Altman, M.
Universidad del Pais Vasco / BCBL
How do we decide what to do next? There is considerable evidence that we benefit from statistical regularities in environmental inputs. But how we encode the past to anticipate upcoming events remains a topic of debate. Do we compute these statistics as ideal observers, or rely on heuristics that provide effective guides in a partially observable world? To investigate these possibilities, we use a 2-alternative eyetracking paradigm examining eye movements to images in series governed by varying transitional probabilities for a return to the same side. We also introduce a new eye measure, fixation bias (FB), which enables us to quantify small deviations in eye position after return to central fixation. FB proves to be a highly sensitive measure of spatial bias that is not influenced by image semantics, inhibition of return (IOR) or the alternation advantage; by contrast, all these factors significantly impact saccade response times (SRTs). Using FB, we show that an extended history of >10 previous trials predicts spatial bias on the current trial, while sensitivity to the transitional matrix governing image locations plays no significant additional role. Our results have both methodological and theoretical implications. First, we show that FB provides a more accurate and precise measure of spatial bias than SRTs, offering a useful additional metric for eyetracking paradigms. Second, our results show that the locus of attention is highly sensitive to the location of recent events: previous trial effects more simply explain behavior apparently consistent with statistical learning. Further research should examine responses to transitional probabilities over longer time frames, and how quickly eye behaviors adapt to changes in the transitional matrix governing an event series. But for short time periods, we have shown that simple heuristics based on what just happened provide efficient adaptations in a dynamic environment.