are simply a consequence, a byproduct, of our need to simplify the world. Although these schemas and heuristics may serve us well most of the time, there are other times that they lead us to errors, which can have serious consequences.

As I mentioned earlier, the two explanations for error are not necessarily incompatible. Indeed, recent theoretical models of error incorporate both aspects.

A few years ago, on the ABC program 20/20, there was a report on the effects of physical attractiveness on job selection, jury decision making, etc. The basic conclusion, which repeats what social psychologists have known for 20+ years, is that people who are more physically attractive are more likely to get jobs, be helped in an emergency, be less likely to be found guilty of a crime, and even when found guilty, are given lighter sentences. [By now, you should be able to figure out how to design a study that would test this hypothesis, right?] The reporter interviews two employers who have offered a job to a very attractive applicant (let's call her Tracey) rather than to a less attractive applicant (let's call her Stacey), whose credentials were very similar/identical. The reporter, sensing the kill (since these two guys were the only ones who agreed to be interviewed), says, "Why did you hire Tracey and not Stacey?" The employers say, "it's because Tracey had a great phone voice, and we need that kind of voice for our company." The reporter says, "Are you sure that it wasn't because Tracey is much more attractive?" And the employers deny this.

What happened? Were the employers lying about the true reasons for their decision? Not necessarily. It turns out that people themselves may not know the "true" reasons for their actions of judgments.

Consider a very famous set of experiments by Nisbett and Wilson (1977). In one study, an experimenter conducted a "consumer study" in a shopping mall. He laid out 4 pairs of panty hose and asked consumers to pick the pair that they liked the best. In reality, all 4 pairs were identical. But it turns out that consumers were significantly more likely to select the right-most pair (of course, the pairs were switched around from subject to subject so that the same pair didn't always end up in the right-most position). For some reason, people seem to show a strong preference for the right-most object [to address a question that crops up when I describe this phenomenon, I don't know whether people in Israel, where reading goes from right to left, show a left position preference].

When the experimenter asked subjects why they chose the right-most pair, they mentioned something about its sheerness, or its strength, or some other physical characteristic of the pantyhose. None of them said that they were influenced by the positioning of the pantyhose, despite the fact that it must have been a causal factor. In fact, when the experimenter explicitly suggested that the position might have been a contributing factor, the subjects looked at him as if there was something desperately amiss. Thus, people in the pantyhose experiment failed to recognize the presence of a real cause of their behavior/judgment.

In other study, subjects watched a videotape of a man with a mustache and Belgian accent giving a lecture. There were two versions of the videotape, and subjects were randomly assigned to watch either the friendly lecture or the hostile lecture. When subjects later rated the lecturer on friendliness, it was indeed the case that those who had watched the friendly lecture rated that lecturer as being more likable, friendly than did those who had watched the same lecturer give the hostile lecture. So far, no problem. All subjects were also asked to rate how attractive the lecturer's mustache was, and how likable the lecturer's Belgian accent was. It turned out that subjects in the friendly condition rated those two features more positively. This is an example of the halo effect--when one dimension of a person is positive, you tend to think that other dimensions are positive as well (thus, the physical attractiveness stereotype--attractive people are also believed to be nice, intelligent, etc.). I'm sure that you have experienced this phenomenon yourself. But although the halo effect is somewhat problematic, that's not what I want to focus on here. What I do want to describe is the following finding: when people were asked why they liked or disliked the lecturer, they thought that it was because of the mustache and accent that they liked/disliked the lecturer. In other words, they thought that:

liking or disliking for the lecturer's mustache and accent----->liking or disliking for lecturer

But there is something wrong here because in the experiment, the experimental manipulation (the independent variable) was whether the lecturer was likable or dislikable (if that's a word!). Subjects were randomly assigned to conditions. Thus, it really was the case that:

liking or disliking for lecturer----->liking or disliking for the lecturer's mustache and accent

In other words, to use the terminology I introduced in my lecture on experimental design, they thought that Y caused X, but it was really X that caused Y. In short--they screwed up the actual causal direction.

Thus, to summarize Nisbett and Wilson's point so far, people sometimes fail to recognize the presence of a contributing cause to their judgments/behaviors (as in the panty hose study), and sometimes get the cause-effect direction reversed (as in the lecturer study).

So what is going on when you ask people why they did something? What are they reporting to you? Nisbett and Wilson explain this as follows: often, you just don't have any idea why you did what you did or why you thought what you thought. So when someone asks you the "why" question, you come up with a plausible explanation, based not on the true reasons, but rather on the reasons why you might have done so. Where does this come from? Nisbett and Wilson say that we have a priori causal theories that govern how we try to construct what might have been responsible for our judgments or behaviors. These a priori causal theories are based on plausibility, rather than on any true ability to introspect and ascertain the true reasons.

This may be pretty abstract so far, so let me take you through a concrete example. Let's go back to the pantyhose study. Let's say that you have chosen the rightmost pair as being the best of the four pairs. Then the experimenter asks you, "why did you choose that one?" What you do at that moment is to answer the question, "why would I (or anyone else) choose a pair of pantyhose?" You rattle off those factors that should be related to pantyhose selection (e.g., sheerness, sheen, and other physical characteristics). In other words, you possess causal theories for why people select consumer goods. Position on the table is simply not one of those causal theories. You may not be aware that this is what you're doing, because if you come up with a plausible explanation for your own behavior, you may say to yourself, "hey, this sounds pretty good. It must be true." [Have you ever been asked the why question, and then begin to give an answer, fumbling around first, but then coming up with a coherent story, which gives you greater confidence that the story is indeed the correct one?].

Getting back to the 20/20 interview, when the reporter asked the two employers whether physical attractiveness was a factor in hiring, and they denied it, they were not deliberately misleading the reporter. They simply may not have known how much they were influenced by the physical attractiveness of the applicants.

Bottom line: sometimes the reasons we give for our actions or decisions may not be veridical, and reflect our theories for why people generally perform those actions or make those decisions rather than reflecting any true insight into our mental processes. In other words, as Nisbett and Wilson put it, we sometimes "tell more than we can know."

According to the cognitive perspective, we make inferential errors because of our limited capacity to process information. We instead use heuristics, or simpler, rules-of-thumb when we make everyday decisions.

Availability Heuristic

Whatever is more available in memory will be more likely to be used in decisions and judgments. What makes something more available in memory? Vividness is one factor. Consider the following thought experiment:

It's time to buy a new car, and you've decided to put more money into one of those good built-to-last Swedish cars, either a Volvo or a Saab. They're both about the same price (pretty high!), and you like the styling of each about the same. So what matters is the repair records. So you go to Consumer Reports to look up the repair records of Volvos and Saabs. As you (may) know, Consumer Reports gathers repair record information from car owners. It turns out on the basis of reports from hundreds of Volvo owners and hundreds of Saab owners that Volvo has a slightly better repair record. So you're thinking about buying a Volvo. Then you go to a party, and mention to your brother-in-law of your intentions. He explodes. "What? I once had a Volvo and let me tell you what happened. First, that fancy fuel injection system went out. 2,000 bucks. Then there were problems with the chassis and the brakes. I finally had to sell it for junk.

What do you do? Do you buy the Volvo or the Saab? Well, whatever your actual decision, this little story certainly troubles you. Why? Because the story is so vivid (you can actually see your beloved Volvo falling apart in front of you). But think about it this way: no matter how vivid the story, your brother-in-law's story is just one car. As compared with hundreds of cars from Consumer Reports. Even if you decide to stick with the Volvo, the story will surely be given more weight than it is worth.

Let's for a moment analyze the Volvo-Saab problem with regard to statistical principles. The base rate data from Consumer Reports is based on hundreds of people. The sample size is in the hundreds. How much more information does your brother-in-law give you? Well--no matter how you slice it, it's still only a single car. This anecdote suggests that we are much more likely to be swayed by vivid and graphic single cases rather than by pallid and dry statistics. The vividness effect refers to the greater influence of vivid information on our decisions and judgments. We pay more attention to vivid information--Stalin said that "a single death is a tragedy. A million deaths is a statistic."

I hear you protesting. Isn't the information from your brother-in-law more valuable? From a trusted source? You know your brother-in-law, but you don't know any of those other anonymous respondents. Without getting into addressing all of these issues (e.g., why would you believe that any of those respondents would somehow be misreporting their experiences? and why would you think that there would be bias, e.g., that the Volvo owners had a greater likelihood of casting a more favorable light on their cars than would the Saab owners?), let's just say that there is the possibility that it is vividness that is among the factors that will influence your decision (and again, we never conclude that X is the only causal factor, or even the most important causal factor).

So how do we conduct an experiment to determine whether vividness alone has an effect on decisions and judgments, which would be evidence for the role of the availability heuristic?

An experiment by Reyes, Thompson, and Bower (1981) demonstrates the vividness effect. Subjects were asked to play the role of jurors in a vehicular manslaughter case. The critical point in the case was whether the defendant was drunk at the time that he left a party. There were two conditions, and as always, subjects were randomly assigned to conditions. In the pallid condition, the facts of the case were described in a pallid fashion. For example, at one point during the party, the defendant knocked a bowl of dip onto the carpet, which may have been an indication of intoxication. Then the defendant went out to his car and drove away. But in the vivid condition, the defendant knocked a bowl of guacamole dip onto the white shag carpet. And he drove away in a bright orange VW bug.

Subjects were asked to give their ratings of the defendant's guilt, either immediately after reading the story or 2 days later. The neat aspect of this study design is that the difference in description between the pallid and vivid conditions should not affect subjects' guilt ratings. After all, the only thing that matters in this case is whether the defendant knocked the bowl of dip over. It doesn't matter whether it was guacamole onto a white shag carpet. The mess that he made should be irrelevant to the judgment of the defendant's guilt (of course, if this was a case about carpet cleaning, then the amount of ugly mess would be relevant!).

OK--here are the results: for those subjects who were asked to render their judgments immediately after reading the case, there was no difference between the pallid condition and vivid condition in guilt ratings. But, for those subjects who rendered their judgments two days after reading the case, the subjects in the vivid condition were significantly higher in their ratings of the defendant's guilt. Why did this happen? According to the availability heuristic (Kahneman and Tversky), information that is more easily called to mind, that is, information that is available in memory, is more likely to be used in subsequent decisions and judgments. Immediately after reading the case, there was no difference between the two conditions because all of the facts were still fresh in their minds; in other words, they were available. But two days after, subjects have forgotten much of the detail of the case, but the vividness of the guacamole dip and the bright orange VW help subjects in the vivid condition remember those facts that are indicative of guilt, and hence those facts are available in memory for use in the judgment, just as Kahneman and Tversky would predict. That's how you get vividness effects.

Vividness effects are enhanced by the media. We get most of our ideas about the likelihood of certain events--crime, winning the lottery, plane crashes, based entirely on media reports. Think about the proportion of crimes that are committed by "ex-mental patients." The fears that people have about ex-mental patients stem entirely from media reports. But there is a very strong reporting bias. After all, how many times do you read about "James Fisher, 37, who has never been a mental patient, was convicted today of a heinous crime..."? Recall the work on perceptual salience--whatever attracts our attention will be considered to a greater extent in the attribution process. Thus, social categories--race, social class, mental status--will tend to be mentioned in media reports because reporters, just like the rest of us, believe that such information may be of causal relevance.

And in the Fall of 1991, Kimberly Bergalis, an AIDS patient, testified at a U.S. congressional committee in support of a bill that would require workers in health care settings--physicians, nurses, technicians--to submit to mandatory testing for HIV, the AIDS virus. Why? Because she had contracted the AIDS virus during oral surgery from an AIDS-infected dentist. Now--the baserate of contracting AIDS from a health care worker is extremely low--in fact, at the time of this writing, there have only been a handful of cases out of many millions of contact opportunities in which AIDS was transmitted from an infected health care worker on the job. Yet, the powerful, vivid image of Kimberly Bergalis testifying before Congress, her body ravaged by AIDS, undoubtedly affected the public to a much greater extent than the base rates would.

By the way, over the years teaching this introductory course, again and again I encounter the vividness effect among undergraduates. Third- and fourth-year students will occasionally see me on campus and recall something about my course. Inevitably, they remember two events: kicking the Bobo doll and the Jeopardy! tape. But then I ask them: "do you remember what the point of the Bobo Doll study was?" "Why did I show the Jeopardy! tape?" Often, they draw a blank.

As a lecturer and teacher, I am concerned with teaching students about the science of psychology and the wonderful and clever ways that psychologists have systematically identified truths about psychological phenomena. But I must admit the first few times my disappointment (in my own failure, rather than in my students' failure) that students don't remember the reason behind the vivid examples I use in lectures.

Representativeness Heuristic

The second important heuristic that Kahneman and Tversky identified is called the representativeness heuristic. We tend to judge how likely something is by judging how similar it is to a prototype, rather than to other information that directly bears on its likelihood. Myers discusses the representativeness heuristic in the book (on pp. 401-02), so I won't go into any detail here about this important heuristic.

Much research has demonstrated that having a theory or hypothesis--that is, a preconceived notion--leads to a set of cognitive processes that lead an individual to tend to confirm the hypothesis, even when there is evidence against that hypothesis.

Here's an example. In a study by Snyder and Swann (1978), subjects were told that they were about to meet a person and talk to them for awhile. Their goal would be to find out about the person. Some subjects were told that the person might be an extravert. Other subjects were told that the person might be an introvert. Subjects then selected questions from a list to ask the person. The list contained extraverted questions (For example, "What would you do to liven up things at a party?"), introverted questions (For example, "What is it about large groups that make you feel uncomfortable?"), and neutral questions (for example, "What kinds of charities do you contribute to?"). The results indicated a very strong confirmatory bias--subjects who were told that the person might be an extravert were more likely to select more extravert questions than were subjects told that the person might be an introvert. The opposite was true for the introvert questions. So even before you have met a person, if you have some hypothesis about what he or she might be like, you will tend to seek information about them that is consistent with your hypothesis.

Where do these hypotheses come from? Sometimes you are given them beforehand (think about blind dates--and the likelihood (probably close to 100%) that the person who sets you up with your blind date has given you all kinds of hypotheses about him/her: "he's really outgoing"; "she's fun to be with"; "she's smart", etc.). But suppose no one has provided you with any hypotheses at all. In a study by Fong and Markus (1982), we pre-selected individuals who were extraverts and those who were introverts. We then gave them the Snyder and Swann procedure. It turns out that extraverts were more likely than introverts to selected extraverted questions, and the introverts were more likely than extraverts to select introverted questions. Thus, the Fong and Markus study demonstrates that, even in the absence of any hypotheses, your own self-concept generates hypotheses about what people might be like: you therefore want to seek information about other people on dimensions that are relevant to your own self-concept.

Snyder and Cantor (1979) had subjects read one week's events in the life of Jane. The story was constructed such that it contained equal numbers of extroverted and introverted behaviors. For example, Jane animatedly conversed with another patient in the doctor's office was one of the extroverted behaviors. And Jane spent her office coffee break by herself was one of the introverted behaviors.

Two days later the subjects tried to remember the behaviors that were relevant for a job for which she was being considered. Those who evaluated her for a job as "research librarian" recalled many more introverted behaviors than extroverted behaviors; those who evaluated her for a job as "real estate salesperson" recalled many more extroverted behaviors than introverted behaviors. Now remember--all subjects read exactly the same story, containing exactly the same number of extroverted and introverted behaviors.

Subjects then rated Jane on her suitability for both jobs. Those who had initially evaluated Jane for the "research librarian" job rated her as much more suitable for that position than for "real estate salesperson." And the opposite was true for those who had initially evaluated Jane for the "real estate salesperson."

So what happened? The Snyder and Cantor study demonstrates that when we test hypotheses, we engage in a confirmatory search--we remember and cite evidence that supports the hypothesis: that's what the memory measures revealed. And then when we make a conclusion, a summary judgment about our hypothesis, we base them on a biased set of information--biased in the direction of our hypothesis.

Other studies have shown that when people test hypotheses and are presented information that is ambiguous--that is, it could either be favorable or unfavorable to the hypothesis or theory, they tend to interpret this ambiguity in favor of the hypothesis.

Finally, let's consider what happens when a theory is completely discounted. Ross, Lepper, and Greene (1975) conducted a study in which subjects were given a "social sensitivity test." They were given 15 hand-written notes and were told that some of them were actual suicide notes, and some were fake suicide notes. They judged which were which.

Subjects were randomly assigned to two conditions. In the success condition, they were given feedback that they had gotten 13 out of 15 correct. In the failure condition, they were given feedback that they had gotten only 7 out of 15 correct--basically at chance levels.

Next, all subjects were told that this was all an experiment. The experimenter told subjects that they had been randomly assigned to conditions, and that no one knew about their real ability to discern real vs. fake suicide notes. Finally, subjects were asked to predict how well they really did on the task--how many out of 15 would they get correct?

Now--the theory that people held about themselves after doing the task--that they were either good at the task or bad at the task, was completely discounted. So there should have been no difference in predicted score in the two conditions. Yet there was a strong effect. This is known as the perseverance effect: even after a theory has been completely discounted, it still affects our beliefs--it perseveres. Why does this happen? Research suggests that when we conjure up explanations that explain the theory (that is, confirmatory search processes), these explanations take on a life of their own after the theory has been discounted. When subjects are asked to think about and explain why an alternative theory might be true, perseverance effects go away.

A stunning example of perseverance comes from Freud's writings. Freud wrote a classic psychological biography of Leonardo DaVinci. In that book, Freud believed that one of Leonardo's dreams, which Freud read from a translation of Leonardo's journals, held the key to understanding Leonardo's psychological make-up, and his purported homosexuality. In that dream, Leonardo was a baby in his crib. Then a vulture flew down and touched his lips with its tail. Well, that's all that Freud needed to make a provocative speculation that Leonardo was really dreaming about his mother touching him on the lips. That's because, vulture is an Egyptian symbol for mother (remember manifest vs. latent content of a dream). But there was one really big problem with this story: Freud read about Leonardo's dream from a German translation. And there was a rather significant translation error: it turned out that Leonardo did not dream that a vulture touched his lips; he dreamed that a kite (the bird, that is, not the flying toy) had touched his lips.

All of a sudden, none of Freud's brilliant analysis makes any sense, because the premise--the whole starting point for the analysis--is totally discounted. But did Freud's publisher care? Not one bit. He ignored this, indicating that although the translation did pose problems for Freud, the analysis was so brilliant anyway that it was worthy of publication.

So if people are really bad at decisions and judgments, what can we do to improve our reasoning? First, I think it's important to simply realize that, from the literature on errors and biases, that we are worse at decision making than we think we are, and to admit that our judgments could be improved. Second, the results of the perseverance effect provide us with an important way to reduce the effects of theory-driven processing: think about ways that the opposite could be true. Such a strategy may help us avoid confirmatory biases that so often lead to a bolstering of a theory we hold, even in the face of ambiguous or even contradictory evidence.

Since a number of errors can be categorized as failures to reason in accordance with simple statistical principles like the law of large numbers (the more evidence you have, that is, the larger the sample size, the greater your confidence in your judgments), we may be able to train people to understand how statistical concepts like sample size and variability are just as relevant to making decisions and judgments in everyday life as they are in making judgments about research. As series of studies by Fong, Krantz, and Nisbett (1986) indicates that formal training in statistics, whether in the classroom or in a 25-minute laboratory training program, serves to enhance the likelihood that individuals will reason in accordance with the law of large numbers, thereby setting the stage for reducing some of the errors that I have discussed in this lecture.

And it is important that we endeavor to find ways to reduce inferential error. In the domain of confirmatory biases alone, such biases can lead to profound consequences. In a famous study by Rosenhan (see page 645 in Myers), individuals faked their way into a mental institution. Although being able to fake your way into a mental institution might seem to be a big problem, it's really not; after all, how many normal individuals (other than those who are accused of a crime) have any incentive at all to fake a mental illness?

But here's the really amazing and disturbing aspect of the Rosenhan study. Upon being admitted, these "pseudopatients" began acting completely normally. But these perfectly normal behaviors were interpreted by staff members as being symptoms of their supposed illness. For example, when Rosenhan, who was one of the pseudopatients, was taking notes on his experiences in the mental hospital, a nurse who saw him taking notes wrote on his chart, "patient engages in writing behavior." That phrasing certainly makes it sound abnormal. Another example: Rosenhan was walking in the hallway of the hospital one day and a psychiatrist asked him, "are you feeling anxious?" (sounds like a Snyder and Swann prediction!).

Another study by Rosenthal and Jacobson demonstrates the self-fulfilling prophecy--that the ultimate effect of confirmatory biases may be to actually make it so. In their study, students who had been identified (randomly) as "late bloomers" to their teachers scored significantly higher on intelligence tests at the end of the year than their fellow students. This study demonstrates the self-fulfilling prophecy--that the ultimate effect of confirmatory biases may be to actually make it so.

Thus, errors and biases are not just limited to the domain of laboratory experiments. They can have consequences for the decisions and judgments we make in everyday life, and ultimately they can affect not only our behavior but those of others.

In closing, here is a quote by Thomas Jefferson: "The wise know too well their weakness to assume infallibility; and he who knows most, knows best how little he knows."