Perceptual decisions often occur under uncertainty, where rapid and accurate responses are critical. While unisensory paradigms provide foundations, real-world perception is inherently multisensory. A central marker is the redundant signals effect (RSE): where responses are faster to multisensory stimuli than their unisensory components. The mechanisms driving the RSE remain debated, often framed as either cooperation (interaction) or competition (independence). This thesis advances understanding through behavioural experiments and computational modelling. First, using a novel dataset (N=21), I examined semantic congruency by equalising redundancy across conditions, a key innovation of this study. Congruent audiovisual signals produced larger multisensory speed-ups than incongruent ones, a modulation effect that cannot be explained by a competition account. Second, I investigate trial sequence effects in a new trimodal experiment (N=21). Here, the incorporation of tactile stimuli allowed all compared stimuli to be assessed following a full modality switch. Sequential dependencies shape multisensory response times, as predicted by a pure competition account, but do not modulate the magnitude of the RSE nor race model violations, challenging claims that violations may arise from sequence effects alone. Finally, using foundational archival data, I introduce the relay model, a novel two-stage account in which unisensory processes first compete in a race before initiating a second race from cross-modal interaction. This hybrid architecture captures key distributional properties of reaction times and offers a mechanistic explanation of how cooperation and competition jointly shape multisensory responses. Overall, the findings argue against purely cooperative or competitive models, instead supporting a hybrid view where multisensory decisions emerge from dynamic interplay between senses. Beyond clarifying RSE mechanisms, this work highlights the roles of congruency, sequential context, and multi-stage dynamics in perceptual decisions. More broadly, it demonstrates the value of computational modelling for linking behaviour to underlying mechanisms, with implications for understanding variability in multisensory perception.
Kalvin Edward Roberts (Fri,) studied this question.