Gain/Loss Framing Effects on Learning in Economic Decision Making Investigated with Eye Tracking. (September 2017)

• Record Type: Journal Article
• Authors: Fernandez, Kylie; Merz, Melissa; Kuhnen, Camelia; Schmidt, Joseph; Lighthall, Nichole
• Is Part Of: Proceedings of the Human Factors and Ergonomics Society ... Annual Meeting. Volume 61:Part 1(2017)
• Page Start: 1166
• Page End: 1167
• Publication Date: 2017-09
• Subjects: Human engineering -- Congresses; 620.8205
• DOI: 10.1177/1541931213601775
• Languages: English
• ISSNs: 1541-9312

Abstract

Previous research has revealed a domain-based bias when people estimate payout likelihoods for probabilistic choice options that minimize losses versus those that maximize gains (Kuhnen, 2015). For instance, in economic boom situations, people overestimate how valuable a low profit stock is. Conversely, in economic recession situations, individuals underestimate stocks that minimize losses. Cognitive neuroscience posits that gain and loss information is processed differently in the brain (Knutson and Bossaerts, 2007; Kuhnen and Knutson, 2005), but the precise mechanisms of these domain differences are still unclear. The current study investigated two potential causes of this domain-based bias. Bias may be driven by a high magnitude effect, owing to greater salience for large gains and losses and subsequent overweighting in probability estimations. This would be evidenced by enhanced attention and memory for stimuli associated with high-magnitude dividend choice options and payouts. Domain-based bias in probability estimations could also be driven by incongruence between objective probabilities and dividend payout valence (e.g., "bad" choice options in the gain domain; "good" choice options in the loss domain; valence incongruence effect ). If true, we would expect enhanced estimation errors, reaction times (RT), and attention when there is incongruence between the valence of choice payout probabilities and their payouts. To test these hypotheses, 26 students from the University of Central Florida (UCF) participated in an economic decision-making study. The main task involved choosing between pairs of stocks (probabilistic payouts) and bonds (sure-thing payouts). Choice pairs were embedded in either a gain or loss block, with both choice options paying either positive or negative dividends, respectively. Stocks within a block were pseudorandomly drawn from either the "good distribution" (70% high payouts) or the "bad distribution" (30% high payouts). After choosing a security, the stock payout was shown and participants were then asked to estimate the probability that the current stock was drawn from the good distribution. Performance bonus payments were paid based on accurate stock probability estimates and 10% of the total earned from stock/bond choices. The study was approved by the UCF Institutional Review Board. Eye movements were recorded throughout to measure overt visual attention as a potential mechanism of domainbased bias. Measures included the fixation duration on each stimulus (dwell time) and average number of oscillations between choice options to determine when and where one looks (Carpenter and McDonald, 2007). Interest areas were created a priori around each critical stimulus in the choice and stock payout phases. To test memory for choice and stock payout phases, participants completed an incidental memory test at the end of the experiment. Here memory was assessed for fractal images associated with each stock and bond option, as well as face images associated with each stock payout ("stockbrokers"). The critical dependent variables to measure domain-based bias were estimation error, response time, oscillation between choice stimuli, and stimulus dwell time. The impact of memory, attention, and congruence of information on measures of domain-based estimation bias was examined with 2 x2 domain (gain, loss) by dividend payout (high, low payout) repeated-measures analysis of variance models (ANOVA). Mixed effects modeling was used to examine the power of outcome RT and visual dwell time to predict probability estimate bias. Behavioral results. Consistent with the valence incongruence hypothesis, absolute errors for stock payout probabilities were relatively higher when gain-domain stocks had worse expected values (gain stock was "bad") than associated bonds and when loss-domain stocks had better expected values than associated bonds (loss stock was "good"). In addition, RT during the choice phase was greater in the loss domain, as participants had to update their estimations the stock came from the "good distribution" even though it only lost money. For stock payout RT, the mixed effects model found an interaction of domain, payout magnitude, and outcome RT where the longer participants spent on gain outcome screens, the more positive their bias and the longer they spent on loss outcome screens, the more negative their bias. Results from the two incidental memory test scores did not reveal any main effects or interactions of domain or dividend payout, lessening support for the high magnitude hypothesis. The data provide support for both attentional effects. Eye tracking data . Greater oscillations between stock and bond options at choice was observed in the loss condition, suggesting greater choice uncertainty when stocks lose money. Stimulus dwell times were higher in the loss domain during the choice phase but did not differ by dividend payout. However, the mixed effects model found an interaction of domain and stock dwell times where the longer participants spent on gain information, the more positive their bias and the longer they spent on loss information, the more negative their bias. The mix of results provide support for both attentional effects. The behavioral results were in line with previous research (Kuhnen, 2015). Together with the eye tracking data, the results support the both the valence incongruence and high magnitude effects. We have evidence that one effect influences overall error rate (incongruence) and the other drives the direction of the error (magnitude). Thus, future interventions should consider both effects when seeking to improve decision making.