Introduction

Temporal integration is a fundamental operation in vision: it allows the nervous system to accumulate sensory evidence over brief time intervals so that perception remains robust despite internal noise and external uncertainty¹. Yet, we still lack a clear account of how the brain sets the temporal window within which sensory signals are bound. In other words, how does the system decide whether two sequential inputs should be integrated or segregated?

A prominent proposal is that rhythmic activity in the alpha band (~8–13 Hz) provides a temporal scaffold for perception. On this view, alpha oscillations define temporal excitation windows that allow for the integration and accurate processing of sensory information^2,3,4,5. Accordingly, faster individual alpha frequencies (IAFs) would enable finer temporal resolution, whereas slower alpha rhythms would enlarge the integration window and promote binding. However, despite its conceptual appeal, the relationship between alpha oscillations and perception remains debated. Empirical evidence has been mixed, with several recent studies reporting null or even contradictory findings, calling into question the generality of this framework^6,7.

Several factors may underlie these inconsistencies, both methodological and theoretical^5,8,9,10. One methodological issue is that most previous studies have relied on variants of the two-flash fusion task, in which participants judge whether two brief, identical flashes presented in close succession are perceived as one or two distinct events². Within the framework of rhythmic perception, it is hypothesized that stimuli falling within the same alpha cycle are bound into a single percept. However, this interpretation of two flash perception is not unambiguous: because each flash can occur at different phases of the ongoing alpha rhythm, one stimulus might simply be processed suboptimally, yielding an apparent fused percept that reflects detection failure rather than genuine temporal integration^3,9,11,12,13. One way to address this challenge is to use paradigms in which the stimuli are shown on screen for several seconds^9,10.

On a more theoretical level, there is disagreement over the fundamental mechanisms that link alpha oscillations to visual temporal perception. From a Bayesian perspective, for example, alpha frequency may shape perceptual inference by altering perceptual sensitivity or by adjusting top-down priors^6,14. However, this framework often overlooks that such priors may be dynamically shaped by recent perceptual experience, which continuously updates internal models and biases the interpretation of incoming sensory information (i.e., serial dependence^15,16). Critically, recent evidence shows that temporal binding mechanisms are not fixed within individuals but adaptively expand or contract based on prior perception, with events following a binding percept more likely to be experienced as a single unified event^17,18,19. Interestingly, it has been hypothesized that serial dependence itself may be linked to alpha oscillations. One possibility is that these oscillations facilitate communication, binding current sensory input with recent perceptual history^16,20,21. It is also plausible that the speed of alpha oscillations affects how strongly past experiences shape current perception. When IAF is slower, temporal resolution and sensory precision may decrease²², making the brain’s estimate of current sensory input more uncertain and thereby increasing reliance on prior experience to stabilize perception.

Another critical aspect that has often been overlooked in studies of temporal integration is its metacognitive dimension. Subjective confidence systematically influences pre- and post-perceptual processing across sensory domains^23,24,25. Recent evidence suggests that faster alpha rhythms may enhance perceptual resolution by generating more sampling frames per time unit, and that more frequent sampling of sensory input may yield a more faithful representation of external events^22,26,27, thus enhancing perceptual and metacognitive abilities. In sum, the relationship between alpha oscillations and temporal perception may be more multi-layered, resisting a simple, single mechanistic explanation.

Another critical consideration is that the focus on oscillations may ignore a link to aperiodic activity, since the complex nature of visual temporal perception likely depends on multiple neural mechanisms, not only rhythmic oscillations^5,8,28. Neural activity exhibits a prominent aperiodic 1/f component reflecting nonrhythmic brain dynamics²⁹, and while there is growing evidence for the importance of taking aperiodic spectra into account³⁰, exactly how and why it influences temporal perception is still being explored^5,8. Flatter spectra, indicating greater neural noise, may disrupt alpha-mediated pulsed inhibition, increasing processing variability and weakening temporal coupling^31,32. It has been shown that aperiodic activity can influence how rhythmic oscillations organize sensory information over time^5,8. Under the proposed framework, the association between alpha rhythms and temporal integration emerges only when internal noise is considered, with higher neural noise linked to greater variability of visual temporal processing and longer integration periods^5,8. Thus, oscillatory and aperiodic neural dynamics may jointly shape visual temporal integration, providing a more complete account of how the brain samples sensory input over time and potentially explaining recent null findings. In this study, we investigated this potential interaction between aperiodic structure and oscillatory dynamics, in the context of the different aspects and potential mechanisms involved in temporal perception.

To address this challenge, we developed a novel variation of a sustained visual stimulus paradigm⁹, based on the perceptual fusion of consecutive visual patterns presented in a temporally extended stream (Fig. 1A). A large sample of participants (n = 83) viewed two low–spatial-frequency Gabor patches oriented in opposite directions (±45°) that alternated in the same spatial location in counterphase over one second. Rather than manipulating the interval between discrete flashes, we varied the duration of each Gabor presentation. This created a continuous visual flow in which shorter durations elicited the illusory perception of a single fused plaid (integration), whereas longer durations produced the perception of two alternating gratings (segregation). This design allowed us to estimate individual temporal integration thresholds while avoiding the confounds of traditional paradigms (e.g., detection failures), providing a more direct measure of perceptual integration⁹. An advantage of continuously presented stimulus streams that integration performance is not confounded with missing a rapidly presented stimulus⁹. Before the behavioral task, we recorded resting-state EEG in both eyes-open and eyes-closed conditions to examine how oscillatory and aperiodic neural dynamics jointly shape visual temporal integration. Critically, we considered the influence of prior perceptual judgment and subjective confidence, relating each of these factors to periodic and aperiodic activity. Including all of these measures of temporal integration performance allowed us to test a mechanistic framework in which faster alpha oscillations within a low-noise neural background enhance the rate and fidelity of visual sampling^5,8,22, thereby increasing temporal precision, reducing dependence on prior perception, and strengthening confidence, together shaping how the visual system dynamically integrates sensory information over time.

Figure 1.

Results

Visual temporal binding is dynamically shaped by recent perceptual experience

The sustained stream of visual patterns in our temporal integration task elicited the illusory perception of a single fused plaid (integration) at shorter inter-stimulus intervals (ISIs), whereas longer ISIs induced the perception of two alternating gratings (segregation; Fig. 1A, 1B). This continuous task replicates classical findings from two-flash paradigms while avoiding their confounds⁹, providing a more direct measure of perceptual binding. For each participant, we computed the proportion of separate responses across ISI levels and fitted a logistic psychometric function to derive two key parameters: the 50% threshold (Point of Subjective Equality, PSE) and the slope (b). The PSE reflects the temporal boundary between perceptual integration and segregation. The slope typically indexes perceptual sensitivity or internal noise, with steeper slopes indicating higher temporal precision^5,8. Across participants, the mean temporal integration threshold was centered around ~80 ms (PSE = 77.71 ± 15.66 ms), closely aligning with the frequency of alpha-band oscillations.

Notably, visual temporal processing was not fixed but dynamically modulated by perceptual choice history. To assess this serial dependence effect, trials were binned according to the perceptual outcome of the preceding trial: i) t–1 integration (fused response in the previous trial) and ii) t–1 segregation (separate response in the previous trial). Separate psychometric functions were then fitted for each condition, yielding distinct PSE and slope estimates (Fig. 1B). A repeated-measures ANOVA on PSE values revealed a significant main effect of t–1 judgment (F(1,82) = 187.15, p < 0.001, η²p = 0.695; Fig. 1C), indicating that integration responses were more likely to follow integration trials, whereas segregation judgments increased the likelihood of perceived segregation on the next trial (t–1 integration: M = 85.73, SD = 15.78; t–1 segregation: M = 68.59, SD = 14.58; p < 0.001; Cohen’s d = 1.13). In contrast, no significant effect of t–1 judgment was found on the slope parameter (F(1,82) = 0.34, p = 0.56, η²p = 0.004). These results show that recent perceptual experience biases the temporal binding window without affecting precision. In line with recent evidence^18,19, this suggests that temporal integration mechanisms are flexible within individuals, adaptively expanding or contracting based on prior perceptual outcomes, reflecting a dynamic interplay between sensory evidence and perceptual history.

Confidence dynamically adjusts with temporal evidence and prior perceptual experience

After each temporal binding judgment, participants rated their subjective confidence on a four-point scale. As expected, confidence followed the temporal clarity of sensory evidence: participants reported lower confidence at intermediate ISIs, corresponding to perceptually ambiguous trials near the temporal integration threshold, and higher confidence at shorter and longer ISIs, where temporal evidence more clearly supported integration or segregation decisions. Across participants, mean subjective confidence was inversely related to the temporal integration threshold (PSE), revealing a negative linear correlation between confidence and PSE (r = –0.28, p = 0.009; Fig. 1E). Thus, individuals with narrower integration windows (i.e., higher temporal acuity) reported higher overall confidence in their perceptual judgments. In contrast, mean confidence was not significantly correlated with the slope of the psychometric curve (r = 0.15, p = 0.177). We next examined whether confidence was influenced by recent perceptual history. Trial-wise confidence ratings were sorted by ISI and by the perceptual outcome of the preceding trial (t–1 integration vs. t–1 segregation), and compared using permutation-based paired t-tests (10,000 permutations; FDR-corrected, q = 0.05). Confidence ratings were significantly higher following t–1 integration trials at short ISIs (21–63 ms; q < 0.05; Fig. 1D), whereas confidence was higher following t–1 segregation trials at long ISIs (104–125 ms; q < 0.05; Fig. 1D). No significant effects of prior perception emerged at the extremes of the temporal range (0 ms and 146 ms; q > 0.05; Fig. 1D) or near the point of subjective equality (83 ms; q > 0.05). Together, these findings indicate that confidence was enhanced when the current temporal evidence supported the same perceptual outcome as the previous trial, suggesting that recent perceptual states bias not only perceptual judgments but also their subjective certainty³³.

Alpha oscillations and aperiodic neural dynamics predict temporal integration

Alongside the temporal integration task, we recorded resting-state EEG activity (eyes-closed, EC; eyes-open, EO) from 64 scalp electrodes. In order to examine how oscillatory and aperiodic neural dynamics jointly shape visual temporal processing⁸, we applied the FOOOF algorithm³⁰ to decompose each participant’s power spectrum into oscillatory and aperiodic components (Fig. S1; see Supplementary). First, we tested the oscillatory sampling hypothesis, correlating resting-state IAF with behavioral measures of temporal integration (PSE and psychometric slope) using Pearson correlations with cluster-based correction (10,000 permutations). Faster IAFs, particularly over posterior–central sites, were associated with lower temporal thresholds (PSE) in both EC and EO conditions (p < 0.001; Fig. 2A), supporting the notion that faster alpha cycles correspond to narrower integration windows. By contrast, IAF was not significantly correlated with psychometric slope in either condition (p > 0.05; Fig. 2B). We then examined the contribution of aperiodic neural activity. Higher aperiodic exponents were significantly associated with steeper psychometric slopes in both EC and EO conditions (p < 0.001; Fig. 2D), indicating that a steeper (less flat) spectrum, reflecting lower neural noise, was linked to greater perceptual precision in temporal binding⁸. Additionally, higher aperiodic exponents were positively correlated with lower temporal thresholds (PSE) in the EO condition (p < 0.001; Fig. 2C), suggesting that reduced neural noise promoted finer temporal resolution. The absence of this effect in the EC condition further replicates our previous findings⁹ and might reflect stronger alignment between neural dynamics and visual processing in EO. In order to check the specificity of alpha and aperiodic activity in temporal perception, we also correlated psychometric function measures with spectral power in the delta (1–4 Hz), theta (4–7 Hz), and beta (15–25 Hz) bands. No significant associations were found (p > 0.05). Additionally, because both IAF and the aperiodic exponent predicted individual differences in PSE, we tested whether their effects were independent or mediated. Mediation analyses (see Method) revealed no significant indirect effect of IAF on temporal thresholds via the aperiodic exponent (β indirect = −0.043, CI [−0.154, 0.017]). Instead, IAF exerted a robust direct effect on PSE (β direct = −0.512, p < .001), alongside an independent contribution of the exponent (β = 0.236, p = .035), indicating that the aperiodic component acts in parallel rather than mediating the IAF–PSE relationship. Together, these findings corroborate previous evidence that visual temporal acuity emerges from the joint influence of oscillatory alpha dynamics and aperiodic neural mechanisms⁸.

Figure 2.

Neural predictors of serial dependence and subjective confidence in temporal binding

We next examined whether resting-state EEG activity predicts the strength of serial dependence during temporal binding. For each participant, we quantified the magnitude of serial dependence as the difference between the temporal thresholds (PSE) derived from the logistic fits following t–1 integration and t–1 segregation trials (ΔPSE = PSE t–1 integration – PSE t–1 segregation). This index captures how strongly the percept from the previous trial biases current temporal judgments. Similarly, we computed a slope difference index (Δslope = slope t–1 integration – slope t–1 segregation) to assess whether prior perceptual history affects perceptual precision and internal noise. Correlating these behavioral indices with resting-state EEG measures (IAF and aperiodic exponent) revealed distinct neural signatures of serial dependence. Faster IAFs, particularly over posterior–central electrodes, were associated with lower ΔPSE values in the EO condition (p < 0.001; cluster-based corrected; Fig. 3A), indicating that individuals with faster alpha rhythms (i.e., finer temporal resolution) were less influenced by prior experience, relying more on current sensory evidence. This relationship was absent in the EC condition (p > 0.05), and no significant associations emerged between IAF and Δslope in either condition (p > 0.05). Conversely, lower exponents were linked to higher ΔPSE values in the EO condition (p < 0.001; Fig. 3B), suggesting that individuals with noisier neural activity rely more heavily on prior perceptual outcomes to stabilize perceptual decisions under uncertainty. This association was not significant in the EC condition (p > 0.05), nor was any reliable relationship found between the aperiodic exponent and Δslope in either condition. Additionally, a mediation analysis (see Method) on the relationship between resting-state IAF and serial dependence strength (ΔPSE) revealed no significant indirect effect (β = 0.07, CI [−0.04, 0.22]). Both faster alpha rhythms (higher IAF; β direct = −0.43, p < .001) and aperiodic activity (β = −0.39, p < .001) independently predicted the degree of reliance on prior perceptual history.

Figure 3.

Finally, we tested whether EEG measures predicted subjective confidence in temporal binding judgments. Faster IAFs over posterior–central sites were associated with higher mean confidence ratings in the EO condition (p < 0.001, cluster-corrected; Fig. 3C), indicating that individuals with faster alpha cycles not only exhibited finer temporal resolution but also greater subjective certainty in their perceptual decisions. No significant relationships were observed between confidence and IAF in the EC condition, or between confidence and aperiodic activity in either condition (p > 0.05). Together, these results demonstrate that oscillatory and aperiodic neural dynamics jointly shaped both the perceptual and metacognitive facets of temporal integration, where faster alpha oscillations promoted sharper temporal sampling and higher confidence in perceptual judgments, reducing dependence on the immediate past.

Discussion

The perceptual system segments the continuous flow of sensory inputs into discrete events to construct a coherent experience of an otherwise chaotic visual world. This temporal parsing has long been linked to the intrinsic speed of alpha-band oscillations² and more recently to the aperiodic structure of EEG spectra⁸. However, recent studies have reported contradictory findings that challenge a direct link between alpha oscillations and individual differences in perceptual resolution⁶. Here, by combining a novel continuous temporal integration paradigm with resting-state EEG, we replicate the classical two-flash findings while overcoming their key limitations that can confound genuine perceptual binding from failure to fully process both stimuli. We replicated the finding that oscillatory and aperiodic activity jointly predict the threshold and slope of the psychophysical curve^5,8, now using a sustained visual stream, as well as linking it for the first time to previous perception effects and confidence judgments. Our results reveal that visual temporal integration arises from a multilevel process governed by the interplay of rhythmic and aperiodic neural activity, perceptual history, and subjective confidence, moving beyond a simple account of the role of alpha oscillations in temporal integration.

First, our findings provide support for, but also a more complex understanding of the link between brain rhythms and visual perception. Individuals with faster resting-state alpha rhythms did indeed show faster temporal integration thresholds, in line with previous findings using brief flashes^{2,3,4,5,8,13,26,34}. However, the speed of alpha frequency did not predict the steepness of the psychometric function (slope), suggesting that rhythmic activity alone does not determine perceptual precision or internal noise during temporal binding^35,36. Interestingly, our findings revealed that this aspect of temporal processing was captured by the aperiodic component of neural activity^5,8. Consistent with a recent framework⁸, higher aperiodic exponents were associated with steeper psychometric slopes and lower temporal thresholds, indicating that a steeper spectrum (i.e., lower neural noise) was linked to both greater perceptual precision and finer temporal resolution. These results are also consistent with the view that aperiodic activity indexes the neural excitation–inhibition balance³⁷ and modulates oscillatory synchronization relevant for the temporal organization of perception^5,8,38,39. In this framework, alpha speed may determine the temporal resolution of perception, whereas the aperiodic structure can regulate its reliability by modulating the excitation–inhibition balance, thus enhancing its precision⁸.

It is interesting that the aperiodic structure of electrophysiological signals has been related to various aspect of cognition and perception⁴⁰, such as speed of processing^41,42, working memory⁴³, emotional processing and arousal^44,45, as well as numerous psychopathologies⁴⁶, and aging^30,31,46.

The current findings explain in more detail how and why resting state EEG can predict individual performance in temporal perception tasks. Our findings reveal that visual temporal integration judgments are strongly perceptual history-dependent and shaped by both alpha rhythmicity and aperiodic neural activity. When we separately examined temporal integration thresholds for trials preceded by an integration response and those preceded by a segregation response, we found that temporal binding judgments were biased toward the previous perceptual outcome. Specifically, a previous integration response increased the likelihood of perceiving a subsequent stimulus as a single unified event, and vice-versa for segregation responses. These results demonstrate that recent perceptual experience flexibly biases the temporal binding window, consistent with recent evidence showing that temporal perception reflects a dynamic interplay between the temporal structure of sensory evidence and perceptual history^17,18,19. Recent work has suggested that alpha oscillations may also support serial dependence processes^16,21. In this study, faster resting-state alpha rhythms and higher aperiodic exponents were associated with a lower magnitude of serial dependence effect, indicating reduced influence of prior perceptual outcomes and a greater reliance on current sensory evidence, potentially due to increased sensory precision^22,27. Conversely, individuals with slower alpha rhythms and lower aperiodic exponents showed stronger serial dependence, suggesting that when temporal resolution and sensory precision decrease the brain compensates by weighting prior perceptual information more heavily to stabilize perceptual experience under sensory uncertainty¹⁶.

After each temporal binding judgment, participants rated their confidence, which mirrored the clarity of temporal evidence: confidence was lowest near the temporal integration threshold and highest at short and long inter-stimulus intervals, where evidence more clearly supported either integration or segregation. Additionally, subjective confidence was inversely related to individual temporal thresholds, indicating that observers with narrower integration windows (i.e., higher temporal acuity) reported higher overall confidence in their perceptual judgments. Crucially, confidence also modulated the interaction between prior perceptual history and current temporal evidence. Confidence ratings were significantly higher following integration judgments at short inter-stimulus intervals and higher following segregation responses at long intervals. These findings suggest that confidence is enhanced when current sensory evidence supports the same perceptual outcome as the previous trial, suggesting that recent perceptual states bias not only perceptual judgments but also subjective certainty³³. Finally, we tested whether neural dynamics predicted subjective confidence in temporal binding judgments. Our findings showed that faster alpha rhythms were associated with higher confidence, indicating that individuals with faster alpha cycles not only exhibited finer temporal resolution and less reliance on previous experience, but also greater certainty in their perceptual decisions, supporting previous evidence linking alpha-band oscillations to both temporal resolution and confidence in perceptual decisions^23,24,47.

Overall, the alpha rhythm was distinctly associated with multiple facets of visual temporal processing, from temporal thresholds and serial dependence to subjective confidence. One potential explanation for this multilevel influence of alpha oscillatory speed on temporal binding is that faster alpha rhythms operating within a neural background of lower noise may enhance temporal resolution through more efficient sensory evidence accumulation^{5,8,26,27,34,48,49,50,51}. Recent evidence suggests that faster alpha rhythms promote greater perceptual resolution by generating more sampling frames per time unit, thereby increasing the amount of accumulated sensory evidence^22,26,27. Consequently, more accurate and higher-rate sampling of visual input may yield a more faithful representation of external events, resulting in higher temporal precision, reduced reliance on prior perceptual experience, and greater subjective confidence in perceptual decisions, as observed in the current study.

Together, these results support and expand on our recently proposed framework linking alpha oscillatory speed and aperiodic neural dynamics in perceptual integration^5,8, showing also a link to decisional and metacognitive facets. By linking rhythmic sampling with the stability of the underlying neural background, this framework highlights that temporal perception emerges from the coordinated interplay between oscillatory precision, neural noise regulation, and the brain’s predictive use of recent perceptual history. Beyond temporal processing, this account provides a broader model for how rhythmic and aperiodic neural dynamics jointly sustain the continuity of conscious visual experience across time.

Methods

Participants

A total of 83 volunteers (46 females, mean age=23.4 years, SD=2.58) participated in the experiment and received compensation. Participants presented normal or corrected-to-normal vision. Exclusion criteria were self-reported neurological and attention disorders, epilepsy, and photosensitivity. The sample size was determined a priori based on a recent meta-analysis of 27 studies examining the relationship between individual alpha frequency and temporal resolution². The meta-analysis reported a pooled population correlation of r = 0.394 (95% CI [0.332, 0.465]; SE = 0.034). Using this estimate as the expected effect size, an a priori power analysis (two-tailed, α = 0.05) indicated that a minimum of N = 49 participants would be required to achieve 80% power to detect r = 0.394, and N = 69 for the lower bound of the confidence interval (r = 0.332). To ensure sufficient power to detect effects within this plausible range, we recruited 83 participants, which provided approximately 96% power for r = 0.394. Power analyses were conducted using G*Power 3.1⁵². The experiment was conducted in accordance with the Declaration of Helsinki and approved by the New York University Abu Dhabi Human Research Protection Program Internal Review Board (IRB).

Stimuli and Experimental Design

We designed a novel visual temporal integration task to measure individual integration performance while avoiding confounds associated with brief, discrete two-flash paradigms. Stimuli were generated and presented in MATLAB (MathWorks) using Psychtoolbox on a gamma-corrected monitor (refresh rate: 144 Hz). Observers viewed the display binocularly from a distance of 70 cm in a dimly lit room. On each trial, two Gabor patches were presented at the center of the screen, alternating over ~1 s in the same spatial location. Each Gabor had a low spatial frequency (1.94 cycles/degree; λ = 0.52°) and a Gaussian envelope (σ = 0.62°), presented at full contrast, with orientations of +45° (right-tilted) and −45° (left-tilted). Each Gabor was rendered with a circular aperture and converted to a texture. These two orthogonal Gabors alternated rapidly to manipulate temporal perception. The key manipulation was the duration of each Gabor presentation (i.e., the inter-stimulus interval, ISI) rather than the interval between discrete events. The Gabors alternated in counterphase with an ISI ranging from 0 to 146 ms in 21-ms steps. Each ISI level was repeated 30 times, yielding 240 experimental trials presented in random order. For ISI = 0, a “fully fused” control was shown by presenting a precomputed fused texture with slight spatial jitter to prevent adaptation. The order of presentation (which orientation appeared first) was randomized across trials. This manipulation produced a continuous visual stream in which short presentation durations elicited the illusory percept of a single fused plaid (integration), whereas longer durations induced the perception of two alternating gratings (segregation). Because the stimuli had distinct orientations but occupied the same retinal location within an uninterrupted sequence, this paradigm dissociated genuine temporal binding from simple detection failures. At the beginning of the experimental session, resting-state EEG activity was recorded while participants relaxed for 3 minutes with eyes closed (EC condition) and 3 minutes with eyes open (EO condition). Before the task, participants received on-screen instructions and completed a short practice block to familiarize with the procedure. Each trial began with a central fixation cross (500 ms), followed by the ~1 s alternation sequence. Immediately after stimulus offset, participants made a two-alternative forced choice indicating their percept: fused/single pattern (key F) or separate/alternating patterns (key S). They then rated their confidence on a four-point scale (1 = not confident at all, 2 = not very confident, 3 = quite confident, 4 = very confident). A short break was provided halfway through the session.

Data Analysis

Behavioral Analyses and Psychometric Function

For each participant, we first computed the proportion of separate responses for each inter-stimulus interval (ISI) level and fitted a psychometric logistic function to the percentage of separate responses as a function of ISI. The fitting was performed individually using a nonlinear least-squares method (mean adjusted R² = 0.938; all participants R² > 0.85). We used a logistic equation and a nonlinear least squares method to fit the proportion of separate responses as a function of ISIs. The formula applied was as follows: y = 1/(1 + exp(b × (t - x))). In this equation, x represents the ISI, and y denotes the proportion of separate responses. The lower bound of y was set at 0, and the upper bound was set at 1. The only free parameters of the function were b (the slope of the function) and t (the 50% threshold), both of which were constrained to assume positive values above zero. Individual 50% threshold and psychometric slope values were obtained by fitting the psychometric logistic curve. The threshold (t) corresponds to the Point of Subjective Equality (PSE)—the temporal delay at which participants were equally likely to report a fused or separate percept. Within this framework, a lower PSE reflects greater temporal segregation (narrower integration window), whereas a higher PSE indicates an increased tendency to bind. The slope parameter (b) indexes the steepness of the psychometric function, providing an estimate of perceptual sensitivity or sensory noise: steeper slopes reflect higher temporal precision, whereas shallower slopes indicate noisier, less precise perceptual decisions. To assess the influence of perceptual history on temporal binding, we implemented a serial dependence analysis. Trials were divided into two bins based on the perceptual outcome of the preceding trial: t–1 segregation (previous trial perceived as separate) and t–1 integration (previous trial perceived as fused). To ensure statistical robustness, we verified that the number of trials per ISI was comparable across bins and that all participants contributed sufficient data for stable curve fitting. Separate logistic functions were then fitted for t–1 integration (mean adjusted R² = 0.918) and t–1 segregation trials (mean adjusted R² = 0.92), yielding distinct PSE and slope estimates for each condition. The difference in PSE between the two bins (ΔPSE = PSE t–1 integration – PSE t–1 segregation) was used as a quantitative index of serial dependence strength, reflecting the extent to which percepts from the preceding trial biased current perceptual judgments. In parallel, an analogous measure was derived for the psychometric slope (Δslope = t–1 integration minus t–1 segregation) to assess potential carry-over effects on response precision. Finally, we computed the average confidence rating for each participant and ISI level, separately for all trials, t–1 integration trials, and t–1 segregation trials.

EEG analysis

Resting-state EEG was recorded in two conditions, eyes closed (EC) and eyes open (EO), each lasting 3 min. Signals were acquired from 64 active scalp electrodes using BrainVision Recorder (Brain Products), sampled at 1000 Hz, and referenced online to FCz. Impedances were kept below 10 kΩ. Participants were instructed to relax, minimize movements, and fixate (EO) or keep their eyes closed (EC). Preprocessing was performed in MATLAB. For each participant, noisy channels were identified by visual inspection and interpolated using spherical splines. Oculomotor artifacts were removed by independent component analysis (ICA). The reference was kept at FCz for all subsequent analyses. We intentionally avoided re-referencing to the average reference because broadband noise or a single noisy channel would otherwise be spread over all channels and distort the estimation of the aperiodic (1/f) component⁸. For spectral analysis, continuous resting-state data were segmented into 1-s epochs. Remaining segments were submitted to Welch’s method, yielding a power spectral density (PSD) estimate. Similar to our recent studies^5,8, to separate oscillatory from aperiodic activity, we applied the “fitting oscillations and one-over-f” (FOOOF) algorithm³⁰ to the PSD of each participant, condition (EC, EO), and electrode. FOOOF models the log-power spectrum as the sum of (i) an aperiodic component following a 1/f function (offset and exponent) and (ii) a set of Gaussian peaks capturing periodic oscillations. We fitted spectra in the 1–70 Hz range to avoid oscillations and spectral plateau crossing the fitting range borders. Default settings were used (peak width limits: 0.5–12 Hz; minimum peak height: 0; maximum number of peaks: Inf; peak threshold: 2), as in previous works^30,53. The aperiodic exponent (slope) from this fit was taken as our index of aperiodic neural activity and, by extension, of excitation/inhibition balance^31,37. Alpha power was defined as the spectral peak within this frequency range in both the full and aperiodic-corrected spectra. The individual alpha frequency (IAF) was identified as the frequency of the maximum spectral peak within the 8–13 Hz range. For control analyses, we additionally computed the power of delta (1–4 Hz), theta (4–7 Hz), and beta (15–25 Hz) oscillations to examine potential relationships with temporal perception outside the alpha band.

Statistical Analysis

After computing individual temporal thresholds (PSE) and slopes, we examined choice-history effects (i.e., serial dependence) using a repeated-measures ANOVA on PSE values, with the preceding trial type (t–1 judgment: integration vs. segregation) as a within-subjects factor. Greenhouse–Geisser corrections were applied where necessary to adjust for violations of sphericity, and all post-hoc p-values were Holm-corrected for multiple comparisons. To quantify serial dependence strength at the individual level, we computed a PSE difference index (ΔPSE = PSE t–1 integration – PSE t–1 segregation) and a slope difference index (Δslope = slope t–1 integration – slope t–1 segregation). Larger ΔPSE and Δslope values reflect stronger dependence on the previous perceptual judgment. These single-value indices served as summary measures of serial dependence strength, used in subsequent correlational analyses with EEG-derived parameters.

For confidence analyses, trial-wise confidence ratings were extracted for each participant and sorted according to ISI and preceding trial type (All trials, t–1 integration, t–1 segregation). Confidence scores were z-scored within participants (row-wise normalization) to preserve within-subject relative variability while removing between-subject scale biases. Differences in confidence between t–1 integration and t–1 segregation trials were assessed using permutation-based paired-sample t-tests conducted separately for each ISI. In each permutation (N = 10,000), participant-wise differences were randomly sign-flipped and a new t-statistic was computed to form an empirical null distribution. Two-tailed empirical p-values were defined as the proportion of permuted t-values with absolute magnitude greater than or equal to the observed t-statistic. To correct for multiple comparisons across ISIs, p-values were adjusted using the Benjamini–Hochberg False Discovery Rate (FDR) procedure (q = 0.05). Effect sizes were calculated as Cohen’s d for paired samples.

For EEG data, we first assessed differences between the eyes-closed (EC) and eyes-open (EO) resting-state conditions. For each electrode, we conducted two-sided dependent-sample t-tests using nonparametric, cluster-based permutation methods⁵⁴ to control for multiple comparisons (10,000 permutations). This analysis was performed separately for the aperiodic exponent and individual alpha frequency (IAF).

Next, for each channel and resting-state condition, we examined EEG–behavior relationships by computing Pearson’s correlation coefficients between EEG measures and behavioral indices. Using the same nonparametric cluster-based permutation framework⁵⁴ (10,000 permutations), we correlated the aperiodic exponent and IAF with the temporal integration threshold (PSE) and psychometric slope derived from the behavioral data across all trials. To assess whether EEG parameters predicted the strength of serial dependence, we further correlated EEG measures with ΔPSE (t–1 integration - t–1 segregation). We also tested whether EEG features were related to metacognitive aspects of temporal perception. For this, mean confidence ratings were computed for each participant and correlated with EEG spectral measures (aperiodic exponent and IAF) using Pearson’s correlations with cluster-based nonparametric correction (10,000 permutations). Finally, as control analyses, we correlated behavioral measures (PSE, slope, ΔPSE, Δslope, and confidence) with resting-state oscillatory power in the delta (1–4 Hz), theta (4–7 Hz), alpha (8–13 Hz), and beta (15–25 Hz) bands to evaluate potential relationships outside the alpha-frequency range. Although a priori hypotheses specified the expected direction of effects, all statistical tests were two-sided. All EEG analyses were implemented in MATLAB using custom scripts in combination with the FieldTrip Toolbox⁵⁵ and the FOOOF-mat toolbox³⁰. To examine how oscillatory and aperiodic neural dynamics jointly contribute to perceptual behavior, we conducted a series of mediation analyses using the lavaan package (version 0.6–19) in RStudio⁵⁶. All variables were z-standardized prior to analysis to facilitate the interpretation of path coefficients. Structural equation models were estimated using maximum likelihood estimation with Full Information Maximum Likelihood (FIML). The significance and confidence intervals of indirect effects were assessed using 10,000 bias-corrected bootstrap resamples, from which 95% confidence intervals (CIs) were derived. Path coefficients are reported as standardized regression weights (β). Separate mediation models were specified to test directional hypotheses regarding how resting-state EEG parameters predicted behavioral measures. We first tested whether the relationship between oscillatory alpha speed and temporal thresholds could be explained by aperiodic neural activity, by estimating mediation models in both directions: IAF - aperiodic exponent - PSE and the reverse aperiodic exponent - IAF - PSE. These analyses examined whether either oscillatory speed or the aperiodic component served as an intermediary mechanism linking resting-state EEG dynamics to temporal integration precision. Next, we applied the same bidirectional framework to serial dependence strength (ΔPSE), testing IAF - aperiodic exponent - ΔPSE and aperiodic exponent - IAF - ΔPSE, to assess whether these neural parameters jointly contribute to the degree of reliance on prior perceptual experience. Finally, we estimated a parallel mediation model including the aperiodic exponent, serial dependence strength (ΔPSE), and confidence as simultaneous mediators of the association between IAF and PSE. This comprehensive model was used to assess whether the relationship between oscillatory dynamics and temporal integration could be explained by the combined or independent contributions of neural noise, reliance on prior perceptual history, and subjective confidence. All models were tested on EEG values averaged over posterior electrode clusters showing significant correlations in the eyes-open (EO) condition. Effects were considered significant when the 95% bootstrap confidence interval of the indirect effect did not include zero. Model fit was evaluated using standard indices (CFI, TLI, RMSEA, SRMR), and explained variance (R²) was reported for each dependent variable.

Supplementary Information

EEG activity in eyes-closed and eyes-open Resting-State

To characterize the spectral components of neural activity, we applied the FOOOF algorithm (Donoghue et al., 2020) to decompose each participant’s power spectrum into oscillatory and aperiodic components (Fig. S1A). We first compared EC and EO conditions using two-sided dependent-sample t-tests with cluster-based permutation correction for multiple comparisons (10,000 permutations; Maris & Oostenveld, 2007). The analyses revealed no significant difference in individual alpha frequency (IAF) between conditions (p > 0.05; Fig. S1B). In contrast, the aperiodic exponent was significantly higher across widespread scalp regions in the EC condition compared to EO (p < 0.001; Fig. S1C), consistent with previous reports of increased aperiodic activity during eyes-closed rest (Dehghani et al., 2010; Donoghue et al., 2020; Deodato & Melcher, 2024).

Figure S1.

Mediation analyses linking neural dynamics and temporal binding

Our results revealed that both IAF and the aperiodic exponent predicted individual differences in temporal thresholds (PSE) and their modulation by prior perceptual experience (ΔPSE). However, it was important to also check whether these effects were independent, or that one mediated the other. We conducted mediation analyses testing whether the effects of IAF on behavioral measures were mediated by aperiodic activity. Path coefficients were estimated with maximum likelihood and evaluated using 10,000 bias-corrected bootstrap resamples (95% confidence intervals). Analyses were performed on the average electrode activity showing the strongest correlations in the EO condition, as identified in previous results.

The first mediation analysis indicated no significant indirect effect of IAF on temporal thresholds via the aperiodic exponent (β indirect = −0.043, CI [−0.154, 0.017]). Instead, IAF exerted a robust direct effect on PSE (β direct = −0.512, p < .001), with an additional independent contribution of the exponent (β = 0.236, p = .035; R² PSE = .36). Thus, the IAF’s influence on temporal integration is not mediated by aperiodic activity but rather operates in parallel through an IAF-specific pathway. A second mediation analysis on the relationship between resting-state IAF and serial dependence strength (ΔPSE) also revealed no significant indirect effect (β = 0.07, CI [−0.04, 0.22]). Both faster alpha rhythms (higher IAF; β direct = −0.43, p < .001) and aperiodic activity (β = −0.39, p < .001) independently predicted the degree of reliance on prior perceptual history. Finally, a parallel mediation model including confidence, aperiodic exponent, and ΔPSE as simultaneous mediators of the IAF–PSE relationship showed no reliable indirect pathways, whereas the direct effect of IAF on PSE remained robust (β = −0.48, p < .001; R² PSE = .37).

Data availability

Data are shared open-access at https://osf.io/4wdmc/.

Acknowledgements

This work was supported by the NYUAD Center for Brain and Health, funded by Tamkeen under NYU Abu Dhabi Research Institute grant CG012.

Additional information

Author contributions

Conceptualization, G.M., M.D., D.M.; Methodology G.M., M.D., D.M.; Formal Analysis, G.M.; Software, G.M.; Visualization, G.M.; Funding Acquisition, D.M.; Writing - Original Draft Preparation, G.M. D.M.; Writing – Review & Editing, G.M., M.D., D.M.; Supervision, D.M.

Funding

New York University Abu Dhabi (NYUAD) (CG012)

• Michele Deodato

• David Melcher

• Gianluca Marsicano