The influence of subjective confidence on speed perception in peripheral vision Authors: Riku Hirano, Tomohiro Kizuka, Seiji Ono Published: 2026 ABSTRACT This study aimed to determine whether speed perception in peripheral vision is modulated by subjective confidence and whether pupil size can serve as an objective indicator of confidence. Previous studies have reported that perceived speed decreases with increasing eccentricity. Based on the principles of a Bayesian model, we employed a two-alternative forced-choice task. Observers judged whether a test stimulus was faster or slower than an 8 deg/s reference stimulus at 5 deg eccentricity, then rated their confidence on a 1–4 scale. The test stimulus appeared at either 5 or 15 deg eccentricity, moving at speeds ranging from 5 to 11 deg/s. As a result, mean confidence was significantly lower at 15 deg eccentricity than at 5 deg, suggesting that blurrier visual input at greater eccentricities reduces perceptual certainty. Furthermore, hierarchical Bayesian modeling showed that lower confidence was associated with slower perceived motion and greater perceptual uncertainty. Higher confidence was also accompanied by greater phasic pupil dilation, indicating that pupil responses reflect subjective confidence. These findings highlight the interactive contribution of sensory and cognitive factors to visual motion perception and provide insight into neural mechanisms linking confidence, visual eccentricity, and speed estimation. Introduction In the present study, we investigated whether subjective confidence influences the perception of speed in peripheral vision and whether phasic pupil responses can serve as an objective indicator of confidence. We acquire visual information across a wide range of the visual field to perceive the environment outside ourselves. However, subjective perception does not always align with physical phenomena. Previous studies have reported that perceived speed decreases with increasing retinal eccentricity, and consequently, objects appear to move more slowly in peripheral vision1–3. Recent theories suggest that human perception is based on Bayesian inference, which integrates sensory input and prior expectations to estimate the characteristics of an object as a posterior probability. According to this theory, when sensory input is uncertain, such as when the contrast of visual stimuli is low, perception tends to rely more on prior expectations. Uncertainty is typically conceptualized as measurement noise in sensory representations, reflecting variability in sensory input4. When sensory measurements are noisy or imprecise, the likelihood function becomes broader, and perceptual estimates rely more strongly on prior expectations. Such uncertainty can arise from multiple sources, including stimulus-dependent sensory noise, stimulus-independent sensory noise, and non-sensory noise5. Since the natural environment contains more stationary objects and fewer fast-moving ones, it has been suggested that observers are thought to have a bias toward slower speeds4,6. This suggests that under conditions of reduced sensory precision, such as in peripheral vision, the influence of priors becomes more pronounced, potentially leading to perceptual biases to underestimate speed. Therefore, because visual acuity and contrast sensitivity decline with increasing eccentricity, the underestimation of speed perception in peripheral vision would be attributed to increased sensory uncertainty and a greater reliance on prior information. Instead, the observed perceptual bias may reflect differences in how sensory evidence is weighted under uncertainty. Within a Bayesian framework, perceptual estimates arise from the integration of sensory likelihood and prior expectations, with the relative influence determined by each of their precision. We hypothesized that subjective confidence reflects the observer’s internal estimate of sensory precision and therefore modulates the weighting of prior information during perceptual inference. Under peripheral viewing, sensory signals are inherently less precise. Yet, the degree to which priors influence perception may depend not only on objective sensory degradation but also on the observer’s subjective estimate of uncertainty. Specifically, lower confidence may indicate a broader likelihood function (reduced precision), thereby increasing the relative weight of the slow-speed prior and enhancing underestimation biases. Conversely, higher confidence may reflect a narrower likelihood, reducing prior influence even under peripheral conditions. Thus, eccentricity-dependent changes in perceived speed may be dynamically modulated by confidence-dependent weighting within Bayesian inference, rather than being determined solely by low-level sensory limitations. Recent studies have shown that changes in pupil diameter can serve as indicators of mental activity and internal state. Specifically, pupil dynamics have been used to estimate attention7–9, cognitive load and mental effort10,11, physical effort12, prediction error and surprise13, and decision making14. Joshi et al. (2016) further showed that pupil diameter is closely linked to activity in the locus coeruleus (LC), a brainstem nucleus that releases norepinephrine broadly across the cortex15. Additionally, Doll et al. (2024) reported that both auditory cortical activity and pupil dilation increased when near-threshold auditory stimuli were perceived more strongly16. These findings suggest that the strength of sensory cortical responses is associated with the strength of perceptual experience. In decision-making, confidence is thought to increase when the available evidence is more reliable, and to decrease when the evidence is uncertain. Therefore, stronger subjective perceptual experiences should lead to higher confidence, whereas weaker perceptual experiences should result in lower confidence. Based on this framework, increasing eccentricity, which leads to a decrease in visual reliability, would alter the subjective strength of visual perception. Taken together, these insights suggest that confidence is grounded in the subjective strength of perceptual experience, and that higher confidence may be accompanied by stronger pupil responses. Importantly, unlike the psychometric slope, which reflects behavioral precision averaged across trials, phasic pupil responses provide a trial-by-trial physiological index of internal uncertainty. Thus, pupil dilation may offer complementary information about confidence related modulation of perceptual inference beyond what can be inferred from psychophysical measures alone. In this study, we investigated whether subjective confidence modulates the perception of speed in peripheral vision. Based on Bayesian theories of perception, we hypothesized that lower confidence would be associated with a stronger bias toward perceiving slower speeds. Furthermore, if phasic pupil dilation reflects subjective confidence, we expected larger pupil dilations in trials with higher confidence. Results Effects of eccentricity on speed perception Observers performed a two-alternative forced-choice (2AFC) speed discrimination task using circular motion stimuli, in which a single reference speed (8 deg/s) was compared against variable test speeds (5–11 deg/s) presented at two eccentricities (5 and 15 deg). Figure 1A shows psychometric functions depicting the proportion of faster responses as a function of test stimulus velocity, separately for the 5 and 15 deg conditions. Each thin curve represents a logistic function fitted to an individual participant’s data, and thick curves indicate group averages. Although the proportion of “faster” responses increased as stimulus velocity increased in both conditions, the 15 deg condition showed a tendency for more slower responses. Figure 1B shows individual and mean PSEs for the two eccentricity conditions. A paired-samples t-test revealed a significant increase in PSE in the 15 deg condition compared to the 5 deg condition (t (9) = − 2.73, p < 0.05, d = 0.86), indicating that stimuli were perceived as slower in the 15 deg condition compared to the 5 deg condition. Figure 1C shows individual and mean JNDs for the two eccentricity conditions. A paired-samples t-test revealed a significant increase in JND in the 15 deg condition compared to the 5 deg condition (t (9) = − 4.68, p < 0.05, d = − 1.48), indicating that stimuli were perceived as slower in the 15 deg condition compared to the 5 deg condition. Figure 1D shows the relationship between delta JNDs and delta PSEs. Robust linear regression revealed that eccentricity related changes in sensory uncertainty (ΔJND) significantly predicted changes in perceptual bias (ΔPSE) across observers (R2 = 0.523, p = 0.031). These findings replicate and are consistent with results reported in previous studies1,2. Fig. 1Psychometric functions and eccentricity effects on PSE and JND. (A) Psychometric functions showing the proportion of “faster” responses as a function of test stimulus velocity for the 5 deg (red) and 15 deg (blue) conditions. Thin lines indicate individual observers, and thick lines represent the group average. Asterisks denote a significant difference between conditions (p < 0.05). (B) Individual PSEs for each eccentricity condition. Asterisks denote a significant difference between conditions (p < 0.05). (C) Individual JNDs for each eccentricity condition. (D) Relationship between individual eccentricity induced changes in PSE and JND. Each dot represents one observer’s difference score (15–5 deg) for PSE (y-axis) and JND (x-axis). The solid line indicates the robust linear regression fit. Psychometric functions and eccentricity effects on PSE and JND. (A) Psychometric functions showing the proportion of “faster” responses as a function of test stimulus velocity for the 5 deg (red) and 15 deg (blue) conditions. Thin lines indicate individual observers, and thick lines represent the group average. Asterisks denote a significant difference between conditions (p < 0.05). (B) Individual PSEs for each eccentricity condition. Asterisks denote a significant difference between conditions (p < 0.05). (C) Individual JNDs for each eccentricity condition. (D) Relationship between individual eccentricity induced changes in PSE and JND. Each dot represents one observer’s difference score (15–5 deg) for PSE (y-axis) and JND (x-axis). The solid line indicates the robust linear regression fit. Effect of eccentricity on perceptual confidence Figure 2 shows the mean confidence ratings for each eccentricity condition. A linear mixed-effects model was conducted with confidence as the dependent variable, eccentricity as a fixed effect, and observer as a random effect. The analysis revealed a significant main effect of eccentricity (Estimate = − 0.254, p < 0.001). This indicates that perceptual confidence was lower in the 15 deg condition compared to the 5 deg condition. Figure 3 shows the mean response times for each eccentricity condition and confidence rating. A generalized linear mixed-effects model with a Gamma distribution and log link function was fitted, with response time as the dependent variable, eccentricity, confidence and their interaction as fixed effects, and observer as a random intercept. The analysis revealed a significant main effect of confidence (Estimate = − 0.255, p < 0.001), indicating that higher confidence ratings were associated with shorter response times. The main effect of eccentricity was not significant (p = 0.507). Importantly, the interaction between eccentricity and confidence was significant (Estimate = 0.0030, p = 0.026), suggesting that the confidence related reduction in response time was attenuated at larger eccentricities. Fig. 2Individual confidence ratings across eccentricity conditions. Each line represents data from an individual observer, showing the mean confidence rating at two eccentricity conditions (5 and 15 deg). Red circles indicate the 5 deg condition, and blue circles indicate the 15 deg condition. Individual confidence ratings across eccentricity conditions. Each line represents data from an individual observer, showing the mean confidence rating at two eccentricity conditions (5 and 15 deg). Red circles indicate the 5 deg condition, and blue circles indicate the 15 deg condition. Fig. 3Individual response times and group means as a function of confidence rating and eccentricity. Response times are plotted across confidence ratings (1–4) for each eccentricity condition (5 deg: red; 15 deg: blue). Circles represent individual observers, and lines connect confidence levels within each observer. Bars indicate group means across observers. Individual response times and group means as a function of confidence rating and eccentricity. Response times are plotted across confidence ratings (1–4) for each eccentricity condition (5 deg: red; 15 deg: blue). Circles represent individual observers, and lines connect confidence levels within each observer. Bars indicate group means across observers. Perceptual confidence modulates speed judgment To quantify the effect of perceptual confidence on speed perception, we estimated the posterior distributions of the PSE and JND using a hierarchical Bayesian model. The model parameter estimates are summarized in Table 1. For PSE, the posterior median was 8.39 deg/s [7.79, 8.95] under low confidence and 7.93 deg/s [7.70, 8.14] under high confidence (Fig. 4). The posterior difference (High–Low) was negative (ΔPSE of − 0.45 deg/s [− 0.88, − 0.03]). This indicates that observers required a higher physical speed to perceive the stimulus as faster when their confidence was low, consistent with an underestimation of perceived speed under low-confidence conditions. For JND was substantially larger under low confidence (median: 2.72 deg/s [2.04, 3.89]) compared to high confidence (median: 1.06 deg/s [0.93, 1.21]; Fig. 4). The posterior difference was strongly negative (ΔJND = − 1.66 deg/s [− 2.75, − 1.04]). This indicates that perceptual sensitivity was markedly higher when observers reported high confidence. Together, these results demonstrate that perceptual confidence is associated with both reduced perceptual bias and increased perceptual sensitivity, consistent with the interpretation that confidence reflects trial-by-trial variations in internal sensory uncertainty. Table 1Parameter estimates from the hierarchical Bayesian psychometric model. The model assumed a Bernoulli likelihood with a logit link to predict binary speed judgments (“faster” = 1). Fixed effects included test stimulus speed, confidence, and their interaction. Random effects included varying intercepts and speed slopes across observers. Estimates represent posterior means, with effective sample size (ESS), convergence diagnostic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document}), and 95% credible intervals (CI) reported for each parameter.Fixed predicterEstimateESS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document} I-95%CIU-95%CIIntercept− 0.3131471.00– 0.740.15Speed0.8151961.000.561.08Confidence0.45121891.000.160.73Interaction (Speed: Confidence)1.26105571.001.021.51Random-effects parameterEstimateESS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document} I-95%CIU-95%CISD (Intercept)0.6135761.000.341.04SD (Speed)0.2936341.000.090.60Correlation (Intercept, Speed)– 0.1463501.00– 0.730.52 Table 1Parameter estimates from the hierarchical Bayesian psychometric model. The model assumed a Bernoulli likelihood with a logit link to predict binary speed judgments (“faster” = 1). Fixed effects included test stimulus speed, confidence, and their interaction. Random effects included varying intercepts and speed slopes across observers. Estimates represent posterior means, with effective sample size (ESS), convergence diagnostic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document}), and 95% credible intervals (CI) reported for each parameter.Fixed predicterEstimateESS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document} I-95%CIU-95%CIIntercept− 0.3131471.00– 0.740.15Speed0.8151961.000.561.08Confidence0.45121891.000.160.73Interaction (Speed: Confidence)1.26105571.001.021.51Random-effects parameterEstimateESS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document} I-95%CIU-95%CISD (Intercept)0.6135761.000.341.04SD (Speed)0.2936341.000.090.60Correlation (Intercept, Speed)– 0.1463501.00– 0.730.52 Parameter estimates from the hierarchical Bayesian psychometric model. The model assumed a Bernoulli likelihood with a logit link to predict binary speed judgments (“faster” = 1). Fixed effects included test stimulus speed, confidence, and their interaction. Random effects included varying intercepts and speed slopes across observers. Estimates represent posterior means, with effective sample size (ESS), convergence diagnostic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\mathrm{R}}}$$\end{document}), and 95% credible intervals (CI) reported for each parameter. Fig. 4Posterior distributions of perceptual bias (PSE) and sensitivity (JND) as a function of subjective confidence. Left panel: posterior distributions of the PSE for low- and high-confidence trials. Right panel: posterior distributions of the JND, reflecting perceptual sensitivity. Posterior distributions were derived from a hierarchical Bayesian psychometric model. Posterior distributions of perceptual bias (PSE) and sensitivity (JND) as a function of subjective confidence. Left panel: posterior distributions of the PSE for low- and high-confidence trials. Right panel: posterior distributions of the JND, reflecting perceptual sensitivity. Posterior distributions were derived from a hierarchical Bayesian psychometric model. The pupil reflects the perceptual confidence To examine the effect of confidence on phasic pupil response, a linear mixed model (LMM) was conducted with mean pupil slope and diameter as the dependent variables, and eccentricity and confidence as fixed effects, including their interaction, with observer included as a random effect. The model predictor estimates are summarized in Table 1. The time course of pupil responses is depicted in Fig. 5A. Pupil diameter began to rise following the onset of the test stimulus. Higher confidence ratings were associated with greater phasic dilation. These results indicate that pupil diameter reflects subjective confidence during perceptual decision-making. Figure 5 B and C show that pupil diameter increased as the confidence level increased. Each color dot represents the standardized mean pupil diameter for an individual observer, and lines show individual-level regression fits. Colored bars represent group means, showing an increase in pupil slope and diameter from confidence level 1 to 4 (Table 2). Fig. 5Time course of phasic pupil responses and their relationship with confidence ratings. (A) Grand-average time courses of pupil diameter for each confidence rating in the 5 deg (upper panel, red) and 15 deg (lower panel, blue) eccentricity conditions. The colored lines indicate the mean response time at each level of confidence. (B) Relationship between confidence ratings and pupil slope in the analysis window (500–1000 ms). Pupil slope was computed for each observer by fitting a linear regression to the pupil time course within this window. Dots represent individual observers’ slope values, bars indicate group means, and solid lines represent individual linear regression fits across confidence ratings. (C) Relationship between confidence ratings and mean pupil diameter in the analysis window (1000–1500 ms). Dots represent individual observers’ pupil diameter, Bars indicate group means, and solid lines represent individual linear regression fits across confidence ratings. Time course of phasic pupil responses and their relationship with confidence ratings. (A) Grand-average time courses of pupil diameter for each confidence rating in the 5 deg (upper panel, red) and 15 deg (lower panel, blue) eccentricity conditions. The colored lines indicate the mean response time at each level of confidence. (B) Relationship between confidence ratings and pupil slope in the analysis window (500–1000 ms). Pupil slope was computed for each observer by fitting a linear regression to the pupil time course within this window. Dots represent individual observers’ slope values, bars indicate group means, and solid lines represent individual linear regression fits across confidence ratings. (C) Relationship between confidence ratings and mean pupil diameter in the analysis window (1000–1500 ms). Dots represent individual observers’ pupil diameter, Bars indicate group means, and solid lines represent individual linear regression fits across confidence ratings. Table 2Summary of the linear mixed-effects model (LMM) predicting pupil slope and size. Both models included fixed effects of confidence, eccentricity, and their interaction, with a random intercept for the observer. Confidence significantly predicted both pupil slope and pupil size (p < 0.001), whereas neither eccentricity nor the interaction reached significance. Estimates are reported with standard errors (SE) and 95% confidence intervals (CI).AFormulaPupil slope ~ (Intercept) + Confidence + Eccentricity + (Confidence * Eccentricity) + (Intercept|Observer)PredictorEstimateSE p 95%CI(Intercept)– 0.41750.0986< 0.001– 0.2243– 0.6108Confidence0.12770.0291< 0.0010.18480.0706Eccentricity0.02710.13370.8390.2893– 0.2351Confidence*Eccentricity0.00330.04090.9370.0835– 0.0770BFormulaPupil size ~ (Intercept) + Confidence + Eccentricity + (Confidence * Eccentricity) + (Intercept|Observer)PredictorEstimateSE p 95%CI(Intercept)– 0.43810.0985< 0.001– 0.2451– 0.6312Confidence0.13700.0291< 0.0010.19410.0800Eccentricity0.02040.13360.8790.2823– 0.2415Confidence * Eccentricity– 0.00020.04090.9960.0799– 0.0804 Table 2Summary of the linear mixed-effects model (LMM) predicting pupil slope and size. Both models included fixed effects of confidence, eccentricity, and their interaction, with a random intercept for the observer. Confidence significantly predicted both pupil slope and pupil size (p < 0.001), whereas neither eccentricity nor the interaction reached significance. Estimates are reported with standard errors (SE) and 95% confidence intervals (CI).AFormulaPupil slope ~ (Intercept) + Confidence + Eccentricity + (Confidence * Eccentricity) + (Intercept|Observer)PredictorEstimateSE p 95%CI(Intercept)– 0.41750.0986< 0.001– 0.2243– 0.6108Confidence0.12770.0291< 0.0010.18480.0706Eccentricity0.02710.13370.8390.2893– 0.2351Confidence*Eccentricity0.00330.04090.9370.0835– 0.0770BFormulaPupil size ~ (Intercept) + Confidence + Eccentricity + (Confidence * Eccentricity) + (Intercept|Observer)PredictorEstimateSE p 95%CI(Intercept)– 0.43810.0985< 0.001– 0.2451– 0.6312Confidence0.13700.0291< 0.0010.19410.0800Eccentricity0.02040.13360.8790.2823– 0.2415Confidence * Eccentricity– 0.00020.04090.9960.0799– 0.0804 Summary of the linear mixed-effects model (LMM) predicting pupil slope and size. Both models included fixed effects of confidence, eccentricity, and their interaction, with a random intercept for the observer. Confidence significantly predicted both pupil slope and pupil size (p < 0.001), whereas neither eccentricity nor the interaction reached significance. Estimates are reported with standard errors (SE) and 95% confidence intervals (CI). Discussion The purpose of this study was to investigate whether subjective confidence modulates the perception of speed in peripheral vision. Based on Bayesian theories of perception, we hypothesized that lower confidence would be associated with a stronger bias toward perceiving slower speeds. Thus, modulation of speed perception by eccentricity would involve not only sensory input factors but also variations in subjective confidence. Our behavioral and physiological results supported this hypothesis. The subjective confidence was significantly lower in the 15 deg eccentricity condition compared to the 5 deg condition. Moreover, when confidence was low, observers were more likely to judge the stimulus as moving more slowly. This pattern aligns with previous findings that low contrast visual stimuli tend to be perceived as slower17. According to Bayesian inference frameworks, perception integrates sensory input with prior expectations to form a posterior estimate. When sensory information is imprecise or noisy as in the case of low contrast and luminance or peripheral vision, perception relies more heavily on prior knowledge4,18–20. Since the natural environment contains more stationary or slow-moving objects, humans are thought to possess a prior preference for lower speeds. Thus, when confidence in the sensory input is reduced in the peripheral, observers may rely more on this prior and perceive it as slower. However, it is important to note that not all manipulations of sensory uncertainty produce perceptual biases consistent with Bayesian predictions. For example, luminance-induced changes in perceived speed do not always follow the expected relationship between sensory uncertainty and perceptual bias21. Previous studies have shown that incorporating physiologically grounded constraints into Bayesian observer models improves their ability to predict perceptual performance. This suggests that perceptual inference is influenced not only by abstract probabilistic computations but also by the underlying neural mechanisms that shape sensory representations19. Accordingly, the present findings may reflect not only bayesian integration of sensory uncertainty and prior expectations but also eccentricity-dependent changes in neural encoding of motion signals. These results therefore suggest that the influence of sensory uncertainty on perceptual bias may depend on the specific neural mechanisms affected by different stimulus manipulations. Nevertheless, eccentricity related reductions in confidence were systematically associated with changes in both perceptual bias and sensitivity, consistent with bayesian accounts of perceptual inference under uncertainty. The decline in confidence at greater eccentricities can be attributed to a reduction in sensory precision. In our task, the reference stimulus was always presented at 5 deg, while the test stimulus appeared either at 5 or 15 deg. At the 15 deg condition, stimuli fall outside the fovea, resulting in a blurrier retinal image and reduced spatial resolution, which likely impairs the accurate perception of speed differences. We observed a significant effect of eccentricity on JND, indicating a reduction in sensitivity to perception. Furthermore, a significant relationship was observed between JND and PSE. The JND and slope of the psychometric function reflect how accurately observers can discriminate between reference and test stimuli. When sensory noise is greater, internal measurements or estimates derived from the same stimulus vary more across trials. As the variability of these estimates increases, the probability of correctly judging the difference between the two stimuli decreases. Consequently, the psychometric function becomes shallower, and the JND increases. Given that the reference stimulus was presented at a constant speed in the present study, the difference between eccentricity conditions likely reflects differences in the magnitude of sensory noise. Specifically, in the 15 deg condition, the sensory likelihood distribution is wider, reflecting reduced sensory precision. In fact, analysis using hierarchical Bayesian models revealed that when confidence was high, the posterior probability distribution of the estimated PSE was narrower. Conversely, when confidence was low, the posterior probability was broader and flatter. This indicates an increase in the uncertainty of sensory input. This sensory degradation could contribute to lower confidence in judgments under the 15 deg condition. After comparing the reference and test stimuli, observers provided a confidence rating for each judgment. Since there were seven conditions of test stimulus velocity, the speed difference between the reference and test stimuli served as a source of sensory evidence for the speed judgment. Importantly, because all velocity conditions were included and the distribution of velocity conditions did not differ across eccentricity conditions, the effect of eccentricity on confidence cannot be attributed to differences in speed evidence. After visual motion information is received in the retina, it is transmitted through the primary visual cortex to the MT/MST areas. These areas contain neurons that are selective for speed and direction, each exhibiting maximal firing rates at optimal speed22–24. Indeed, it has been reported that electrically stimulating MT neurons biases perception toward the speed preferred by those neurons25, suggesting a close relationship between activity in MT and speed perception. Anatomical and inactivation evidence indicates that MT receives dominant magnocellular drive via LGN and V1, and that magnocellular contributions are relatively preserved at moderate eccentricities26,27. These findings suggest that cortical input to MT remains mostly unchanged at 15 deg eccentricity compared to 5 deg eccentricity. These findings suggest that eccentricity-related effects may not be fully explained by a simple reduction in overall cortical input strength to MT. Rather, the eccentricity may alter the precision and shape of the sensory likelihood function by modifying tuning characteristics and collective coding structures. Previous studies investigating the effect of eccentricity on speed discrimination have shown that discrimination performance relative to a reference speed follows a U-shaped function, with the highest sensitivity observed at intermediate speeds (approximately 8–64 deg/s)28. Importantly, as eccentricity increases, discrimination performance deteriorates markedly in the low-speed range, whereas performance at higher speeds remains largely unaffected. Therefore, the U-shaped function shifts toward higher speeds with increasing eccentricity. The authors suggested that this shift may reflect a bias toward higher preferred speeds in speed-selective neurons within V1 and MT. Consistent with this interpretation, low-pass cells are more prevalent in central vision, whereas broadband cells are more dominant in peripheral representations within MT29. A similar pattern has been reported in V1, where low-pass cells are more common in the central and broadband cells are more frequently observed in the periphery30. Therefore, when comparing 5 and 15 deg eccentricities, the differences are unlikely to reflect a simple reduction in cortical input to MT. Rather than indicating a decrease in the strength of neuronal signals, they are more plausibly interpreted as changes in the tuning properties, precision, and speed-dependent information content of the neuronal populations used for speed estimation. Such changes would not only reduce the amplitude of the response but also alter the variability and distribution of neural responses in relation to speed. Within a Bayesian framework, speed perception can be described as the integration of a sensory likelihood and a prior distribution. The eccentricity-related shift of the U-shaped discrimination function toward higher speeds, together with the selective impairment at low speeds, suggests that changes in speed-tuning properties in V1/MT alter the shape of the likelihood in a speed-dependent manner. Specifically, in peripheral vision, the likelihood appears to broaden, indicating reduced sensory precision. As a result, confidence decreases, and the posterior estimate becomes more strongly influenced by a slow-speed prior, leading to systematic changes in perceptual bias. Physiological evidence from pupil dilation further supports the relationship between confidence and speed perception. Our pupil analysis revealed that phasic pupil dilation was significantly greater in trials rated as high in confidence. Recent research by Doll et al. (2024) demonstrated that both auditory cortical activity and pupil dilation increased when near-threshold auditory stimuli were perceived more strongly. These findings suggest that pupil dilation may serve as an indirect physiological marker of perceptual strength and confidence. In this light, our observation of larger pupil dilation during high-confidence trials may reflect a stronger or clearer underlying sensory representation. In the present study, two pupillary indices (the slope and mean pupil diameter) showed significant effects of subjective confidence, whereas no significant effect of stimulus eccentricity was observed. If pupil diameter directly reflected the sensory noise or perceptual uncertainty inherent in sensory input, one would expect systematic modulation under the 15 deg condition, where the reliability of visual information is reduced. However, no such effect was detected. These findings suggest that pupil responses don’t simply track the quality of sensory input, but rather reflect neural processes associated with subjective confidence. Pupil diameter is closely linked to the activity of the locus coeruleus, the primary source of cortical noradrenalin15. Noradrenaline has been reported to suppress neural noise while enhancing task-relevant neural responses, thereby increasing the signal-to-noise (S/N) ratio of neural activity. From this perspective, trials associated with higher subjective confidence may have involved elevated LC–NE activity, resulting in stronger suppression of neural noise and amplification of visual stimulus related signals, ultimately leading to an increased S/N ratio. Specifically, under 5 deg conditions, higher subjective confidence may have been accompanied by greater noise suppression, whereas under 15 deg conditions, lower confidence may have been associated with relatively weaker noise suppression. Taken together, these results raise the possibility that the LC–NE system, as indexed by pupil dynamics, does not merely reflect the amount of sensory noise in the input. Rather, it may dynamically regulate the effective S/N ratio within neural circuits, thereby modulating the generation of posterior probability estimates. A methodological limitation of the present study is that confidence ratings were collected after the perceptual speed judgment. Subjective confidence often mirrors first-order perceptual evidence, and therefore may not constitute an independent process31. In the present study, subjective confidence decreased with increasing eccentricity, suggesting reduced perceptual precision in peripheral vision. This pattern is consistent with Bayesian accounts in which increased sensory uncertainty influences perceptual inference, although eccentricity-related changes in sensory encoding may also contribute. Future studies manipulating stimulus properties, such as contrast or size, to equate confidence across eccentricities would help further clarify these relationships. Nevertheless, the present findings provide behavioral and physiological evidence that confidence systematically varies with eccentricity and is associated with corresponding changes in perceptual sensitivity and bias. In our paradigm, confidence likely reflects the observer’s internal estimate of decision reliability based on sensory evidence. Although neural activity was not directly measured, the observed behavioral and physiological patterns are consistent with established principles of sensory uncertainty and perceptual inference. Overall, the results demonstrate a systematic relationship between eccentricity, confidence, and perceptual performance, consistent with Bayesian models of sensory uncertainty. Conclusion The present study demonstrates that subjective confidence plays a role in modulating perceived speed in peripheral vision. When confidence is low, whether due to sensory degradation, observers tend to perceive moving stimuli as slower, which is consistent with Bayesian perceptual models. These findings highlight the importance of considering both sensory and cognitive factors in understanding visual motion perception and offer insight into the neural mechanisms linking confidence, eccentricity, and speed estimation. Methods Observers Ten observers [five men, mean age: 26.3 (SD: 2.4) years] with normal or corrected-to-normal vision and no known visuomotor deficits participated in the study. The observers were diagnosed with neither stereoscopic problems nor strabismus. All observers were right-handed. All methods were carried out in accordance with relevant guidelines and regulations. All the observers provided written informed consent in accordance with the Declaration of Helsinki. All protocols were approved by the Research Ethics Committee of the Institute of Health and Sport Sciences, University of Tsukuba. Apparatus The experiment was conducted in a dimly lit room. The observer sat approximately 57 cm in front of the monitor, with their head on a chinrest to suppress head movement. Visual stimulus was generated by Psychtoolbox extensions in MATLAB (MathWorks, Natick, MA) and presented on the 26.5-in. LCD monitor (MPG 271QRX QD-OLED, Micro-Star INT’L CO., LTD., Taiwan; refresh rate: 360 Hz; spatial resolution: 2560 * 1440 pixels). Eye movements and pupil size were detected with a video-based eye-tracking system (Eyelink 1000 plus, SR Research Ltd., Canada; sampling rate 1000 Hz) from the right eye. The eye position signals were detected by reflections of infrared light on the pupil and cornea, and a black image of the pupil was captured by the camera. Before starting each block, the eye position was calibrated by fixing the gaze on the nine calibration points on a gray background in a binocular condition. Stimulus and procedure The observers performed a speed discrimination task using a two-alternative forced-choice method (Fig. 6). The timing of the start of each trial was determined by the observer. The trial started when the observer pressed the upper central button on the keyboard with five buttons (MilliKey Response Box, LabHackers Research Equipment, Canada). Then, a fixation point (0.5 deg) and reference stimulus appeared in the center of the monitor for 1500 ms, and the observer was instructed to fixate on the fixation point. After that, the reference stimulus was moved at random times between 1500 and 2000 ms. In this study, the aim was to verify the effect of eccentricity, where a visual stimulus moved to maintain a constant eccentricity (distance from the fixation point) was used. After the reference stimulus was presented, a 1000 ms blank screen was shown, followed by the appearance of the fixation point and test stimulus. After 1500 ms, a test stimulus that moved around the fixation point while maintaining a constant eccentricity was presented. The observer was asked to compare whether the test stimulus was faster or slower than the reference stimulus, and when they had made their decision, they responded using the keyboard corresponding to their decision. The test stimulus continued to move until the observer responded. The reference stimulus was always at an eccentricity of 5 deg and moved at a speed of 8 deg/s. The test stimulus was at an eccentricity of 5 or 15 deg and moved at one of the following speeds: 5, 6, 7, 8, 9, 10, or 11 deg/s, chosen at random. To account for the effects of orientation and adaptation, the stimulus was set to alternate between clockwise and counterclockwise movement. The start position of the visual stimulus was also randomized. After blank, the degree of confidence in the judgment was reported by pressing the key corresponding to the value of 1, 2, 3, or 4. The score means that 1 is the lowest confidence and 4 is the highest confidence. Fig. 6Experimental paradigm. (A) Experimental paradigm. The fixation point is presented at the center of the monitor (0 deg) and the reference stimulus is presented at a random position with 5 deg eccentricity. The test stimulus is presented at a random position with an eccentricity of 5 or 15 deg. (B) Example of eccentricity for a monitor. The fixation point, the reference and the test stimulus are white on a gray background. (C) The average eye position during the presentation of the reference and test stimulus, time normalized to 100 points. The black line represents the average across observers. Each colored line represents the average eye position of an individual observer. Experimental paradigm. (A) Experimental paradigm. The fixation point is presented at the center of the monitor (0 deg) and the reference stimulus is presented at a random position with 5 deg eccentricity. The test stimulus is presented at a random position with an eccentricity of 5 or 15 deg. (B) Example of eccentricity for a monitor. The fixation point, the reference and the test stimulus are white on a gray background. (C) The average eye position during the presentation of the reference and test stimulus, time normalized to 100 points. The black line represents the average across observers. Each colored line represents the average eye position of an individual observer. There was a total of fourteen conditions [two eccentricities (5 or 15 deg) * seven test speeds (5, 6, 7, 8, 9, 10 and 11)], and each condition was conducted 20 trials. Therefore, each observer conducted 280 trials. There were 28 trials per block, and there were 10 blocks in total. Data analysis The eye position was processed using a low-pass filter with a cutoff frequency of 30 Hz, and the pupil size was processed using a band-pass filter with a cutoff frequency of 0.05–4 Hz. Furthermore, to normalize the change in pupil diameter, it was standardized based on the pupil diameter of each observer throughout the experiment and set so that the minimum value was zero and the maximum value was 1. Blinks during the trial were detected using the Eyelink’s automatic detection algorithm, and eye position and pupil diameter data from 200 ms before to 200 ms after each blink were linearly interpolated. If the position of the fixation changed by more than 2 deg from the time the reference stimulus was presented until it was occluded or the test stimulus was presented until the subject pressed the button for speed judgment, it was excluded from the analysis as an invalid trial. A total of 94 trials (3.36%) were excluded from the 10 observers, and 2706 trials (96.64%) of the data were used for analysis. Psychometric functions were fitted to each observer’s speed judgement data using logistic regression with a binomial distribution and logit link function (MATLAB glmfit function, maximum likelihood estimation). From these functions, the point of subjective equality (PSE) and the just-noticeable difference (JND) were estimated. The PSE was defined as the test stimulus speed at which observers judged the test and reference stimuli as equal with 50% probability. Therefore, a higher PSE than the reference stimulus indicates that the test stimulus is perceived as slower, whereas a lower PSE indicates that it is perceived as faster. The JND was defined as the difference between the test stimulus speeds corresponding to 75% and 25% probability of judging the test stimulus as faster than the reference stimulus. The JND represents sensory uncertainty and serves as an index of discrimination ability, as it reflects the slope of the psychometric function. Larger JND values correspond to shallower slopes and greater sensory uncertainty, indicating poorer discrimination ability, whereas smaller JND values correspond to steeper slopes and lower sensory uncertainty, indicating better discrimination ability. PSEs and JNDs were then compared between the two eccentricity conditions (5 deg vs. 15 deg) using a paired-samples t-test. To examine the relationship between changes in sensory uncertainty and perceptual bias, robust linear regression was performed using iteratively reweighted least squares with bisquare weighting, as implemented in MATLAB (fitlm function). This approach reduces the influence of outliers by assigning lower weights to observations with large residuals, resulting in more reliable parameter estimates. To examine the effects of perceptual confidence on speed judgments, we modeled speed discrimination responses (slow = 0, fast = 1) using hierarchical Bayesian logistic regression implemented in the brms package, which interfaces with Stan. Speed judgments were modeled using a Bernoulli likelihood with a logit link function. Confidence (low: < = 2, high: > = 3) and centered test stimulus speed were included as fixed effects, along with their interaction, to assess whether perceptual confidence modulated the psychometric function. Random intercepts and random slopes for test stimulus speed were specified for each observer to account for between observer variability in baseline response bias and sensitivity. Weakly informative priors were used to regularize estimation and improve numerical stability. Fixed-effect coefficients and intercepts were assigned normal priors centered at zero, while standard deviations of random effects were assigned exponential priors, and correlation matrices were assigned LKJ priors. Posterior distributions were estimated using Markov chain Monte Carlo (MCMC) sampling with four chains, each consisting of 4,000 iterations including 1,000 warm-up iterations. Convergence was assessed using the Gelman-Rubin statistic (R̂), and all parameters showed good convergence (R̂ < 1.01) and effective sample sizes (Bulk- and Tail- ESS > 1000). Posterior distributions of psychometric parameters were used to compute derived quantities, including the PSE and JND, allowing direct comparison of perceptual bias and sensitivity between confidence conditions. Effects were quantified using posterior medians and 95% credible intervals, and effects were considered credible when credible intervals excluded zero. To examine the effects of eccentricity and perceptual confidence on pupil responses, we fitted linear mixed-effects models. Mean pupil diameter (1000–1500 ms post-stimulus onset) and pupil slope (500–1000 ms post-stimulus onset) were analyzed as dependent variables in separate models. Pupil measures were z-scored within each observer prior to analysis. Each model included fixed effects of eccentricity, confidence, and their interaction, with observer specified as a random intercept. Statistical analyses were performed with MATLAB and R with a significance level of 0.05.