European DS Seminars Series
June 4th: Nicolò Bertani (Catolica Lisbon)
April 2nd: Konstantinos Katsikopoulos (Uni. of Southampton Business School)
March 5th: Ceren Cibik (University of Warwick & Ofcom): “Avoiding Dissonant Information”
Abstract: I examine whether prior exposure to information that contradicts one’s beliefs drives information avoidance. More specifically, I focus on the outlook towards abortion and the two main beliefs on abortion rights: “pro-life” (opposes abortion rights) and “pro-choice” (advocates abortion rights). In experiments with US respondents, I first vary the prior exposure to information: whether the information participants receive is in line with (consonant information) or contrary to their beliefs (dissonant information). I then measure avoidance of dissonance information using a willingness to pay measure. I find that a strikingly high proportion of participants are willing to avoid dissonant information at a material cost, using up almost half of their experimental budget. Prior exposure to dissonant information is insignificant in driving information avoidance. What matters most are beliefs: Pro-life participants are willing to spend a substantially higher proportion of their experimental budget to avoid dissonant information than pro-choice participants. An attempt to use text analysis to examine the reasoning behind dissonant information avoidance suggests that anticipation of negative emotions is a key mechanism driving information avoidance. This can also explain the difference in willingness to pay among opposing belief groups. These findings have implications for policies aiming to reduce political polarisation through information provision.
Feb-06, 2024 - Itzhak Gilboa (HEC): “Reasoning in Face of Uncertainty”
Abstract: How should we reason about uncertainty in the absence of objective probabilities? The common approach in economic theory is that the rational way to deal with such situations is Bayesian: using subjective probabilities when objective ones are not given and cannot be estimated. We take issue with this view, arguing that it is sometimes more rational to admit that we do not know a distribution than to pretend we do. Taking this viewpoint, one is faced with the question, what is a rational way to select a set of distributions, given a database of observations? We offer an axiomatic approach to this selection problem, yielding likelihood regions: sets of distributions that are monotonic with respect to the likelihood function. Starting with an abstract set of theories, we propose conditions on choice functions (across different databases) for which there exists a statistical model such that the choice function is a likelihood region relative to that model -- for the general case and for the case of a fixed likelihood-ratio threshold. We interpret the results as supporting the notion of likelihood regions for the selection of theories.
Jan-09, 2024 - Miguel Lobo (INSEAD) with Asher Lawson (INSEAD): “The Wisdom of Crowds and Jensen’s Inequality”
Abstract: The wisdom of crowds effect, whereby an estimate based on the average of individual estimates performs better than the average performance of the individual estimates, has been documented to work in many settings, but also to not work in others. In some problems, a solution obtained by some combination of individual solutions may perform worse that the average performance of the individual solutions. We argue that, in both cases, this is a consequence of Jensen’s inequality, applied to either a convex or concave cost function. This framework allows for a reliable prediction of the direction and strength of the effect. For instance, in forecasting a real-valued quantity with a convex cost function such as the squared prediction error, the wisdom of crowds effect always applies.
We validate this framework with ranking tasks from field data and from experimental data: 11,175 MBA students and executive education participants completing a class exercise where they are asked to prioritize items in a survival at sea simulation, and 1,270 experimental subjects ranking sets of 10 S&P 500 companies by revenue. For the experimental task, different sets of 10 companies for the ranking task were constructed based on a pre-test to estimate the bias (average error) and dispersion (how much person-to-person variation) in people’s perception of each company’s size. This allowed us to design ranking tasks that elicit different averages in individual performance, and where there is more or less diversity across individual views.
With sum-of-absolute-deviations scoring (l1 or Manhattan distance), we determine the areas where the cost function is convex vs. concave, and based on this predict that the wisdom of crowds effect is strongest when 1) the average individual performance is better, and 2) when the diversity in individual views is higher. For tasks where average individual performance is especially poor, the cost function becomes concave and the effects can be reversed. These predictions are strongly confirmed in both the field and experimental data.
Dec-05 - Ahti Salo (Aalto University): “Applications of Adversarial Risk Analysis and Cross-Impact Analysis in Defence”
Abstract: Models of adversarial risk analysis (ARA) help address decision problems in which there are several antagonistic players and the role of the analyst is to support one of the players. This player may be, for example, the Defender who can take pre-emptive actions before the Attacker decides whether to proceed with an aggressive act. In such a setting, it is pertinent to explore how sensitive the ARA results are to alternative assumptions about the Attacker. Specifically, when the Defender considers what portfolio of pre-emptive actions should be implemented, it is instructive to determine all portfolios of actions that are non-dominated in view of plausible assumptions about the Attacker. A detailed examination of these portfolios helps reveal those actions that are robust, in the sense that they would be selected for all these assumptions. We illustrate this approach with a realistic case study on military planning.
We also present a case study in which probabilistic cross‐impact analysis was employed to explore the impacts of three‐dimensional (3D) printing on the Finnish Defense Forces. In this case study, leading technological and military experts were consulted to obtain cross-impacts statements about the interdependencies between eleven key uncertainty factors representing technological progress, industry growth, and standardization, among others. These statements were then employed as parameters in an optimization model to derive probability estimates for the joint probability distribution over all scenarios, defined as combinations of possible outcomes for these uncertainty factors. We highlight some insights suggested by this structured process of scenario analysis.
Nov-07 - Anisa Shyti (IE): “A Brief Account on Attitudes Toward Ambiguous Time”
Abstract: Time plays a crucial role in life and business. Many business decisions are rooted in transactions that involve known outcomes (such as cash inflows and outflows), but for which the exact timing remains unknown. While the measurement of attitudes towards ambiguity has predominantly centered around probabilities, this paper shifts the focus to the timing uncertainty of known outcomes, examining attitudes toward ambiguous time. Given that time is naturally bounded only by the present, we construct time intervals to operationalize ambiguous time. We present a method that allows us to isolate attitudes toward ambiguous time, by eliminating from the equation the discount utility function. This approach enables us to measure how people perceive ambiguous time and how they respond to the resolution of uncertainty. We report results from lab-in-the-field experiments with MBA students engaged with entrepreneurial projects over a term. Participants faced binary choices involving precise-time prospects and ambiguous- time prospects that yielded the same known outcome. The primary experimental manipulations included a chance condition, where uncertainty resolution depended on a random event, and a quality condition, where it depended on project quality. The experimental design also included treatments that considered the average delay or proximity to the present (near vs. far) and the length of the time interval (short vs. long), where longer intervals represent higher ambiguity. The results show that attitudes toward ambiguous time are contingent on the mechanism that resolves uncertainty, the average delay, and the degree of ambiguity. For chance-based timing of outcomes, decision makers avoid ambiguous time. In contrast, for quality-based timing of outcomes, decision makers seek ambiguous time, and more so for outcomes in the distant future. To the best of our knowledge this is one of the few studies that focuses attitudes toward ambiguous time and provides evidence of ambiguity seeking towards time.
October 2023 - Cem Peker (NYU Abu Dhabi): “Robust recalibration of aggregate probability forecasts using meta-beliefs”
Abstract: Previous work suggests that aggregate probabilistic forecasts on a binary event are often conservative. Extremizing transformations that adjust the aggregate forecast away from the uninformed prior of 0.5 can improve calibration in many settings. However, such transformations may be problematic in decision problems where forecasters share a biased prior. In these problems, extremizing transformations can introduce further miscalibration. We develop a two-step algorithm where we first estimate the prior using each forecasters’ belief about the average forecast of others. We then transform away from this estimated prior in each forecasting problem. Evidence from experimental prediction tasks suggest that the resulting average probability forecast is robust to biased priors and improves calibration.
June 2023 - Daniel Banki (UPF) - "Trust Across Contexts".
Abstract: We investigate how well commonly used measures of trust capture trust across different contexts. In a preregistered online study, we measure trust in three different ways: using the trust game, two measures commonly used in surveys, and 14 real-life scenarios (asked hypothetically) that cover many aspects of people’s personal and professional lives. We find that the trust game is a poor predictor of how people behave in real-life scenarios. Survey measures predict behavior in individual real-life scenarios better, and their accuracy increases further when predicting a person’s average behavior across the situations. As a further demonstration that trust is context-dependent, we compare two culturally distinct countries – the US and Japan – using our measures. We find that Americans are more trusting according to survey measures and real-life scenarios – but less trusting if we base this comparison on the trust game instead. Taken together, our results suggest that not all measures of trust are created equal, and researchers need to think carefully about the measures of trust that are best suited to the contexts they are interested in. Joint work with Daniel Navarro-Martinez (Universitat Pompeu Fabra) and Shohei Yamamoto (Hitotsubashi University).
May 2023 - Simona Botti (LBS) - “What is the value of knowing in advance of an undesirable, unavoidable future?”.
Abstract: This paper examines the consequences on well-being of knowing in advance, versus not, of the occurrence of an undesirable, unavoidable future event (e.g., a genetic illness, an aversive task), as well as how these consequences explain preference for advance knowledge. We find that advance knowledge hurts psychological well-being before the event and only marginally improves it after the event; moreover, the longer the time since knowing, the more aversive the effect on well-being. Nevertheless, consumers prefer knowing, versus not knowing, in advance. This preference reveals a hyperopic tendency: consumers they overvalue the more distant and more uncertain benefits of advance knowledge relative to its more proximal and more certain detriments. We study this phenomenon with hypothetical scenarios, real data from individuals who tested for a genetic disease (Huntington’s Disease), and lab studies involving consequential decisions. Joint work with Selin Göksel (VU Amsterdam) and Nazli Gurdamar Okütur (Koç University).
April 2023 - Ploutarchos Kourtidis (LSE) - “Confirmation bias and mitigating strategies in vaccine decision making”.
Abstract: Vaccine hesitancy is a contributing factor to the (re)occurrence of vaccine-preventable diseases. Confirmation bias, people’s tendency to selectively seek or distort information in favour of their own beliefs, has been linked to vaccine hesitancy. In this research, we measure people’s (N=1,641) attitudes towards a developing vaccine against Nipah, a “priority disease” according to WHO. We also measure whether people exhibit confirmation bias when they process information related to Nipah. Finally, we test whether confirmation bias and vaccination choice are sensitive to either analytical or more intuitive corrective strategies (debiasing and nudging). In two preregistered studies we measure people’s vaccination willingness and evaluation of factual but ambiguous information about Nipah. We subsequently a) direct participants’ attention towards disconfirming information to facilitate consideration of alternative hypotheses (debiasing), or b) provide them with a subtle nudge that focuses on regret salience and social norms (nudging). We found that more 30% of our sample were hesitant towards a Nipah vaccine. Importantly, we found that vaccine-willing and vaccine-hesitant participants evaluate health information similarly, but in different directions, suggesting that confirmation bias is prevalent, and equally strong, among both groups. Only 10% of our sample revised their vaccination choice post-intervention, with vaccine-hesitant participants being significantly more likely to revise their choice than their vaccine-willing counterparts. Neither debiasing nor nudging significantly reduced bias or produced a revision in vaccination choice. We conclude that confirmation bias is associated with both vaccine hesitancy and vaccine willingness. Furthermore, while vaccination choices appear generally stable, vaccine-hesitant individuals may be more open to change their attitude in light of new information. Finally, we conclude that public health information should be communicated with caution, given that it can be misinterpreted in favour of one’s own vaccination belief.
March 2023 - Ralph Hertwig (MPI, Center for Adaptive Rationality) - “How Experimental Methods Shaped Views on Human Competence and Rationality”.
Abstract: Within just 7 years, behavioral decision research in psychology underwent a dramatic change: In 1967, Peterson and Beach (1967) reviewed more than 160 experiments concerned with people’s statistical intuitions. Invoking the metaphor of the mind as an intuitive statistician, they concluded that “probability theory and statistics can be used as the basis for psychological models that integrate and account for human performance in a wide range of inferential tasks” (p. 29). Yet in a 1974 Science article, Tversky and Kahneman rejected this conclusion, arguing that “people rely on a limited number of heuristic principles which reduce the complex tasks of assessing probabilities and predicting values to simple judgmental operations” (p. 1124). With that, they introduced the heuristics-and-biases research program, which has profoundly altered how psychology, and the behavioral sciences more generally, view the mind’s competences and rationality. How was this radical transformation possible? We examine a previously neglected driver: The heuristics-and-biases program established an experimental protocol in behavioral decision research that relied on described scenarios rather than learning and experience. We demonstrate this shift with an analysis of 604 experiments, which shows that the descriptive protocol has dominated post-1974 research. Specifically, we examine two lines of research addressed in the intuitive-statistician program (Bayesian reasoning and judgments of compound events) and two lines of research spurred by the heuristics-and-biases program (framing and anchoring and adjustment). We conclude that the focus on description at the expense of learning has profoundly shaped the influential view of the error-proneness of human cognition.
February 2023 - Peter Wakker (Erasmus School of Economics) - “The SIMPLICITY of (Non-)Discounted Welfare if You Have Unbounded Time”.
Abstract: Ramsey argued that impatience/discounting is unfair to future generations, especially if infinitely many. However, the resulting unweighted average utility (“patience”) leads to paradoxes: no Pareto optimality & Parfit’s repugnant conclusion. We introduce preference foundations to resolve these paradoxes. The common thinking in all of the literature as yet is that there are two problems here: (1): Infinite time is harder than finite. So, first handle finite and then try to extend. (2): We need continuum-richness to do our maths. But Suppes, Brouwer (my fixed-point countryman), and others showed that we don’t know its empirical/ethical meaning. The time has come for a surprise: infinite time is easier! Both problems disappear! We get a preference foundation both more general and simpler, and therefore ethically clearer, than all preceding ones in the literature.
January 2023 - Emanuele Borgonovo (Bocconi) - “Information Density: from Decision Analysis to Probabilistic Sensitivity”.
Abstract: Information value is widely studied and applied in decision analysis. We discuss and extend the notion of information density, originally introduced in Hazen 2014. We discuss existence, uniqueness, and formulation in association with alternative definitions of information value. We then discuss its formulation for reporting problems with proper scoring functions. We analyze how this extended notion blends with four statistical properties that aid the interpretation of measures of statistical association. We discuss analytical results for Gauss-linear models. We illustrate the graphical insights through a well-known realistic case study in the medical sector and a recently developed model to support planning during the COVID-19 pandemic.
Joint works with Gordon Hazen, Xuefei Lu, and Elmar Plischke.
December 2022 - Freddy Lim (INSEAD) - “Loyalty Currency and Mental Accounting: Do Consumers Treat Points Like Money?”.
Abstract: Loyalty programs have greatly expanded in scale and scope, and loyalty points issued by firms serve as a new form of currency alongside the traditional currency of money. In this paper, we study how consumers decide to pay with points or money for a purchase and how these decisions are affected by consumers’ points earning characteristics. We develop a model of consumers’ payment choices and estimate it on proprietary loyalty program data from a major U.S. airline company using a hierarchical Bayesian framework. Our results demonstrate that mental accounting, the subjective perceived value of points, and the reference exchange rate play important roles in consumers’ payment choices. Moreover, the primary points earning source and the total earning level are jointly associated with consumers’ attitudes toward points and money: Consumers who earn many points and mostly with the focal firm tend to value points more than money, while those who earn few points or mostly through a co-branded credit card tend to value money more than points. To better understand heterogeneity in consumers’ attitudes toward points, we propose a probabilistic segmentation of consumers and identify four behavioral segments with distinctive characteristics. Through counterfactual analysis, we demonstrate how a firm can implement money and point pricing policies to efficiently target and influence consumers’ payment choices.
November 2022 - Götz Giering (LOUGHBOROUGH) - "PROCESS TRACING CHOICE QUALITY IN RISKLESS MULTI-ATTRIBUTE DECISIONS."
Abstract: Empirical findings from behavioural decision research suggest that individuals employ a range of strategies to construct their preferences when faced with multi-attribute choice problems. However, it is not well understood what information decision makers use when making a multi-attribute choice, and if training decision makers can improve multi-attribute choice quality. I adopted a process-tracing approach to evaluate preference construction processes in terms of quality in two behavioural experiments. Both studies involved a set of hypothetical choice problems, and utilised normative standards based on attribute range sensitivity as benchmark to compare against observed decision processes. Choice heuristics displayed efficiency and performed at least as good as strategies based on Multi-attribute Value Theory (MAVT). This result was observed consistently, independent of the prescriptive intervention stage, across experiments. MAVT-based strategies failed to unleash their theoretical potential. The study findings suggest that level of strategy execution had higher impact on choice quality than the sophistication of the selected decision strategy.
October 2022 - Ayse Onculer (ESSEC) - "Looking a Gift Horse in the Mouth: The Dark Side of Uncertain Price Promotions."
Abstract: A growing number of brands use uncertain price promotions to improve their sales. This research introduces the “Looking a gift horse in the mouth” effect by probing uncertain price promotions as a potential liability. The results of a pretest field experiment and a series of eight studies with incentive-compatible designs reveal that when consumers receive an inferior prize in an uncertain price promotion (low-win), they question the value of the attained outcome (i.e., they “look a gift horse in the mouth”) which reduces their appraisal of the promotional brand. Receiving a low-win triggers mixed emotions which have an adverse effect on the appraisal of the promotional brand. Our studies provide both indirect and direct support for an account of counterfactual comparison and mixed emotions. Arash Talebi, Sonja Prokopec and Ayse Onculer.
September 2022 - Selin Goksel (LBS) - "Consumer - Service Provider Interactions In The Face Of Embarassement"
Abstract: What types of interactions do embarrassed consumers prefer to have with service providers? Current research guides service providers to approach their customers in a sociable manner. However, in contrast, we find that embarrassed consumers shy away from sociable interactions. Specifically, we find that the more consumers feel embarrassed, the less likely they are to prefer interacting with providers who adopt sociable communication styles and the less likely they are to prefer utilizing communication media that facilitate sociable interactions with service providers. Focusing more specifically on the medical domain, we further show that healthcare providers do not fully account for this distaste and provide a more sociable interaction than what patients desire. Finally, based on our findings, we provide a series of managerial recommendations to guide service providers in how to best adjust their offerings to a client base that contain a mix of embarrassed and non-embarrassed patients. Selin Goksel, London Business School;Sydney E. Scott, Washington University in St. Louis; Jonathan Z. Berman, London Business School.
June 2022 - Sarat-Chandra Akella (HEC) - "Ambiguity Attitude: Utility vs Probability Weighting"
Abstract: An individual’s ambiguity attitude characterizes her decision-making in situations where uncertainty cannot be directly represented with probabilities. In two-stage models such as the smooth ambiguity model, expected utility (EU) applied recursively can generate non-neutral ambiguity attitudes through the utility scale, if the transformation function is non-linear. However, if EU is violated in practice, these models lose descriptive power. We design an experiment to elicit a generalized version of the smooth ambiguity model where both utility and probability weighting can generate ambiguity attitude. Preliminary results at the aggregate level indicate that the utility scale exhibits greater descriptive power in the first stage, determining ambiguity attitude. However, probability weighting remains significant in the second stage.
May 2022 - Aurelien Baillon (Erasmus School of Economics) - "Source Theory: A Tractable and Positive Ambiguity Theory"
Abstract: This paper introduces source theory, the first ambiguity theory that is tractable and empirically grounded. Unlike most current ambiguity theories, it does not involve multistage gambles (whose evaluation is problematic under nonexpected utility) or expected utility for risk. To obtain tractability and predictive power, it neither involves high-dimensional subjective parameters such as sets of priors, second-order probabilities, or nonadditive set functions, by using (and axiomatizing) sources that are uniformly ambiguous. We can thus analyze ambiguity aversion generally and empirically more realistically than done before. We introduce matching partitions as a new and powerful tool for analyzing uncertainty attitudes in transparent manners. Thus, we fully formalize and axiomatize insensitivity under ambiguity (similar to perception), a prevailing empirical finding. We axiomatize within-subject between-source probability weighting transformations that are the analogs of Pratt-Arrow utility transformations, both for aversion and for insensitivity. (Joint work with Han Bleichrodt, Chen Li, and Peter P. Wakker).
April 2022 - Pranadharthiharan Narayanan (IE) - "Group decision making with a minority stakeholder"
Abstract: In many real-world environments, that involve multiple stakeholders, minority preferences matter. In these contexts, simple decision rules such as majority voting typically fail to attend to the diversity of stakeholder preferences. We propose a prescriptive group diversity rule that measures the similarity of stakeholder preferences by comparing the perceived relative importance assigned to attributes. In order to account for minority views, our rule assigns the highest utility to those alternatives that maximize the diversity of stakeholder preferences satisfied. We study the properties of this rule in simulated decision environments and benchmark its empirical performance in two experimental studies involving a hiring scenario as well as a radioactive waste disposal problem. (Joint work with Enrico Diecidue and Matthias Seifert).
March 2022 - Rahil Hossein (UPF) - "The scale Effect: How Rating Scales Affect Product Evaluation"
Abstract: Review websites rely on different scales, such as the 5-point scale, the 10-point scale, or the 100-point scale. How do consumers aggregate ratings from sources that use different scales as they form purchase intentions and make consumption decisions? In 8 studies (N = 2,976), we found that ratings expressed on larger scales have a stronger effect on product evaluations than ratings expressed on smaller scales. We call this phenomenon the ‘scale effect.’ This finding is surprising in light of the ubiquity of the 5-point scale and research on the role of fluency in information processing. The scale effect results from a combination of deliberate and non-deliberate cognitive processes. Evidence suggests that, by default, people tend to give more weight to large-scale ratings, partly because they find them more informative. People can overcome this default aggregation strategy if there exist reasons to give more weight to small-scale ratings. (Joint work with Gaël Le Mens).
February 2022 - Francis de Vericourt (ESMT) - "Is your machine better than you? You may never know"
Abstract: AI systems are increasingly demonstrating their capacity to make better predictions than human experts. Yet, recent empirical studies suggest that professionals sometimes doubt the quality of these systems, and as a result overrule machine-based prescriptions. This paper explores the extent to which a decision maker (DM) can properly assess whether a machine produces better recommendations. To that end, we analyze an elementary dynamic Bayesian framework, in which a machine performs repeated decision tasks under a DM’s supervision. The task consists in deciding whether to take an action or not. Crucially, the DM observes the accuracy of the machine’s prediction on the task only if she ultimately takes the action. As she observes the machine’s accuracy, the DM updates her belief about whether the machine’s predictions outperform her own. Depending on this belief, however, the DM sometimes overrides the machine, which affect her ability to assess it.
In this set-up, we characterize the evolution of the DM's belief and overruling decisions over time. We identify situations under which the DM’s belief oscillates forever, i.e., the DM always hesitates whether the machine is better. In this case, the DM never fully ignores the machine but regularly overrules it. We further find that the DM’s belief sometimes converges to a Bernoulli random variable, i.e., the DM ends up wrongly believing that the machine is better (or worse) with positive probability. We fully characterize the conditions under which these failures to learn occur. These results highlight some fundamental limitations in our ability to determine whether machines make better decision than experts. They further provide a novel explanation for why humans may collaborate with machines – even when one may actually outperform the other. Joint work with Huseyin Gurkan (ESMT).