Neural mechanisms of attention, not expectation, govern spatial selection by probabilistic cueing

Abstract

Spatial probabilistic "Posner" cueing is widely employed in studies of endogenous spatial attention. Such cueing guides attention by providing prior knowledge about the likely spatial location of task-relevant events. Yet, it has been compellingly argued that such spatial priors also elicit expectation effects, rendering Posner cueing unsuitable for measuring attentional effects in isolation. We address this debate by combining signal detection theory models of behavior with concurrent electrophysiological recordings, and directly compare Posner cueing effects with those of attention and expectation cueing. Participants performed two tasks: a dual cueing task, with orthogonal relevance (attention) and probability (expectation) cues, as well as a Posner cueing task. Relevance and probability cueing independently modulated distinct behavioral parameters - perceptual sensitivity and decisional criterion, respectively - whereas Posner cueing modulated both sensitivity and criterion. However, only sensitivity modulations by Posner cueing were correlated with those of relevance cueing. Criterion modulations by Posner cueing were uncorrelated with those of probability cueing. Both Posner and relevance cueing, but not probability cueing, modulated various neural markers of spatial attention, including steady-state visually evoked potential (SSVEP) amplitude and alpha-band (8-12 Hz) oscillation power. Representational similarity analysis and cue-label prediction with deep convolutional neural networks revealed dissociable underpinnings of relevance and probability cueing, and identified Posner cueings neural representations with those of relevance cueing. Our results address a long-standing debate in the attention literature and clearly demonstrate that spatial selection by probabilistic cueing is governed by neural mechanisms of attention, not expectation. Significance StatementHow does our brain select relevant information for accomplishing task goals? "Posner cueing" is a popular choice for studying brain mechanisms of goal-driven attention in the lab. Posner cues guide attention by providing advance information about the likely location of task-relevant events. Yet, Posner cueing has been criticized because it engages brain mechanisms not only of attention but of event expectation, as well. We address this critique by employing state-of-the-art neural decoding approaches, including deep convolutional neural networks, and show that behavioral and neural underpinnings of Posner cueing closely match those of attention, not expectation. The results address a long-standing debate in the attention literature and validate Posner cueing as a reliable method for measuring attentions effects in the brain.