Context & Consistency

The ‘attraction effect’ or ‘asymmetric dominance effect’ is a widely studied decision-making phenomenon, challenging the regularity principle in rational choice theory. It posits that introducing a third option, similar but inferior to one of the available options in a binary choice set, increases the choice share of the dominating option. While the standard attraction effect is pervasive, researchers have identified conditions where the effect is absent or reversed. One such condition is a strong prior trade-off, indicating a biased preference for one of the choices in the binary set, which mitigates the attraction effect. We suggest that measurements of the attraction effect are likely confounded by differences in experimenter-assumed indifference curves (EAIC) and the true subjective indifference curves (SIC) of experiment participants, which is what produces the strong prior trade-off. We show via simulations that in the presence of this baseline bias, the triplet-triplet design that has been extensively used in the literature would produce false alarms, i.e., the measures can show context effects even in the absence of any. We also artificially introduce context effects in our simulation and show that even in the presence of violations of regularity, the measures would give misses. These false alarms and misses result simply because the two decoys placed on the attribute space in the triplet-triplet design are no longer symmetrical to the SICs. Finally, we offer guidelines to prevent such errors in future studies.

This project evaluates changes in risk-seeking behavior across multi-stage decision tasks. Decision-makers were presented with a two-stage decision tree. At each stage, participants chose between a sure reward or a risky gamble with equal expected value. Participants who chose the gamble option continued to the second decision stage, where they again chose between a sure reward or a gamble. Rational choice theories require that behavior at the second stage be consequentialist, meaning that choices should only depend on the possible future outcomes. According to this view, participants should display a consistent risk preference across all consequentially equivalent decision nodes. However, our findings indicate widespread violations of consequential consistency, with past events (i.e. the outcomes of earlier, but consequentially irrelevant chance events) significantly affecting participants’ willingness to gamble at the second decision stage. Specifically, participants made riskier choices after experiencing good outcome than they did after bad outcome. We examine two alternative explanations for observed inconsistencies: one based on changing subjective probabilities in line with the Gambler’s Fallacy, and another based on shifting reference points.

Although most research into risky decision making has focused on simple scenarios – where isolated choices are made independent of one another – many important decisions in life play out across sequences of interdependent events and actions. Despite the ubiquity and importance of such decision problems, we know relatively little about how people manage the complexities of dynamic, multistage decisions. Our work combines techniques from two lines of research to investigate how people handle the challenges of dynamic decision making (DDM). From the econometric tradition we rely on a family of truth-and-error models that can be used to estimate the distribution and stability of preference profiles, and the presence of errors. In a complementary analysis, we use cognitive modeling to investigate the psychological processes underlying DDM. Decision Field Theory-Planning provides a unified framework for testing competing hypotheses about how people collect information and plan for the future. Results from both sets of analyses identify distinct groups of individuals. We discuss the behavioral and cognitive factors distinguishing groups from one another, including degree of planning, biased information sampling, and strategy shifts. We highlight the strengths and weaknesses of each modeling approach, and examine where their results support conflicting conclusions.

Modelling the role of experience in the formation of preferences is an increasingly popular topic in numerous research areas. Thus far, however, there have been limited applications using physiological sensor data to unpick the role of experience in preferential change under experimental conditions, a particularly challenging task in real-world settings where the effect of an experience can be confounded with changes in other external factors. This motivates the work in the present study, where we design a novel virtual reality (VR)--based data collection process that allows us to collect physiological sensor data to measure the effect of experience in a controlled setting. It also allows us to collect data on hypothetical futuristic scenarios where preferences may be more subject to change. Specifically, we ask participants to complete stated preference (SP) tasks on travel mode choice. After each SP choice, the participant `experiences' their chosen mode in Virtual Reality (VR), where the level-of-service attributes (e.g. travel time, waiting time, level of comfort, etc.) are mapped with those presented in the SP. They are then asked to reevaluate their choices. Electroencephalogram recordings, eye-tracking data, and skin conductance data are recorded to (a) better understand the initial decision-making process, (b) better evaluate the participant's experience within the virtual reality settings and (c) better predict preferential change. We develop different versions of decision field theory to evaluate how to best capture the impact of experiencing the chosen travel mode. This also allows us to test models for their capabilities in predicting preference reversal.