Judgement and Decision Making Exam 3

"Dialectical bootstrapping"

"The wisdom of the crowd" -Under the right conditions, the average of many people's estimates is usually better than any one person's estimate -Different people rely on different knowledge. -People's errors tend to cancel out. -Francis Galton's example: Guessing the weight of a slaughtered and dressed ox at a country fair But what if there is no crowd? -Can one person benefit from making a second estimate? -Is there a smart way to generate the second estimate? -Maybe a strategy like "consider the opposite" would help.

calibration

how accurately do we make probability estimates For continuous judgments, correspondence is often called "calibration."

Decision affect theory

is a theory of anticipated and experienced emotion that incorporates versions of regret theory and disappointment theory Rather than maximizing expected value or expected utility, people make choices to maximize the expectation of anticipated emotions.

The normative model (MAUT)

is an exhaustive, compensatory model designed to choose the best (optimal) option

People tend to overestimate ____ probability events and underestimate ___ probability events.

low, High At least in part, this reflects regression toward the mean. But that's not the whole story. Typical method -Collect probability estimates for lots of cases or questions. -For example, for each binary item on a trivia test, how confident are you that you got it right? -Group cases into subjective-probability bins. -Plot the objective probability for each of the subjective-probability bins, as on the next slide.

If the regret and disappointment functions have the right forms, they can explain...

many violations of SEU

Coherence

means consistency among a set of judgments (i.e., whether the judgments are consistent with each other). Expected utility theory, probability theory, and Bayes' theorem are mostly about coherence. We are often not very coherent in our judgments. We overestimate conjunctions, underestimate disjunctions, underweight base rates, and so on...

Correspondence

means consistency with some external measure of truth. Often, correspondence is low as well. Clinical versus actuarial judgments, misjudging correlations,...

Three types of overconfidence(Moore & Healy, 2008)

overprecision, overestimation, overplacement

choice overload

too much choice can be debilitating

Overconfidence

A (self-flattering) lack of correspondence between one's assessment of one's skill or knowledge and one's actual skill or knowledge.

Affect and the valuation function (Hsee & Rottenstreich, 2004)

Affective stimuli lead to "valuation by feeling." -Like System 1 -Presence or absence is important. -Number or magnitude is less important(insensitivity to scope). Less affective stimuli lead to "valuation by calculation." -Like System 2 -More linear value function (greater sensitivity to scope)

Information search

Alternative-based versus attribute-based strategies

MAUT procedures: Problem structure, scores, and utilities

Assemble or create a set of alternatives. -Consider only the "decent" alternatives. Choose a set of relevant attributes. -Some authors recommend that you do this first,because it can lead to creative "new" alternatives. Score each alternative on each attribute. -Scores can be in real units (e.g., $) or on simple scales. It's often useful to convert these scores to utilities. -Puts all attributes on the same scale (e.g., 0-100) -Allows for nonlinear utility functions -Larger apartments are better, but only up to a point.

More defaults as nudges

Automobile liability insurance (Johnson et al., 1993) -New Jersey and Pennsylvania both offered lower insurance premiums for policies with a reduced right to sue for pain and suffering. -In NJ, the limited-right-to-sue policy was the default (chosen by 79%). -In PA, the full-right-to-sue policy was the default (chosen by 70%). Other examples -Assignment to healthcare or retirement plans -Automatically re-enrolling in a service or plan -Online shipping policies (air or ground) Online privacy policies -McKenzie et al. (2006) showed that people use defaults to infer the recommended course of action ("information leakage").14

The affect heuristic

Background: Anticipatory affect in the Iowa Gambling Task -In a risky-choice task involving different decks of cards, people felt anxious just before choosing from one of the "bad" decks. -This anticipatory affect steered people away from poor choices. -Patients with damage to a decision-making area of the brain did not have the anticipatory affect and continued to make poor choices. The affect heuristic -"As used here, affect means the specific quality of 'goodness' or 'badness' (1) experienced as a feeling state (with or without consciousness) and (2) demarcating a positive or negative quality of a stimulus. Affective responses occur rapidly and automatically.... We argue that the reliance on such feelings can be characterized as the affect heuristic" (Slovic et al., 2002, p. 397). -Affect may be used as a cue for judgment. -Affect may be substituted for more difficult assessments.

Misperceptions of randomness (Gilovich et al., 1985)

Basketball fans considered sequences of Xs (hits) and Os (misses). -Fans classified sequences as "chance shooting" or "streak shooting." Sequences with high alteration rates were usually perceived as random (chance shooting). Sequences with 50% alternation rates (more like random coin flips) were usually perceived as streak shooting. People expect random sequences to alternate more (and have shorter streaks) than they really do. -So, if we think a process is random, we may also think (incorrectly) that an ongoing streak should end.

Using frequencies instead of probabilities improves and reduces what?

Bayesian reasoning (Gigerenzer & Hoffrage, 1995) and reduces conjunction errors (Fiedler, 1988). Providing information in frequency terms can be viewed as a nudge, though Gigerenzer would disagree.

MAUT procedures:Summary evaluations

Calculate a summary evaluation or "global utility" for each alternative. The xij are the scores of the alternatives (i) on the various attributes (j). The uj(xij) are the single-attribute utility functions. The wj are the weights for the various attributes. The Ui are the summary evaluations of the alternatives. Choose the alternative with the highest evaluation. -Consider doing sensitivity analyses.

Choice architecture and "nudges"(Thaler & Sunstein, 2008)

Change the choice environment rather than the person. -As with incentives and accountability. -These are changes to the environment, but not really "nudges."

Three approaches to debiasing (Soll et al., 2015)

Change the person. -Education and improved cognitive strategies Change the environment. -Incentives, accountability, and choice architecture ("nudges") Use a statistical model or decision aid.

The compromise effect:Results from a previous semester

Choice set 1 -Battery A lasts 32 hours and costs $5.20 18 (58%) -Battery B lasts 38 hours and costs $6.20 8 (26%) -Battery C lasts 24 hours and costs $4.40 5 (16%) Choice set 2 -Battery A lasts 32 hours and costs $5.20 5 (19%) -Battery B lasts 38 hours and costs $6.20 14 (52%) -Battery C lasts 42 hours and costs $8.80 8 (30%) Battery C affects the relative attractiveness of A and B. -The ratio of Battery B/Battery A increased from 0.44 to 2.80 in choice set 2. -Similar to (but weaker than) Huber and Puto's (1983) results.

Featured of MAUT

Compensatory, exhaustive, and very high on mental effort (exhausting!) It should lead to the optimal (best) choice. That's why it's considered normative. Recommended for important decisions, but not for less important ones. MAUT is used for lots of important policy deci

Iyengar & Lepper (2000, study 1)

Customers at an upscale grocery could sample jams from an array of either 6 or 24 jams at a tasting booth. All customers who sampled jams got a $1-off coupon to use to buy any jam. Results -DV #1: More people stopped at the 24-jam booth than at the 6-jam booth (60% vs. 40%). -DV #2: Overall (not just among those who stopped), more people bought a jam from the 6-jam booth than from the 24-jam booth(30% vs. 3%).

Decoy effects and reasons(Shafir, Simonson, & Tversky, 1983)

Decoys give people a reason for preferring one of the other options.

Basketball fans' beliefs about the "hot hand" (Gilovich et al., 1985, study 1)

Does a player have a better chance of making a shot after having just made the last two or three shots than after having missed the last two or three shots? -91% of fans said yes. -For a 50% shooter, the estimated chance was 61% after having made a shot versus 42% after having missed a shot. Is it important to pass the ball to someone who has just made several shots in a row? -84% said yes. -Players hold similar beliefs. Gilovich et al. (1985) found that basketball players actually do not get the hot hand. -NBA field goals and free throws,Cornell players' unguarded shots -These results were surprising and controversial. -But they were backed up by many additional studies and analyses. Tentative conclusion: There is little, if any, evidence of streak shooting in basketball, but we believe that streak shooting is common and important. Generalization based on more recent data -Evidence for "getting hot" is more likely in individual sports with repetitive motions, like billiards, darts, and golf putting (Oskarsson et al., 2009). -Not for team sports.

Adaptiveness

Does the use of resources match the demands of the choice task (e.g., time, stakes)?

Defaults as nudges for organ donation(Johnson & Goldstein, 2004)

Donation rates are higher when consent is the default (in "opt-out" programs). -Survey results (left) and observed consent rates (right). The U.S. has an "opt-in" system, but it may operate more like a neutral or "mandated choice" system. -The consent rate was 49% in 2019; 59% in Ohio

Punch lines from studies of carryover emotions

Emotions from one situation carry over to other judgments and decisions. -Another example: Charitable donations are larger on sunny days (Cryder & Weber, 2010) Emotions are more nuanced than simply positive or negative. -Different negative emotions (sadness, disgust, anger, fear) have different effects. -Another example: After the 9/11 attacks, Americans who were angry (rather than fearful) reported lower risk estimates and precautionary actions and supported more vengeful (rather than conciliatory) policies (Lerner et al., 2003).

Disgust

Endowment effect eliminated Lower selling and choice prices, consistent with an "expel" goal

Sadness

Endowment effect reversed Consistent with a "change" goal

Debiasing "association-based" errors (Arkes, 1991)

Errors based on what information is currently activated Examples -Hindsight bias -Overconfidence -Conjunction errors based on representativeness What (often) works -Activation of different associations -State reasons supporting different possibilities. -"Consider the opposite." -Accountability (see Lerner & Tetlock, 1999) -Accountability often leads to thinking about counterarguments. -Other incentives (e.g., monetary stakes) don't have this effect.

Debiasing "psychophysically based" errors (Arkes, 1991)

Errors that result from functions like prospect theory's utility function -They arise from loss aversion or diminishing marginal sensitivity. Examples -Framing effects -Sunk cost effects -Psychophysics of spending -A $400 stereo upgrade may seem reasonable in a $30,000 car What (often) works -Frame situations in more than one way. -Think about what else you could do with the money. -Think about "opportunity costs."

Information use

Exhaustive versus non-exhaustive strategies

elation

For good outcomes, the opposite of disappointment

rejoicing

For good outcomes, the opposite of regret

A major correction to the hot-hand literature (Miller & Sanjurjo, 2018)

Gilovich et al.'s (1985) paper was shocking because it showed that, despite people's strong beliefs, basketball players actually do not get the hot hand. -These results have been confirmed in numerous follow-ups with additional data. 30 years later, however, Miller and Sanjurjo (2018) pointed out that Gilovich and others had used a biased statistical test. With an unbiased test, there IS evidence for the hot hand in basketball, even in the original data Unfortunately, Miller and Sanjurjo's analyses are complicated and even a simple explanation is very time-consuming. -For the curious, I've posted their paper, some extra PowerPoint slides and a lecture video on the topic, and a link to an informative ESPN magazine article. -This material is NOT required. Also, these new results do NOT necessarily mean that people's perceptions are accurate. -In some cases, people's estimates of the magnitude of hot-hand effects are still too large.

Regret and the umbrella problem

If it rains, you'll regret not taking your umbrella. If it doesn't rain, you'll regret taking it. The amount of regret is a function of the difference between what you got and what you would have had if you had chosen differently. For good outcomes, the opposite of regret is "rejoicing."

Disappointment and the umbrella problem

If you don't take your umbrella and it rains, you'll be disappointed because you got something worse than you expected. For good outcomes, the opposite of disappointment is "elation."

Decision affect theory (Mellers, 2000)

Imagine a choice between two independent gambles. -Gamble 1 yields either outcome A or outcome B. -Gamble 2 yields either outcome C or outcome D. Imagine that you choose gamble 1 and outcomes A and C occur. -Disappointment (or elation) involves a comparison of A and B. -Regret (or rejoicing) involves a comparison of A and C. -In decision affect theory, both of these effects are greater if the outcome is surprising. In the studies, participants judged their anticipated pleasure (happiness) with many different combinations of outcomes.

Irrelevant or distracting options

Imagine that you're given two alternatives, A and B. -You prefer A Now imagine that you're given those same two alternatives along with a third alternative, C. -You may or may not like C, but the presence of C should not affect your preference of A over B. Yet sometimes it does. -The "decoy effect" There are various versions of the decoy effect. -Compromise effect = Extremeness aversion -Attraction effect = Asymmetric dominance

Too much choice?

In a recent meta-analysis of 99 studies, the overall effect of having many options was zero. -Chernev, Böckenholt, & Goodman (2015) Sometimes, having many options leads to choice overload. -Studies at the top of this graph But other times, having many options is helpful. -Studies at the bottom of this graph It's complicated: There are several moderators of the effect. -E.g., complexity of the option set, preference uncertainty

Summary of advice for debiasing

Match the remedy to the cause of the error. -Work with our psychology, not against it (e.g., with frequency formats). -More effort may not be enough; more balance may be needed. -Activate other associations by "considering the opposite" Change the person. -Education and better cognitive strategies: reframe situations, consider the opposite, consider opportunity costs, etc. Change the environment. -Incentives and accountability -Choice architecture (nudges): defaults, scale transformations, frequency formats, etc. Consider a statistical model or decision aid.

Accountability(Lerner & Tetlock, 1999)

Most accountability research involves predecisional accountability to audience with unknown views. -Basic effect: More intensive, more balanced processing -Whether this helps or not depends on the source of the bias(like Arkes, 1991). Attenuates biases that result from a lack of effort: Anchoring, Overconfidence, Conjunction errors,Sunk costs (sometimes) Has no effect when people lack the ability or skill to do better. -Insensitivity to base rates or sample size, Preference reversals Amplifies biases that seem justifiable -Ambiguity aversion, Loss aversion, Decoy effects

The mugger study

Mugger had either 0 or 4 previous convictions Empathy and no-empathy conditions -"Put yourself in the position of the victim(s) and think about how you would feel when being mugged at night. Please write a sentence below to describe your feelings." Results (average length of sentence) No empathy -2.6 years if 1st conviction -5.8 years if 5th conviction Empathy -3.4 years if 1st conviction -4.6 years if 5th conviction The other three studies had similar crossover interactions.

Multiattribute utility theory or technique- MAUT

Normative model Consider all relevant attributes of all decent alternatives. -Use a weighted linear model to calculate summary evaluations. Lots of short-cut strategies (heuristics)

When decomposing the problem makes things worse

Often, we do not have much insight into our own cognitions. -Judges lack insight into their "policies." -Prospect theory predicts choices well, but we wouldn't normally describe our choices that way. -Decision utility ≠ experienced utility Sometimes, thinking about the reasons for our choices can lead us to make worse choices, because we focus on the wrong reasons.

Regression toward the mean: Predicting posttests from pretests

On Test 1 (pretest), the average scores was 70(range = 40 to 100). Same distribution for Test 2 (posttest) Scores are correlated, but not perfectly. Regression toward the mean -Those who scored above average on Test 1 score lower on Test 2. -Those who scored below average on Test 1 score higher on Test 2.

Maximization

Optimal versus "good enough" choices

Punch lines from Mellers' decision affect theory

Our affective response to an outcome depends on... -The objective outcome -What we would have received if we had chosen differently -Regret or rejoicing -What else we might have received, given the choice we made• Disappointment or elation -How surprising the outcomes are We anticipate these emotions and use them to guide our choices. -This works because we are pretty good at correctly anticipating our emotional responses. -E.g., gamble outcomes, exam grades, weight changes, pregnancy tests -Gilbert and Wilson would say that we focus way too much on our immediate emotional responses, rather than what we'll feel a bit later.

The cancer-cluster myth (Gawande, 1999)

Outbreaks -Legionnaire's disease, AIDS, E. coli in spinach, disease from occupational exposures - Initial concerns often lead to a biological or chemical culprit. Residential cancer clusters -There are "too many" cancers in a particular neighborhood. -The coincidence seems very unlikely, so we search for a cause. -But the search is almost always unsuccessful. What are the chances? -There are many different types of cancer and many different neighborhoods, so the likelihood of a cluster of some kind of cancer in some neighborhood is quite high. -"Given a registry of 80 different cancers, you could expect 2750 of California's 5000 census tracts to have statistically significant but perfectly random elevations of cancer." -For any given tract, P(Some cancer appears elevated) > 0.5 -If there is really no elevated cancer risk, and α = 0.01,then P = 1 - P(No cancer appears elevated) = 1 - 0.9980 = 0.55.

Other overconfidence results(based mostly on calibration studies)

Overconfidence in judgment is a common, robust finding. -Replicated many times across different cultures -Not related to intelligence -Not everyone is overconfident. -Weather forecasters, bookies, and bridge players do very well. Gambling on trivia questions (real stakes involved) -Same degree of overconfidence -Suggestion that overconfidence even increases with incentives Accuracy is often uncorrelated with confidence and experience. -Eye-witness testimony -Clinical psychologists' judgments (Oskamp, 1965)

Overestimation

Overestimating one's ability, performance, level of control, or chance of success (e.g., chance of being correct) Often assessed in "calibration" studies

Carryover emotions and the endowment effect (Lerner et al., 2004)

Part 1: Film clips and subsequent writing were used to induce disgust, sadness, or neither (neutral). -Manipulation checks at the end of the study (after Part 2) verified that participants felt the intended emotions Part 2: Tested the endowment effect for a highlighter set -Selling condition -Keep the highlighter set or trade for an amount of money -$0.50 to $14 in $0.50 increments -The true selling price was then determined randomly. -The seller's earlier choice at that price was binding. -Choice condition (rather than a buying condition) -Like the selling condition, but without the endowment of the highlighter set.

"Dialectical bootstrapping"(Herzog & Hertwig, 2009)

Participants estimated the dates of 40 events from the 16th to 19th centuries. -E.g., the discovery of electricity Three conditions for generating second estimates -Reliability: Participants made second estimates without any particular guidance. -Dialectical bootstrapping: "First, assume that your first estimate is off the mark. Second, think of a few reasons why that could be. Which assumptions and considerations could have been wrong? Third, what do these new considerations imply? Was the first estimate rather too high or too low? Fourth, based on this new perspective, make a second, alternative estimate." -Other person: Use another random person's first estimate as the second estimate. The first and second estimates were then averaged. Playing devil's advocate with oneself can improve judgments, but not as much as getting a "real" second opinion.

Shah & Wolford (2007)

Participants evaluated some number of pens, ranging from2 to 20. Afterward, they could buy a pen at about half price($1 instead of about $2). Results -Some choice increases the likelihood of purchase, but too much choice leads to indecision.

Misperceptions of randomness

People expect random sequences to alternate more (and have shorter streaks) than they really do. -We expect them to look random even at very small scales. -"Local representativeness"

Illusory correlation

People focus on positive co-occurrences, particularly if they "fit the theory

Elimination by aspects (another choice heuristic)

Pick the most important (or most salient) attribute and set a cutoff. -No more than $700 rent -Throw out options that don't make the cutoff. Carry others forward. Pick next most important attribute and set a second cutoff. -At least 800 square feet -Throw out options that don't make this cutoff. Carry others forward. Continue through the attributes until only one alternative remains. -At least a 4 on attractiveness -No more than a 10-minute commute Consider all alternatives on one attribute at a time. -An attribute-based strategy Can be used to select a subset for further consideration.

Two reasons to study debiasing (Larrick, 2017)

Practical -To help people make better judgments and decisions Theoretical -To help understand the various sources of biases -If we learn how to moderate a bias (to increase it, decrease it, or even eliminate it), then we learn about its psychological causes.

What Joint versus separate evaluation(Hsee, 2000) means

Preference reversals again-Context and the focus of attention matter. A practical implication of the JE-SE difference -When purchasing, we usually engage in joint evaluation. -When we're shopping for a new TV, minor differences in picture quality may seem important. -When consuming, we usually engage in separate evaluation. -The picture looks fine at home. -But we may notice that the remote control is hard to use. -We may pay attention to the "wrong" attributes when purchasing. This is a bit like an affective forecasting error.

Asymmetric dominance:Results from this semester

Q16, Choice set 1 -Battery A lasts 32 hours and costs $5.20 17 (53%) -Battery B lasts 40 hours and costs $7.60 12 (38%) -Battery C lasts 30 hours and costs $5.80 3 (9%) Q16, Choice set 2 -Battery A lasts 32 hours and costs $5.20 10 (38%) -Battery B lasts 40 hours and costs $7.60 16 (62%) -Battery C lasts 38 hours and costs $8.20 0 (0%) Battery C affects the relative attractiveness of A and B. -The ratio of Battery B/Battery A increased from 0.71 to 1.60 in choice set 2.

Changing the scale as a nudge (Larrick & Soll, 2008)

Q3. Which of the following would save more gas for a person who drives 12,000 miles per year? A. Switching from a vehicle that gets 34 miles/gallon to one that gets 50 miles/gallon [a 47% increase] B. Switching from a vehicle that gets 16 miles/gallon to one that gets 20 miles/gallon [a 25% increase] Most people choose A. -Class data: 43 of 58 students (74%) chose A. But the opposite is true. -This misperception is the "MPG illusion." -The Miles per Gallon (MPG) metric makes it hard to judge the effect of increasing fuel economy. For reasoning about fuel usage, "Gallons per 100 miles" is a much better metric. Lower is better. In Study 3, Ss made choices about upgrading a fleet of vehicles. -In the MPG condition, only 25% of Ss correctly chose an increase from 15 to 19 MPG over an increase from 34 to 44 MPG. -In the GPM condition, 64% of Ss correctly chose a reduction from 6.67 to 5.26 gallons/100 miles over a reduction from 2.94 to 2.27 gallons/100 miles.

Overprecision

Reporting confidence intervals that are too narrow, as in the previous task

Performance of choice heuristics

Satisficing, EBA, and MCD are "boundedly rational." Disadvantages -They may not choose the best overall alternative. -The choice can be affected by irrelevant factors, such as search order. -They sometimes lead to the violation of "important" axioms. Advantage -Big savings in time and effort Performance -When time and attention are limited, simple choice heuristics can perform better than exhaustive strategies. -People are adaptive when choosing among strategies (Payne, Bettman, & Johnson, 1988).

The "Texas-sharpshooter fallacy" (a term from epidemiology)

Seeing patterns in randomness ...

Joint versus separate evaluation(Hsee, 2000) ice cream

Serving A:7 oz in5 oz cup separate eval: $2.26 Serving A:7 oz in5 oz cup Joint eval: $1.56 Serving B:8 oz in10 oz cup separate eval: $1.66 Serving B:8 oz in10 oz cup Joint eval: $1.85

Satisficing (a choice heuristic)

Set cutoffs on all important attributes. -No more than $700 rent -At least 800 square feet -At least a 4 on attractiveness -Ignore less important attributes (e.g., commute time) Consider alternatives (apartments) one at a time. -E.g., Start with Apt. A, then consider Apt. B, etc. -An alternative-based strategy Pick the first alternative that exceeds all of the cutoffs Can be used to select a subset for further consideration.

Evaluability hypothesis

Some attributes are easy to evaluate in isolation (e.g., torn cover), whereas others are not (e.g., number of entries). More evaluable attributes play a greater role in the separate evaluation mode.

Choice among options with many features

Some choices involve only one attribute at a time. -Money, life expectancy, Hershey's kisses, etc. -Many of these choices also involve uncertainty (gambles). -EV, EU, and PT But many choices involve options with many different attributes. -Selecting a car, an apartment, an insurance plan, a significant other,... -Many don't involve uncertainty (at least explicitly). Normative model- MAUT

Neither emotion (neutral)

Standard endowment effect

Iyengar & Lepper (2000, study 2)

Students in an undergraduate class could get extra credit by writing a two-page essay. Students were given either 6 or 30 choices of potential topics. Results -DV #1: Assignment done?• 6-choice group: 74%• 30-choice group: 60% -DV #2: Grades• On average, students in the 6-choice group got higher grades for "content" and "form" on the assignment.

Give your 90% confidence intervals(based on Russo & Shoemaker, 1989)

Students' intervals included the true value much less than 90% of the time, implying that the intervals were too narrow. This always happens.

Choosing jams(Wilson & Schooler, 1991)

Subjects tasted and rated 5 strawberry jams. -Reasons group: Rated the jams after listing their reasons for liking or not liking each jam -Control group: Did not list reasons Ratings were compared to those of "experts." -Trained sensory panelists for Consumer Reports evaluated jams on 16 characteristics. Results: Average Spearman correlation with experts -Control group 0.55 -Reasons group 0.11 -But agreement with professional "jamologists" is not a particularly good standard for correctness.

Choosing posters(Wilson et al., 1993)

Subjects viewed and rated 5 posters, then chose 1 poster to keep. -Reasons group: Rated the posters after listing their reasons for liking or not liking each poster -Control group: Rated the posters after an irrelevant task Pretest: The "art" posters were clearly better than the "humorous" posters. Results: Subjects in the reasons group... -Had much weaker preferences for art posters over humorous posters. -Were more likely than control subjects to choose the humorous posters. -Were less satisfied than control subjects weeks later, on both subjective and behavioral questions. To me, this study is more convincing than the jam study, because it has a better standard for correctness. Not knowing what the important attributes are may be a bigger problem for aesthetic choices than for other choices.

Gambler's fallacy

The belief that a particular random event is "due" if it has not occurred for a while. More specifically: The belief that a streak of several similar outcomes (e.g., coin tosses landing on "heads") decreases the chance of the same outcome on the next trial. -Coins, dice, roulette, sexes of children, 100-year floods, etc. However, if a process really is random, then the events in the sequence are independent. -A string of several similar outcomes (e.g., heads, girls) does not predict the next outcome in the sequence.

The "hot-hand belief

The belief that a streak of several similar outcomes (e.g., shots made in basketball) increases the chance of the same outcome on the next trial (e.g., shot). This belief may or may not be correct, depending on the circumstances. If we think that skill, intention, or control is involved, then longer streaks seem more likely, and more likely to continue.

Overplacement

The belief that one is better than others (e.g., more skilled) The "better than average" effect Most people think that they drive better than the average person. Better than the median

Disappointment (or elation

The comparison is to the expectation for the choice you made. The more surprising your outcome, the greater the disappointment or elation. The expectation includes what would have happened in the other state of the world for the same choice. Within a row in the previous tables

Regret (or rejoicing)

The comparison is to what would have happened in the same state of the world if you had made a different choice. Within a column in the previous tables

Regression toward the mean: Predicting pretests from posttests

The same is true in reverse Those who scored above average on Test 2 score lower on Test 1. Those who scored below average on Test 2 score higher on Test 1. You don't need a causal explanation for these results. -Imperfectly correlated variables always work this way.

Differences among choice strategies

There's usually no option that's best on all attributes. -E.g., the biggest apartment is not the least expensive. -At a particular price point, desirable attributes are often negatively correlated. Tradeoffs -Compensatory strategies -"The house is expensive, but it's worth it for the view." -Noncompensatory strategies -"It doesn't matter how safe and fast and durable the Volvo is;I can't afford to spend $50,000 on a car." -It's raining, so let's just go to whichever restaurant is closest.

Bounded rational

They don't adhere exactly to principles of rationality. But they approximate those principles (to different degrees).

Features of nudges

They don't forbid any options. They don't significantly change the incentives. They must be easy and cheap to avoid. The goal is to steer people toward better choices while still preserving people's freedom of choice. "Libertarian paternalism"

Reducing overconfidence

Training helps. -When people are given intensive feedback after each judgment, they learn to reduce overconfidence. -Weather forecasting versus medical diagnosis? Stop to consider reasons why your judgment might be wrong. -Participants asked to give reasons for and against each of the alternatives before giving an answer and a confidence rating became more accurate and better calibrated. -It's the reasons against the preferred answer that matter.

Punch lines from Hsee and Rottenstreich's studies

We can and do evaluate prospects in two different ways. -Affectively, like System 1 -Analytically, like System 2 -The evaluation method is often determined by what we're evaluating, but we can use either way or both ways. Prospect theory's functions are more curved for affect-rich stimuli and more linear for affect-poor stimuli. -The utility function -The probability weighting function Affective evaluation is more qualitative. -Some or none, possible or impossible,certain or uncertain

Post-hoc judgments of probability

We often judge the probability of a particular pattern or sequence, after it has occurred. Instead, we should judge the probability of any similar pattern or sequence. We should also consider the processes that lead to seemingly unlikely and therefore impressive outcomes. -A stock trader who beats the market several years in a row must be very skilled, right? -Taleb (2004) says that the success may be due to"survivorship bias" instead.

Results for experts (Plous, 1993)

Weather forecasters' predictions of precipitation are remarkably well-calibrated. Physicians' estimates of the likelihood of pneumonia reflect gross overconfidence (overestimation).

MAUT procedures: Attribute weights

Weight each attribute by its importance. Here's one procedure... -Rank the attributes in terms of importance. E.g., "1" is most important, "2" is next most important, etc. -Create "raw weights." E.g., zero(?), 10, 15, 40, 100, 150 Standardize the weights by dividing each one by the total weight. E.g., 40/315 = 0.127 -The standardized weights sum to 1.0. When assigning attribute weights, keep the ranges of the attributes in mind. E.g., square footage might be important "in principle," but it will be less important if the available apartments don't vary much on that dimension.

Regression toward the mean: Not for perfectly correlated scores

When scores are perfectly correlated, there is no regression toward the mean. But variables are (almost) never perfectly correlated.

When does MAUT perform poorly?

When there is time pressure or other constraints. When it makes you pay too much attention to unimportant attributes. For some aesthetic judgments, where we have trouble identifying what the important attributes are (examples in an upcoming lecture). It's not a good descriptive model of what people actually do in most situations.

Regression toward the mean: Other examples

When two variables are not perfectly related, very high or very low scores on one scale are associated with more average scores on the other scale. Extreme scores and performances tend to be followed by less extreme scores and performances. Examples -Heights of fathers and sons -Height and weight -Sports Illustrated "jinx" -"Sophomore slump" in sports, music, and movies

Debiasing "strategy-based" errors(Arkes, 1991)

When you know how to do better, but you just don't Examples -Satisficing and lexicographic strategies in MAUT tasks -Even when there is time to use a more complete strategy -Errors in assessing covariation -E.g., between symptom and disease What (often) works -Raise the stakes or "level of involvement" to increase the benefits of accurate judgments. -Leads to the use of more data in a more sensible manner -Leads to consideration of the merits of arguments rather than just the number of arguments

Anticipated emotions:The Allais paradox

Which would you choose, A or B? -A $1,000,000 for sure -B 89% chance of getting $1,000,000 10% chance of getting $2,500,000 1% chance of getting nothing Which would you choose, C or D? -C 11% chance of getting $1,000,000 89% chance of getting nothing -D 10% chance of getting $2,500,000 90% chance of getting nothing Many people choose A and D, which violates SEU. One explanation: They choose A because they anticipate that they would regret passing up the sure $1,000,000 if they chose B and lost.

A real-life example(Simonson & Tversky, 1982)

Williams-Sonoma catalog -Bread maker was $275 -Sales were pretty weak. -Then they added a second, similar bread maker to their catalog, priced at $429. -Few people bought the $429 model, but sales of the $275 model nearly doubled. Think about that the next time you're shopping for a smart phone, a TV, a computer, or a car The punch line is that the choice context matters, sometimes a lot. -Constructed preferences again

Four nifty studies(Hsee & Rottenstreich, 2004)

Willingness to pay for 5-CD or 10-CD sets of Madonna -After a priming task involving feelings or calculations Number of hours one would work at the bookstore -For $30 or $60 cash (affect-poor) -For a music book worth $30 or $60 (affect-rich) Donations to save 1 or 4 pandas -Represented by dots (affect-poor) -Represented by pictures (affect-rich) Prison sentence for a mugger

Joint versus separate evaluation(Hsee, 2000)

Willingness to pay for music dictionaries Dictionary A:1993,10,000 entries,Like new separate eval: $24 Dictionary A:1993,10,000 entries,Like new Joint eval: $19 Dictionary B:1993,20,000 entries,Torn cover Separate Eval: $20 Dictionary B:1993,20,000 entries,Torn cover Joint Eval: $27

Seeing correlation and causation Smedslund (1963)

conducted a similar study of nurses involving 100 "patient cards" with disease and symptom information. 86% incorrectly reported a positive association. Illusory correlation

local representativeness

does not predict whether people think a streak is likely to end (the gambler's fallacy) or continue (the hot-hand belief). You also need a separate judgment regarding the randomness, intentionality, or controllability of the process (Burns & Corpus, 2004; Oskarsson et al., 2009; Tyszka et al., 2008).

Nudges

factors which encourage people to think and act in particular ways. Nudges try to shift group and individual behaviour in ways which comply with desirable social norms Nudges are designed to be more subtle. -Example: Putting fruit at eye level