Abstract
In reproducing the orientation of a visual stimulus, perceptual decisions are systematically biased towards previous stimuli, a signature of serial dependence. This bias depends on whether previous stimuli were attended, but can also be modulated by uncertainty. For instance, cardinal orientations are represented with less uncertainty than obliques, leading to well-known orientation anisotropies. Often, anisotropies are residualized before the analysis of serial dependence. Here we show that this practice may be detrimental as it can camouflage differences between conditions. We presented sequences of relevant and irrelevant Gabors, intermixed in single trials. Participants performed a dual-task. During the sequence, they had to quickly detect a Gabor of the same colour as the fixation spot (target), while ignoring Gabors of an irrelevant colour (non-target). When the sequence ended, they had to reproduce the orientation of the last Gabor, which was always of the target colour. We found that adjustment responses were more variable when the Gabor preceding the last one (e.g., the inducer) was also of the target colour, potentially indicating an increase in response uncertainty due to the dual-task required by targets. Overall, target and non-target inducers caused comparable serial dependence, revealing an unexpected immunity of this bias to attention. However, by separately considering serial dependence for obliques and cardinal orientations, we found a clear dissociation. Only when the last Gabor was preceded by a non-target cardinal inducer, perceptual decisions were strongly biased toward cardinal orientations, likely reflecting anchoring effects triggered by response uncertainty. For obliques, serial dependence was only evident after target inducers. Interestingly, by removing orientation anisotropies before the analysis, this seemingly important component of perceptual decisions was largely altered.