The standard relationship between choice frequency and choice time is violated in multi-attribute preferential choice

Abstract

Many decision making theories assume a principle of sequentially sampling decision-relevant evidence from the stimulus environment, where sampled evidence is dynamically accumulated toward a threshold to trigger a decision in favour of the threshold-crossing option. A core prediction of sequential sampling models is that options more likely to be chosen are chosen more quickly. This result has been empirically supported hundreds of times for low-level speeded perceptual decisions — the traditional domain of sequential sampling models. More recently, sequential sampling models have been generalised and applied to higher-level preferential, or value-based, decisions — decisions for which there is no objectively correct option. Preferential options are typically composed of multiple attributes, like a phone defined by its price, camera quality, memory capacity, and so on. Here, we show that decisions for such multi-attribute preferential options with defined features violate the core prediction of sequential sampling models: options more likely to be chosen are not chosen more quickly. We find this invariance across 4 data sets spanning multi-attribute choices made in unconstrained conditions, under time pressure, and for multi-attribute options with artificial or marketplace compositions. The result remains whether the relationship between choice frequency and choice time is inspected at the lower level of component attributes or the higher level of whole options. Our finding places critical constraints on the capacity to generalise sequential sampling models from low-level perceptual decisions to high-level multi-attribute preferential choice.