optionaSanchezDunning-LearningaLittleAboutSomethingMakesUsOverconfident.pdf

    PSYCHOLOGY

    Research: Learning a Little AboutSomething Makes UsOvercondentby Carmen Sanchez and David Dunning

    MARCH 29, 2018

    HBR STAFF/TOM KELLEY ARCHIVE/GETTY IMAGES

    As former baseball pitcher Vernon Law once put it, experience is a hard teacher because it gives the

    test first, and only then provides the lesson.

    Perhaps this observation can explain the results of a survey sponsored by the Association of

    American Colleges & Universities. Among college students, 64% said they were well prepared to

    work in a team, 66% thought they had adequate critical thinking skills, and 65% said they were

    proficient in written communication. However, among employers who had recently hired college

    students, less than 40% agreed with any of those statements. The students thought they were much

    further along in the learning curve toward workplace success than their future employers did.

    Overcondence Among Beginners

    Our research focuses on overconfidence as people tackle new challenges and learn. To be a beginner

    is to be susceptible to undue optimism and confidence. Our work is devoted to exploring the exact

    shape and timeline of that overconfidence.

    One common theory is that beginners start off overconfident. They start a new task or job as

    “unconscious incompetents,” not knowing what they don’t know. Their inevitable early mistakes

    and miscues prompt them to become conscious of their shortcomings.

    Our work, however, suggests the opposite. Absolute beginners can be perfectly conscious and

    cautious about what they don’t know; the unconscious incompetence is instead something they

    grow into. A little experience replaces their caution with a false sense of competence.

    Specifically, our research focused on the common task of probabilistic learning in which people

    learn to read cues from the environment to predict some outcome. For example, people must rely on

    multiple signals from the environment to predict which company’s stock will rise, which applicant

    will do the best job, or which illness a patient is suffering from. These can be hard tasks — and even

    the most expert of experts will at times make the wrong prediction — but a decision is often essential

    in many settings.

    In a laboratory study, we asked participants to imagine they were medical residents in a post-

    apocalyptic world that has been overrun by zombies. (We were confident that this would be a new

    scenario to all our participants, allowing them all to start as total novices.) Their job, over 60

    repeated trials, was to review the symptoms of a patient, such as whether the patient had glossy

    eyes, an abscess, or brain inflammation, and diagnose whether the patient was healthy or infected

    with one of two zombie diseases. Participants needed to learn, by trial and error, which symptoms to

    rely on to identify zombie infections. Much as in a real-world medical diagnosis of a (non-zombie)

    condition, the symptoms were informative but fallible clues. There were certain symptoms that

    made one diagnosis more likely, but those symptoms were not always present. Other potential

    symptoms were simple red herrings. Participants diagnosed patients one at a time, receiving

    feedback after every diagnosis.

    The Beginner’s Bubble

    We found that people slowly and gradually learned how to perform this task, though they found it

    quite challenging. Their performance incrementally improved with each patient.

    Confidence, however, took quite a different journey. In each study, participants started out well-

    calibrated about how accurate their diagnoses would prove to be. They began thinking they were

    right 50% of the time, when their actual accuracy rate was 55%. However, after just a few patients,

    their confidence began skyrocketing, far ahead of any accuracy they achieved. Soon, participants

    estimated their accuracy rate was 73% when it had not hit even 60%.

     

    It appears that Alexander Pope was right when he

    said that a little learning is a dangerous thing. In our

    studies, just a little learning was enough to make

    participants feel they had learned the task. After a

    few tries, they were as confident in their judgments

    as they were ever going to be throughout the entire

    experiment. They had, as we termed it, entered into

    a “beginner’s bubble” of overconfidence.

    What produced this quick inflation of confidence? In

    a follow-up study, we found that it arose because

    participants far too exuberantly formed quick, self-

    assured ideas about how to approach the medical

    diagnosis task based on only the slimmest amount of

    data. Small bits of data, however, are often filled

    with noise and misleading signs. It usually takes a

    large amount of data to strip away the chaos of the

    world, to finally see the worthwhile signal. However,

    classic research has shown that people do not have a

    feel for this fact. They assume that every small

    sequence of data represents the world just as well as long sequences do.

    But our studies suggested that people do eventually learn — somewhat. After participants formed

    their bubble, their overconfidence often leveled off and slightly declined. People soon learned that

    they had to correct their initial, frequently misguided theories, and they did. But after a correction

    phase, confidence began to rise again, with accuracy never rising enough to meet it. It is important

    to note that although we did not predict the second peak in confidence, it consistently appeared

    throughout all of our studies.

     

    A Real-World Bubble

    The real world follows this pattern. Other research

    has found that doctors learning to do spinal surgery

    usually do not begin to make mistakes until their

    15th iteration of the surgery. Similarly, beginning

    pilots produce few accidents — but then their

    accident rate begins to rise until it peaks at about

    800 flight hours, where it begins to drop again.

    We also found signs of the beginner’s bubble outside

    of the laboratory. As with probabilistic learning, it

    has been shown that most people under the age of

    18 have little knowledge of personal finance. Most

    primary and secondary educational systems do not

    teach financial literacy. As such, personal finance is

    something most learn by trial and error.

    We found echoes of our laboratory results across the

    life span in surveys on financial capability

    conducted by the Financial Industry Regulatory

    Authority. Each survey comprised a nationally

    representative sample of 25,000 respondents who

    took a brief financial literacy test and reported how knowledgeable about personal finance they

    believed they were. Much like in the laboratory, both surveys showed that real financial literacy

    arose slowly, incrementally, and uniformly across age groups.

    Self-confidence, however, surged between late adolescence and young adulthood, then leveled off

    among older respondents until late adulthood, where it began to rise again — a result perfectly

    consistent with our laboratory pattern.

    It is important to note that our work has several limitations. In our experiments, participants

    received perfect feedback after each trial. In life, consistent feedback like this is often unavailable.

    Also, our tasks traced how confidence changed as people learned truly novel tasks. There are plenty

    of tasks people learn in which they can apply previous knowledge to the new task. We do not know

    how confidence would change in these situations. Relatedly, we cannot be certain what would

    happen to overconfidence after the 60th trial.

    With that said, our studies suggest that the work of a beginner might be doubly hard. Of course, the

    beginner must struggle to learn — but the beginner must also guard against an illusion they have

    learned too quickly. Perhaps Alexander Pope suggested the best remedy for this beginner’s bubble

    when he said that if a few shallow draughts of experience intoxicate the brain, the only cure was to

    continue drinking until we are sober again.

    Carmen Sanchez is a PhD candidate in Social and Personality Psychology at Cornell University. She studies howperceptions of abilities change as people learn, cultural differences in self-enhancement, and nancial decision-making.

    David Dunning is a Professor of Psychology at the University of Michigan. His research focuses on the psychology ofhuman misbelief, particularly false beliefs people hold about themselves.

                                                                                                                                      Order Now