关键词: 注意, 价值决策, 证据积累, 加法机制, 乘法机制

Abstract: In value-based decision making, individuals must sample, compare, and integrate information from different options and their attributes under constraints of limited time and cognitive resources. As a critical mechanism linking information processing to choice formation, attention has become central to understanding how evidence is accumulated during decision making. Existing accounts have proposed two major computational explanations. The multiplicative model suggests that attention enhances the weight of attended information, thereby amplifying its contribution to evidence accumulation. In contrast, the additive model proposes that attention provides an additional directional bias toward the currently attended option, exerting an influence that is relatively independent of its subjective value. However, these two mechanisms have often been conceptualized as competing explanations, making it difficult to account for inconsistent attentional effects observed across different experimental contexts. Here, we argue that attentional processing in value-based decision making is inherently dynamic and hierarchical. Rather than exerting a fixed computational influence throughout the decision process, attention may serve different computational functions depending on the progression of decision formation and the level of information representation being processed. By systematically integrating evidence from behavioral studies, computational modeling, and neuroscience, we propose a “time × information hierarchy” framework to explain how attention dynamically regulates evidence accumulation through two complementary computational principles: value amplification and directional bias. Specifically, we propose that attentional influences on value-based decisions do not rely on a single mechanism, but instead exhibit a gradual transition from multiplicative to additive processes over the course of decision formation. During the early stage of decision making, individuals face substantial uncertainty and engage in broad information sampling and value comparison across multiple attributes and options. At this stage, attention primarily operates on lower-level information representations, including attribute-level features (e.g., price, reward, and health value) and option-level subjective value representations. By increasing the relative weight of relevant value signals entering the evidence accumulation process, attention enhances the impact of high-value or goal-relevant information, consistent with a multiplicative value-gain mechanism. This account aligns with the attentional drift diffusion model (aDDM), which proposes that attention influences choice by modulating the relative contribution of attended and unattended information during evidence accumulation. As decision formation progresses, uncertainty within the decision system gradually decreases, and information processing shifts from broad exploration toward focused comparison among dominant options. We propose an intermediate filtering stage to characterize this transition from open-ended information sampling to choice convergence. Importantly, this stage does not represent a fixed temporal window or an abrupt switch between two mechanisms. Instead, it reflects a dynamic computational state in which both mechanisms may coexist. During this period, multiplicative value amplification continues to contribute to evidence accumulation, while attentional effects increasingly involve directional reinforcement toward currently favored options. During the late stage of decision making, when accumulated evidence approaches the decision threshold, processing priorities shift from value comparison toward evidence integration and choice commitment. At this stage, attentional effects are more likely to exhibit an additive pattern. Rather than continuously amplifying the strength of value signals, attention may provide an additional directional bias that facilitates the convergence of evidence toward a particular choice and supports final decision commitment. Importantly, we further emphasize that late-stage additive effects should not be attributed exclusively to attention itself. Such effects may also reflect the joint contribution of attentional allocation, decision commitment, action preparation, and urgency-related processes. Therefore, late-stage attentional influences should be understood as the outcome of interactions among attentional mechanisms, evolving decision states, and behavioral output processes. The theoretical contribution of this framework does not lie in proposing the first computational integration of multiplicative and additive mechanisms. Existing models, such as the Gaze-weighted Linear Accumulator Model (GLAM), have already incorporated both value weighting and directional bias components. Instead, the present framework advances existing theories by explaining why these two mechanisms may dominate under different conditions through a temporal and hierarchical perspective. We propose that multiplicative and additive mechanisms are not mutually exclusive alternatives, but rather represent distinct computational functions emerging at different stages and levels of information processing within the same decision system. By establishing the “time × information hierarchy” framework, this review provides a unified account of existing behavioral, computational, and neural findings and offers a new perspective for understanding how attention dynamically shapes value representation, evidence accumulation, and choice formation.

Key words: attention, value-based decision-making, evidence accumulation, additive mechanism, multiplicative mechanism