IND 641 Engineering Psychology CHAPTER 8 Decision Making

IND 641 Engineering Psychology CHAPTER 8. Decision Making q FEATURES AND CLASSES OF DECISION MAKING q Uncertainty – involving risk q Familiarity and Expertise – rapidly and little deliberation; experts not always more accurate q Time – one shot vs. evolving decisions; time pressure q Classes of Decision-Making Research q rational or normative decision making – decisions according to optimal framework – optimal beta q cognitive or information processing approach – biases, limitations, heuristics q naturalistic decision making – decision making under real environment -expertise, complexity q AN INFORMATION PROCESSING MODEL OF DECISION MAKING (fig 8. 1) q cue (ambiguous & incorrect) seeking -- selective attention – experiences and attentional resource q diagnosis – situation assessment or situation awareness 1. external cues filtered by selective attention (bottom-up processing) 2. LTM – hypotheses and the estimation of the likelihood or expectancy (topdown processing) q often incorrect -- uncertain nature of cues or vulnerabilities for selective attention and WM q iterative – search for further info -- feedback loop to cue filtering , confirmation 고려대학교 산업공학과

IND 641 Engineering Psychology q WHAT IS “GOOD” DECISION MAKING? q the expected value of a decision q produce “good” outcomes q expertise q DIAGNOSIS AND SITUATION AWARENESS IN DECISION MAKING q Quality of diagnosis q the role of perception in estimating a cue q the role of attention in selecting and integrating the info by the cues q the role of LTM to establish possible hypotheses or beliefs q the role of WM to update and revise beliefs or hypotheses q Estimating Cues: Perception q human as intuitive statistician q perception of mean – relatively well q perception of proportions q dichotomous observation – reasonably accurate between 0. 05 and 0. 95 (midrange) q more extreme proportion – conservative, biased away from the extremes of 0 and 1. 0 – conservative tendency (never say never), salience or infrequent event -- overestimate q estimation of the variance (fig 8. 2) q estimate the variability as less if the mean of the values is greater -Weber’s Law of psychophysics -- the concept that a just-noticeable difference in a stimulus is proportional to the magnitude of the original 고려대학교 산업공학과

IND 641 Engineering Psychology q estimation of correlation – underestimation of high correlation, vice versa q estimation in extrapolating nonlinear trends – toward more linear (fig 8. 3) q Evidence Accumulation: Cue seeking and Hypothesis Formation (fig 8. 4) q cue properties 1. cue diagnosticity – how much evidence a cue should offer – value and polarity 2. cue reliability or credibility – the likelihood that the physical cue can be believed – independent of diagnosticity – information value of a cue = diagnosticity x reliability 3. physical features of the cue – conspicuous or salient – important bearing on attention q multiple cues 1. selective attention – different weight according to their info value 2. integration (bottom-up processing) of perceptual features in pattern recognition or dimensions in an object display 3. expectancies biasing -- top down processing in perceptual pattern recognition 4. not parallel to perceptual pattern recognition -- iterative testing and retesting of a belief q Attention and Cue Integration q Information Cues are Missing q what they do not know (missing cues) – seek these cues before a firm diagnosis 고려대학교 산업공학과

IND 641 Engineering Psychology q under time stress – more info deteriorated decision making performance q selective filtering strategy – compete for the time available for the integration of info – more info leads to time consuming filtering process at the expense of decision quality q Cues are Differentially Salient q the salience should be directly related to the info value of the cue in making a decision, not just detecting a fault q the info that is difficult to interpret or integrate – underweighted q the absence of a cue – what is not seen, symptoms not observed q Processed Cues Are Not Differentially Weighted q do not effectively modulate the weight of a cue based on its value �as if they were of equal value �reducing the cognitive effort required to consider different weights q weighting varies in more of an “all or none” fashion (fig 8. 5) q why use as if heuristic? �cognitive simplification q Expertise and Cue Correlation q multiple cues with highly correlated each other and equally weighted (fig 8. 6) �intuitive form of info integration, similar to perceptual pattern recognition � closely associated with expertise q RPD (recognition-primed decision making) q recognizes the pattern of cues as a typical cues in prior experience� rapid and relatively automatic categorization �expert decision makers under high time stress q no correlation, no time pressure, a single cue salience – abandon RPD, 고려대학교 산업공학과 rather a slower, more analytical diagnosis

IND 641 Engineering Psychology q Expectations in Diagnosis: The Role of Long-term Memory q q q q q two aspects of LTM in diagnosis, reflected perception and pattern recognition 1. cue correlation -- RPD 2. hypothesis frequency – most expected, most frequent diagnostic category Representativeness diagnosis by comparing cues, symptoms, or perceptual evidence with the set that is representative of the hypothesis on the basis of experience in LTM � typical of RPD or visual pattern recognition nothing wrong but used when the cues are somewhat ambiguous without adequately considering the base rate, probability or likelihood physical similarity to a prototype hypothesis dominates probability consideration if the physical evidence is ambiguous (missing) – use probability �availability heuristic Availability Heuristic approximating prior probability – people typically entertain more available hypotheses factors influencing the availability of a hypothesis (absolute frequency or prior probability) q recency q hypothesis simplicity q elaboration in memory of the past experience q Belief Changes Over Time: Anchoring, Overconfidence, and the Confirmation Bias 고려대학교 산업공학과

IND 641 Engineering Psychology q Anchoring Heuristic q not all hypotheses are treated equally �“mental anchor” to the initially chosen hypothesis �“first impressions are lasting” �primacy in memory q recency effect in cue integration q primacy is dominant when info sources are fairy simple and integration procedure is one that calls for a single judgment of belief at the end of all evidence (sequentially) q if the sources are more complex and often require an explicit updating of belief after each source is considered, then recency tends to be more likely (simultaneously) q The Confirmation Bias q a tendency for people to seek info and cues that confirm the tentatively held hypothesis or belief, and not seek those that support an opposite conclusion of belief � cognitive tunnel vision q three possible reasons 1. greater cognitive difficulty dealing with negative info than with positive info 2. higher cognitive effort to change the hypothesis 3. influence the outcome of actions taken on the basis of the diagnosis, which will increase their belief that the diagnosis was correct �self-fulfilling prophecy q Implications of Biases and Heuristics in Diagnoses q humans as “a bundle of biases” 1. heuristics are highly adaptive under rapid and not enough mental effort and time 고려대학교 산업공학과

IND 641 Engineering Psychology q CHOICE OF ACTION q Certain Choice (fig 8. 7) q compensatory method �satisficing rule – good enough q EBA (elimination by aspects) – reduce the cognitive effort, satisfactory q Choice Under Uncertainty: The Expected Value Model (fig 8. 8) q costs and values (benefits) – maximizing the expected value of a choice 1. Ps (probability of the state of the world) * Vxy (outcome value) 2. the expected value of each option 3. the greatest expected value as a choice q complicating factors to human decisions under uncertainty 1. maximizing gain/minimizing the loss – minimizing the maximum loss 2. difficult to assign objective values to different outcomes – safety 3. subjective estimates of objective values irrelevant to objective values 4. inconsistency between people’s estimates of probability and the objective probabilities q Biases and Heuristics in Uncertain Choice q Direct Retrieval q past experience, familiar domain, clear state of the world q RPD (recognition-primed decision making) under high time pressure q Distortions of Values and Costs q humans are trying to maximize an expected utility (subjective value of different expected outcomes) 고려대학교 산업공학과

IND 641 Engineering Psychology q prospect theory (Kahneman and Tversky, 1984): q potential loss exerts a greater influence over decision-making behavior than does a gain of the same amount – loss aversion q both positive and negative limbs are curved toward the horizontal -perceived value of Weber’s Law of Psychophysics q Perception of Probability q kahneman and Tversky – a function relating true prob. to subjective prob. q three critical aspects for understanding risky choice (fig 8. 10) 1. subjectively overestimate of the probability of very rare events – insurance and gamble 2. flat slope of its low probability end – reduced sensitivity to probability change 3. perceive probability as less than actual probability �framing effect q The framing effect q people’s preference with a choice between a risk and a sure thing q risk-seeking bias between negatives (avoidance-avoidance conflict) q risk-aversion bias between positives (risk and sure thing) q change in loss or gain, depending on the neutral point or frame of reference for the decision making – framing effect (frames of reference) q “sunk cost” bias q Rationally, the previous history of investment should not enter into the decision for the future. Yet it does. q investors for poor previous decision – sure loss and risky loss q newcomer – “sure thing” option is neither loss nor gain -- 0 utility and expected loss – bias to terminate the investment 고려대학교 산업공학과

IND 641 Engineering Psychology q IMPROVING HUMAN DECISION MAKING q Training Decision Making: Practice and Debiasing q Domains of decision making whether expertise develop from practice or not (table 8. 1) q problems of learning in decision making �the role of feedback in decision making problems 1. feedback is often ambiguous, in a probabilistic or uncertain world 2. delayed feedback – “Monday morning quarterbacking” “hindsight bias” 3. feedback is processed selectively – fig 8. 11 q debiasing – tailoring more specific training to target certain aspects of decision making flaws q reduced overconfidence, diagnostic information from the absence of cues, away from nonoptimal anchoring bias q provide more comprehensive and immediate feedback in predictive and diagnostic tasks (probability rather than frequency) q Proceduralization q a technique for outlining prescriptions of techniques that should be followed to improve the quality of decision making – fault tree and failure modes analysis q successful in certain real-world decisions that are easily decomposable into attributes and values q Automation: Displays and Decision Aids q three major categories of assistance that automation can offer 1. attention and cue perception – pictorial presentation over numerical or verbal, proximity compatibility principle 2. diagnosis – offload working memory, making inferences (expert system ) 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과

IND 641 Engineering Psychology 고려대학교 산업공학과