Cue combination occurs when two independent noisy perceptual estimates are merged together as a weighted average, creating a unified estimate that is more precise than either single estimate alone. Surprisingly, this effect has not been demonstrated compellingly in children under the age of 10 years, in contrast with the array of other multisensory skills that children show even in infancy. Instead, across a wide variety of studies, precision with both cues is no better than the best single cue - and sometimes worse. Here we provide the first consistent evidence of cue combination in children from 7 to 10 years old. Across three experiments, participants showed evidence of a bimodal precision advantage (Experiments 1a and 1b) and the majority were best-fit by a combining model (Experiment 2). The task was to localize a target horizontally with a binaural audio cue and a noisy visual cue in immersive virtual reality. Feedback was given as well, which could both (a) help participants judge how reliable each cue is and (b) help correct between-cue biases that might prevent cue combination. Crucially, our results show cue combination when feedback is only given on single cues - therefore, combination itself was not a strategy learned via feedback. We suggest that children at 7-10 years old are capable of cue combination in principle, but must have sufficient representations of reliabilities and biases in their own perceptual estimates as relevant to the task, which can be facilitated through task-specific feedback.
Audio-visual, Bayesian, Cue combination, Decision making, Perceptual algorithms, Virtual reality