Reward expectation yields distinct effects on sensory processing and decision making in the human brain

The study investigates whether reward expectation influences sensory processing (spatial attention) and decision-making (criterion biases) via shared or distinct mechanisms. Prior work often conflated these effects because the same high-value locations were both perceptually prioritized and decisionally favored. Using signal detection theory (SDT), sensory prioritization is indexed by sensitivity (d′) and decisional prioritization by choice criterion (c). The authors designed a task to decouple reward effects across space versus choices: space-specific reward expectation should modulate sensitivity at locations of higher expected reward; choice-specific reward expectation should modulate decision criterion without altering sensitivity. The purpose is to identify behavioral dissociations and corresponding neural signatures (ERPs, alpha oscillations, microsaccades) of these two forms of reward expectation, clarifying how reward, attention, and choice interact in the human brain. The importance lies in resolving controversies about whether reward-driven attention and decision-making share mechanisms, and in validating attention markers as tied to sensory selection rather than decisional biases.

Prior studies show reward expectation speeds reactions and improves detection, engaging prefrontal and parietal areas (e.g., PFC, LIP). However, many designs failed to separate perceptual sensitivity changes from decision criterion shifts or motor preparation. Monkey studies manipulating relative reward across locations found LIP activity tracking reward values or motivational salience but did not decouple perceptual from decisional consequences. More recent SDT-based monkey work dissociated sensitivity and criterion by manipulating relative rewards across locations or absolute rewards across choices; overlapping PFC populations encoded both, while V4 correlated with sensitivity but not criterion. Another monkey study using probabilistic cueing and anti-saccades showed V4 firing modulated by criterion rather than sensitivity, hypothesizing not all criterion manipulations reflect spatial attention. The present work directly tests this hypothesis in humans, with EEG and eye-tracking, and introduces a fixed reward side to compare attentional allocation across hemifields under space- versus choice-specific reward manipulations.

Participants: 24 healthy volunteers (9 females; age 20–37, mean 26.1), normal or corrected vision; ethics approval from IISc IHEC (4-20182020). Staircasing and training over three days; main experiments on two days with concurrent EEG, eye tracking, and GSR (GSR not analyzed). Task: Two-alternative change detection using two Gabor gratings (one per hemifield). Trial: fixation (1000 ms); gratings plus central reward cue (300–700 ms); blank (200 ms); gratings reappear (200 ms) with independent 50% change probability per side; response probe indicates which side to report Yes/No change within 1500 ms; audiovisual feedback displays current reward/penalty and cumulative score. Orientation change magnitude individualized via staircase to ~69–70% accuracy (range 6–32°, mean 16°). Sessions: Two reward-cueing types—space-specific and choice-specific. Each session: 12 blocks × 48 trials (total 1,152 trials/participant across sessions), with 6 gain blocks (rewards for correct responses) and 6 loss blocks (penalties for incorrect responses). Fixed (FX) side: contingency constant within a block; Variable (VR) side: contingency switches across mini-blocks (10–16 trials, exponential distribution). Cue: colors indicate FX vs VR side; mapping counterbalanced. Space-specific session: equal rewards for hits and correct rejections on each side, but average reward differs across sides within mini-blocks (VR>FX or VR<FX). Loss blocks mirror penalties for incorrect responses. Choice-specific session: equal rewards on FX; on VR, unequal rewards for hits vs correct rejections in alternating mini-blocks (liberal: Yes>No; conservative: Yes<No); loss blocks mirror penalties (FA vs Miss). Participants infer switches from feedback. Responses via keypress (orthogonal mapping to space), not saccades. Eye-tracking: Eyelink 1000 Plus at 1000 Hz; trials rejected for gaze deviations in defined windows; monocular tracking used; microsaccades detected via Engbert-Kliegl algorithm with velocity thresholds, temporal proximity limits, and outlier exclusion. EEG: 128-channel HydroCel net, Cz reference, 1000 Hz, impedances typically <30–50 kΩ; preprocessing with FieldTrip: 0.5–35 Hz IIR filter, downsample 250 Hz, ICA artifact removal, SCADS channel/epoch rejection, interpolation, average reference, epoching -200 to +700 ms around stimulus for ERPs. ERPs: N2pc and P300 from occipitoparietal electrodes (PO3/4, PO7/8, O1/2), quantified at 150–210 ms (N2pc) and 230–480 ms (P300). P2a from frontocentral electrodes quantified 150–210 ms. Alpha power: Chronux multitaper spectrograms from -1000 to +1000 ms; individual alpha frequency (IAF) 7.8–12.2 Hz (mean 10.2 Hz); post-stimulus alpha power measured 450–950 ms; pre-stimulus 500 ms before onset; lateralization indexed as suppression differences contralateral to stimuli across contingencies. Statistics: Three-way and two-way ANOVAs (side × reward contingency × block valence), Wilcoxon signed-rank tests, Bayes factors (JZS priors), cluster-based permutation tests for microsaccade time-series, permutation tests for regression coefficients and R² comparisons. Conserved resource analysis: parameters and markers plotted around inferred contingency switch trials; assess cross-hemifield coupling by fitting de-meaned FX vs VR values to x+y=0 and comparing R² across session types. Regression/prediction: Multiple linear regression of Δd′ or Δc from neural markers (ΔN2pc, ΔP300, Δα_pre, Δα_post) or motoric markers (ΔMSC, ΔRT); leave-one-out predictions assessed via robust correlations.

Behavioral dissociation (n=24): • Space-specific reward expectation selectively modulated sensitivity (d′) at the higher-reward location, not criterion. Sensitivity modulation difference between contingencies: Δd′FX = −0.60 ± 0.12, Δd′VR = 0.43 ± 0.09; Signed rank p < 0.001, BF₁₀ > 10³. Criteria not significantly modulated: ΔcFX = 0.01 ± 0.09, ΔcVR = 0.10 ± 0.08; p = 0.511, BF₁₀ = 0.24. ANOVA: side × reward contingency interaction significant for d′ (F1,23=30.88, p<0.001); no effects for c. • Choice-specific reward expectation selectively modulated criterion at VR, not sensitivity. Criterion modulation: ΔcFX = −0.07 ± 0.05, ΔcVR = −0.46 ± 0.08; p < 0.001, BF₁₀ > 10³; absolute criterion modulation greater at VR (p<0.001). Sensitivity modulation not significant: Δd′FX = 0.05 ± 0.09, Δd′VR = 0.15 ± 0.10; p = 0.391, BF₁₀ = 0.29. ANOVA: criterion main effect of reward contingency (F1,23=24.86, p<0.001) and interaction (F1,23=21.26, p<0.001); d′ main effect of side (F1,23=6.51, p=0.018). Gain vs loss blocks showed virtually identical patterns; pooled analyses matched. Δd′ and Δc modulations were uncorrelated in both sessions (ρ=0.10, p=0.237; ρ=−0.03, p=0.454). Neural attention markers engaged only by space-specific expectation: • ERPs: N2pc modulation larger contralateral to VR than FX with space-specific (ΔN2pcFX = −0.05 ± 0.08 µV, ΔN2pcVR = 0.14 ± 0.08 µV; p=0.018, BF₁₀=8.22); no difference with choice-specific (ΔN2pcFX=0.10±0.11 µV, ΔN2pcVR=0.13±0.15 µV; p=0.753). P300 modulation larger at VR only with space-specific (ΔP300FX = −0.16 ± 0.09 µV, ΔP300VR = 0.13 ± 0.09 µV; p=0.002, BF₁₀=12.96); not with choice-specific (ΔP300FX=0.00±0.13 µV, ΔP300VR=0.09±0.18 µV; p=0.549). • Alpha-band: Post-stimulus alpha lateralization only with space-specific—stronger contralateral suppression at VR vs FX (ΔαFX = 0.83 ± 0.35, ΔαVR = −0.79 ± 0.35; p < 0.001, BF₁₀=25.21). Choice-specific showed no lateralized post-stimulus alpha modulation (ΔαFX = −0.42 ± 0.41, ΔαVR = −0.17 ± 0.32; p=0.753). Pre-stimulus alpha suppression modulated only by choice-specific reward expectation (ΔαFX = 1.52 ± 0.94, ΔαVR = −0.66 ± 1.38; p=0.049), consistent with decisional bias/excitability. • P2a showed no differential modulation (supporting figs/tables). Motoric attention markers: • Microsaccades: Space-specific increased MSC toward higher reward—ΔMSCFX = −0.04 ± 0.02, ΔMSCVR = 0.04 ± 0.01; p < 0.001, BF₀=21.02; significant clusters 150–350 ms post-cue. Choice-specific showed no MSC modulation (ΔMSCFX=0.00±0.01, ΔMSCVR=−0.01±0.01; p=0.224). • Reaction times: Space-specific faster RTs at higher reward—ΔRTFX = 34.14 ± 9.24 ms, ΔRTVR = −19.60 ± 5.97 ms; p < 0.001, BF₀>10³. Choice-specific RT modulation across locations not significant (ΔRTFX = −1.46 ± 6.28 ms, ΔRTVR = 3.97 ± 5.67 ms; p=0.587), but RTs were faster for high-reward (low-penalty) choices at VR (δRT = −24 ± 9 ms; p=0.009, BF₀=8.88). Conserved attentional resource: • Only space-specific sensitivity showed cross-hemifield competition: FX side modulation Δd′FX = −0.51 ± 0.08; p<0.001, BF−₀>10⁴; FX and VR values fit x+y=0 with higher R² than choice-specific. Analogous competitive patterns for post-stimulus alpha (Δα_postFX=0.83±0.35; p=0.037), RT (ΔRT=34.14±9.24; p<0.001), and MSC (ΔMSC=−0.04±0.02; p=0.003) in space-specific, but not in choice-specific. Pre-stimulus alpha showed no such competitive effect. Predictive links: • In space-specific sessions, Δd′ was explained by neural markers (R²=0.34; F(4,43)=5.54; p=0.001), with significant contribution of post-stimulus alpha (β=−0.16; p=0.001), and by motoric markers (R²=0.30; F(2,45)=9.56; p<0.001) with RT (β=−5.15; p=0.003) and MSC (β=1.99; p=0.045). Leave-one-out predictions: neural r=0.399 (p=0.004), motoric r=0.602 (p<0.001). • In choice-specific sessions, Δc was not reliably predicted by neural (R²=0.06; p=0.602) or motoric (R²=0.04; p=0.406) markers; leave-one-out predictions near zero (neural r=−0.017; motoric r=−0.053). Unexpected attentional allocation: In choice-specific sessions, despite equal average rewards across sides, attention-related markers and d′ tended to favor the FX side (e.g., lower RTs, higher MSC, ERP trends), suggesting that criterion shifts reflect response-planning strategies rather than spatial attention deployment.

The findings demonstrate a double dissociation: varying reward expectation across space selectively increases sensitivity (d′) and engages canonical spatial attention markers (N2pc, P300, contralateral alpha suppression, microsaccade biases, faster RTs), whereas varying reward expectation across choices selectively shifts decision criterion without engaging these attention markers. This resolves a longstanding conflation between reward-driven sensory prioritization and decisional biases, showing they are mediated by distinct neural mechanisms. Space-specific modulations reflect a conserved, competitive attentional resource across hemifields, consistent with attention as limited resource allocating sensory gain. Choice-specific modulations align with response bias or strategy (planning Yes/No based on contingencies), supported by faster RTs for more rewarded choices and pre-stimulus alpha suppression indexing baseline excitability, rather than spatial attention. The results clarify that not all criterion shifts are attentional; criterion changes due to reward-contingent response biases are separable from attention, whereas criterion effects linked to spatial choice biases in localization tasks may still reflect attention components. The study also suggests that attention-related EEG and oculomotor markers are tied to sensory selection (d′) rather than reward expectation per se, as they tracked the location with higher sensitivity even when average rewards were equal, highlighting nuanced interactions between reward, uncertainty, and cognitive demands.

This work introduces a task that decouples reward expectation effects on spatial attention versus decision-making and identifies their distinct behavioral and neural signatures in humans. Space-specific reward variations modulate sensitivity and recruit electrophysiological and motoric markers of attention, reflecting a conserved attentional resource across hemifields. Choice-specific reward variations modulate decision criterion and pre-stimulus alpha power, consistent with response-strategy biases, without engaging spatial attention markers. These findings refine theoretical models linking reward, attention, and choice, and reconcile conflicting reports regarding neural correlates of sensitivity versus criterion. Future research should: • Use high-spatial-resolution neuroimaging (e.g., fMRI) to dissociate brain regions processing reward valence and salience and their roles in attentional components. • Apply drift-diffusion modeling to parse mechanisms by which reward alters starting point versus evidence accumulation rate for sensitivity and criterion. • Test generalization to other modalities and tasks, and distinguish reward-induced response biases from spatial choice biases that may constitute attentional components.

• EEG spatial resolution limits dissociating fine-grained brain regions; lack of valence (gain vs loss) main effects on neural/motor measures may reflect this limitation, motivating fMRI for regional specificity. • Early visual ERP components (e.g., P2) showed no modulation; the immediate response window after change gratings may have precluded analysis of post-change epochs. • Monocular eye-tracking can be less robust than binocular tracking; although validated and supplemented by algorithmic criteria, residual measurement limitations may remain. • The paradigm relies on implicit learning of reward contingency switches; individual differences in learning strategies might influence criterion and attention allocation. • Generalizability beyond visuospatial orientation change detection and keypress responses remains to be tested.