Mechanisms of Context Effects in Multi-alternative Decisions

Abstract

Decision-making is one of the most fundamental cognitive processes that is susceptible to systematic biases. Context effects are a well-known family of these biases, wherein the presence of an irrelevant alternative influences preferences. This irrelevant alternative is called a "distractor" in uni-attribute decisions and a "decoy" in multi-attribute decisions. These context effects suggest that preferences are not fixed but dynamically shaped by the composition of the choice set. Thus, they challenge the Independence of Irrelevant Alternatives (IIA) principle of rational choice theory, which states that preference between two high-value alternatives should remain intact, regardless of the presence of a third alternative. Despite decades of research, the mechanisms underlying context effects remain unclear, partly due to their complexity and inconsistent findings. This thesis addresses this issue via two approaches. First, we conducted carefully designed value-learning experiments to distinguish the roles of immediate vs. temporal contexts in shaping distractor effects. Second, we used perceptual rather than value-based stimuli to test the robustness and generalizability of context effects in uni- and multi-attribute decisions. The contribution of immediate vs. temporal contexts on the emergence of the distractor effect has still been debated. To address this, we developed a value-learning paradigm in which participants first learned the value of three alternatives in each context, then reported their subjective value, and finally made binary and ternary choices. Crucially, subjective value reports allowed us to identify whether distractor effects emerged during the learning phase or at the moment of choice. Results revealed no consistent distractor effect in final choices; instead, value distortions appeared in subjective values during learning. Our findings suggest that during value-learning, the value of a distractor alternative distorts value representations over a long timescale spanning several learning and choice trials. Importantly, divisive and range normalization fail to capture the direction and across-participant variability of this distortion. Our data is best captured by a mechanism that constructs subjective value representations via a series of binary comparisons between pairs of alternatives in each context. These findings revealed new information about the timescale and computational mechanisms underlying context-dependent, uni-attribute, multi-alternative choices. Next, we tested the replicability of these distractor effects in perceptual decision-making tasks where all value information was available at the moment of choice. This removed the influence of prior learning. Results from two different perceptual tasks showed no distractor effect on choice, but decision time was prolonged when a high-value distractor was present. Furthermore, analyses of pupil dilation revealed that internal arousal states modulate the direction of the distractor effect on choice. At low arousal, distractors tend to have negative effects. At high arousal, however, the effects of distractors shift toward positive. Current static, context-dependent models cannot intrinsically explain the distractor effect on reaction times. To explain these findings, we proposed a dynamic model that incorporates valuation and decision stages. The model uses a combination of normalization theory and binary comparison, which is based on decision-by-sampling theory, to calculate subjective value in the valuation stage. Normalization theory explains negative distractor effects, while binary comparison captures positive ones. Thus, the combination of these two strategies can account for the null distractor effect on choice. Next, the subjective value is fed into a race model, which ultimately determines reaction time (RT) and accounts for the distractor effect on RT. This framework can also explain arousal findings. The negative distractor effect in low-level arousal may result from the greater influence of normalization on value representation, while the positive distractor effect in higher arousal may result from greater binary comparisons. Overall, we concluded that distractor alternatives influence the decision-making process. This influence may affect either the ultimate choice, as reported in previous studies, or the response time. Beyond distractor effects, we examined the robustness and relationships among classic decoy effects (attraction, compromise, and similarity) using a novel, three-alternative, perceptual task with distinct attributes. The results revealed strong attraction and compromise effects, but no consistent similarity effect. Notably, attraction and compromise were positively correlated, and both were negatively correlated with similarity. This suggests that they may have partly distinct underlying processes. Pupillometry again demonstrated that higher arousal strengthened overall context biases. Together, these findings highlight that decision-making depends on both external choice structures and internal physiological states (e.g., arousal state), the latter of which can amplify or redirect context effects. These findings open new directions for computational models that integrate contextual and physiological factors to better capture the variability of human decision-making.