Response expectations shape serial dependence and stimulus processing

serial dependence, inverted encoding, EEG, perceptual decision-making 29 *Corresponding author: [email protected] 30 31 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 2

32 Perceptual decisions are biased by recent history, yet the balance between attractive and repulsive 33 effects varies across contexts. Here, we tested whether trial -by-trial response expectations shape the 34 direction of history biases in serial dependence during orientation reproduction. Behaviorally, we 35 found that no -response trials —especially when rare —reduced attractive biases and enhanced 36 repulsive biases. EEG results revealed stronger evoked responses and amplified neural 37 representations for stimuli following no -response trials. Together, these findings suggest that 38 interrupting the perception –action cycle fosters a state of re -engagement with current input and 39 disengagement from past stimuli, indicating that serial dependence is a flexible process dynamically 40 modulated by task expectations and transient shifts in sensory processing. 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 3

63 Perceptual decisions are shaped not only by current sensory input but also by the temporal structure 64 and history of recent events. A well -known example is serial dependence, in which decisions about 65 the current stimulus are biased by stimuli encountered a few trials before 1. 66 Over the past decade, numerous studies have documented serial dependence across a wide 67 range of visual tasks. These studies typically reported attractive biases, whereby visual decisions are 68 biased toward prior stimuli see 2,3 for reviews . A classic case is the orientation adjustment task: when 69 reproducing the orientation of the current stimulus, particularly with weak, uncertain , and brief 70 stimuli 4–7, observers make systematic errors in the direction of the orientations presented on recent 71 trials. Similar effects have been observed for other basic features such as motion and color, as well 72 as for more complex stimuli such as facial expressions, emotions, or even aesthetic preferences 8–11, 73 suggesting that serial dependence reflects a general characteristics of human perception and cognition 74 12. 75 Despite its pervasiveness across modalities and even species 13,14, the magnitude and direction 76 of serial dependence vary considerably across individuals and task conditions 3,15–17. In some cases, 77 serial dependence is dominated by repulsive biases—perceptual reports are biased away from 78 previous stimuli —and overall behavioral data suggest that attractive and repulsive components 79 coexist and interact at the individual level 15,18–24. These two opposite forms of serial dependence are 80 thought to arise from influences of prior events at different processing stages 18,25,26, yet the conditions 81 under which each dominates remain unclear. 82 One factor proposed to influence this balance is the continuity of perception –action cycles. 83 Most studies reporting attractive biases involve sequences of stimuli followed by overt responses. 84 Manipulations of response requirements, however, show that attraction is reduced —or can even 85 reverse into repulsion—following trials in which no response was required 15,18,27,28. For example, in 86 orientation adjustment tasks, observers typically show attractive serial dependence after runs of 87 response trials, whereas the same task can yield repulsive biases when the immediately preceding 88 trial required no response 15,18. This pattern suggests that response requirements exert at least a 89 modulatory effect on how previous stimuli influence current decisions, but the underlying 90 mechanisms remain debated 1,18,27–30. 91 We hypothesized that the modulatory effects of prior response requirements reflect trial -by-92 trial expectations that shape processing of the current stimulus and, consequently, the influence of 93 past events. In paradigms manipulating response requirements, participants may anticipate a response 94 trial following a no -response trial—either due to perceived regularities in alternating events 31 or 95 simply because response trials typically occur more frequently than no -response trials 1,18,27. This 96 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 4 expectation to deliver a response could lead to a re -engagement with the task and stimuli, thereby 97 reducing the influence of prior trials or even promoting a stronger tendency to discriminate the current 98 stimulus from previous ones, potentially resulting in repulsive biases. In other words, interrupting the 99 perception–action cycle may induce a state of stronger engagement with the current stimulus and 100 relative disengagement from past events. 101 We tested this hypothesis using behavioral and EEG data. In a first behavioral experiment, we 102 manipulated the ratio of response to no -response trials across blocks of an orientation adjustment 103 task. The rationale is that if no -response trials induce an expectation of an upcoming response, this 104 effect should depend on response frequency: the rarer the no -response trials, the stronger the 105 expectation following them, and consequently, the greater the modulation of serial dependence. In a 106 second experiment, we analyzed EEG data from a task in which response and no -response trials 107 occurred with equal probability. We examined overall changes in EEG activity and in stimulus 108 representations to identify neural correlates of the modulation of serial dependence following no -109 response trials. 110 Across both experiments, serial dependence patterns were systematically altered after no -111 response trials, showing reduced attraction and increased repulsion. This effect was evident when no-112 response trials were rare (Experiment 1) and also when they were equally likely as response trials 113 (Experiment 2). In the latter case, EEG analyses revealed neural signatures indicative of increased 114 attentional engagement and enhanced stimulus processing after no-response trials, consistent with the 115 behavioral reduction in attraction and increase in repulsion. Together, these findings suggest that trial-116 by-trial expectations about response requirements modulate serial dependence: no -response trials 117 appear to trigger a transient reset or re -engagement process that shifts the weighting of past versus 118 current sensory information. These results indicate that serial dependence is not a fixed property of 119 perceptual decision-making but a flexible phenomenon shaped by cognitive state and task context, 120 offering new opportunities to relate behavioral variability and individual differences to fluctuations 121 in internal states. 122 123

124 Experiment 1 125 Participants performed an orientation adjustment task under three response-frequency conditions, run 126 in separate blocks: a low -response condition (25% response trials), an equal -response condition 127 (50%), and a high-response condition (75%). In no-response trials, participants were presented with 128 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 5 a neutral stimulus (a circular frame without response cues , Figure 1A ) and instructed to passively 129 wait for the next trial. 130 We focused on serial dependence in trials following a response (R1–R1) and those following 131 a no-response trial (R0–R1; see Figure 1B) and assessed how these effects varied across response -132 frequency conditions. To this aim, we fitted each condition data with an extended derivative -of-133 Gaussian (δoG) model, implemented as the sum of two δoG functions (see Methods). This approach 134 allows both attractive and repulsive components of serial dependence to be captured within the same 135 function 32. 136 To test whether serial dependence was modulated by the previous trial type (response vs. no-137 response), we included a trial-type modulator on the amplitude and width parameters of the two δoG 138 components. We then compared this full model to a reduced model without the trial -type modulator 139 using the Bayesian Information Criterion (BIC), computing the difference ΔBIC = BIC( reduced) – 140 BIC(full) (see Methods). Positive ΔBIC values indicate evidence favoring the more complex model, 141 whereas negative values favor the simpler one. 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 6 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 Figure 1. Experiment 1: Paradigm and behavioral results. A) Example of the sequence of events in the two trial types. In 180 R1, an oriented Gabor was followed by a mask, and after a brief blank interval participants were presented with a response 181 tool—a circular frame with two triangles on opposite ends —which they rotated to reproduce the perceived orientation. 182 In R0, no triangles were shown on the response tool, and participants passively awaited the next trial, as no response was 183 required. B) Main conditions of interest for measuring serial dependence. In R0 –R1, serial dependence was measured 184 from a previous trial that required no response; in R1 –R1, it was measured between two consecutive response trials. 185 Across separate blocks (bottom bar plot), we varied the ratio of response to no -response trials: 25% responses (rare 186 response trials), 50% responses (equal proportion of response and no -response trials), and 75% responses (frequent 187 responses, rare no-response trials). C) Serial dependence in orientation adjustment errors across blocks. Errors (y-axis) 188 are plotted as a function of the orientation difference between the stimulus in the previous and current trial (Δ, x -axis). 189 Data show the circular running average of aggregated single-trial errors across participants, with shaded areas indicating 190 ±1 SD. Fits result from model comparisons of different formulations of the sum of two derivative -of-Gaussian (δoG) 191 functions, with or without an additional effect of trial type (R0 –R1 vs. R1–R1; see Methods). In plots showing a single 192 curve and fit (25% and 50% conditions), model comparison favored a simpler model without a trial -type effect. In 193 contrast, for the 75% condition (two curves), the preferred model included a trial -type effect, indicating differences in 194 serial dependence between R0 –R1 (gray) and R1 –R1 (green) trials. D) Comparison of serial dependence between 195 conditions with approximately equal numbers of trials: R0 –R1 under the 25% response condition (orange) and R0 –R1 196 under the 75% response condition (gray). This contrast isolates the effect of response frequency on serial dependence (see 197

for model comparison). 198 199 Figure 1C shows the best -fitting model results across conditions. Evidence favored the full 200 model in the 75% response condition (ΔBIC 75% = +3.19), indicating an effect of the previous trial 201 type on serial dependence. In contrast, the 25% and 50% conditions showed evidence favoring the 202 simpler model (ΔBIC25% = –12.5, ΔBIC50% = –8.70). 203 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 7 In the 25% and 50% conditions serial dependence was predominantly attractive, exhibiting 204 the typical positive bias toward the previous orientation. In the 75% condition, serial dependence was 205 attractive following response trials (R1 –R1). However, following no -response trials (R0 –R1), the 206 pattern showed stronger repulsive tails, consistent with prior findings 18,33. 207 Because the number of R0 –R1 and R1 –R1 trials differed across response -frequency 208 conditions (e.g., R0 –R1 trials were rarer in the 75% condition), we performed an additional 209 comparison equating trial counts. Specifically, we compared R0–R1 trials from the 25% condition (≈ 210 18.2% trials) to R0–R1 trials from the 75% condition (≈ 18.1% trials). This confirmed a robust effect 211 of overall response frequency (ΔBIC = +4.74), with stronger repulsive components observed when 212 no-response trials were rare (75% condition). 213 Taken together, these behavioral results indicate that serial dependence is modulated by the 214 presence or absence of a response on the preceding trial, and that this modulation depends on the 215 global frequency of response trials. When no -response trials are rare, serial dependence following 216 such trials shows reduced attraction and enhanced repulsion, particularly for larger Δ values. 217 Experiment 2 218 What can explain the effect of previous response requirements and their dependence on the frequency 219 of responses? One possibility is that after rare no -response trials, participants come to expect that a 220 response will be required on the next trial. This interpretation aligns with extensive evidence showing 221 that trial-by-trial expectations modulate behavior across perceptual and decision -making tasks 31,34. 222 Such expectations may prompt a re-engagement with the task and enhanced processing of the current 223 stimulus following a no-response event. 224 To further investigate this possibility, we analyzed an independent EEG dataset from a similar 225 orientation adjustment task in which response and no-response trials occurred with equal probability 226 (50%; see Methods and Figure 2A ). Behaviorally, participants again showed modulation of serial 227 dependence by previous trial type, with reduced attraction and increased repulsion after no-response 228 trials (ΔBIC = +8640; Figure 2B), consistent with Experiment 1 despite the different response 229 frequency conditions. 230 231 232 233 234 235 236 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 8 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 Figure 2. Experiment 2: Paradigm and behavioral results. A) Sequence of events in the two trial types. In this EEG 255 experiment, the response frequency was 50%, meaning R0 and R1 trials were equally likely to occur. In R1 trial s, 256 participants reproduced the perceived orientation by rotating a response bar, whereas in R0 trial s, a black disk was 257 presented instead of the response tool. B) Serial dependence in orientation adjustment errors in the two types of trials 258 (R0–R1, following no-response trials, in gray; R1–R1, following response trials, in green). Dots represent the mean error 259 across participants for each Δ value (shown in discrete steps; see Methods). Error bars indicate SEM corrected for repeated 260 measures. The curves show the predictions of the best-fitting model, following the same approach as in Experiment 1. 261 262 We next focused on the analysis of EEG activity time -locked to the current stimulus 263 comparing the two trial types . Global Field Power (GFP, see Methods) was significantly higher 264 following no-response trials (R0–R1) than following response trials (R1–R1) in a time window from 265 155 ms to 310 ms after stimulus onset (Figure 3A). This time range and the associated topography 266 difference between conditions (Figure 3B) appear to be suggestive of relatively late attentional 267 modulations (e.g., involving time windows characteristics of the P1/N1 complex). A linear model 268 predicting trial-by-trial bias from GFP and trial type (R0 –R1 vs. R1 –R1, see Methods ) revealed a 269 significant positive effect of trial type, indicating larger positive (attractive) bias in orientation 270 adjustment responses in the R 1–R1 condition compared to R 0–R1 (β = 0.83 ± 0.21 SE, t(5517) = 271 3.92, p < .001). In contrast, GFP fluctuations were negatively associated with bias magnitude (β = –272 0.85 ± 0.34 SE, t(5517) = –2.48, p = .013), suggesting that higher GFP was predictive of stronger 273 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 9 repulsive biases (Figure 3C). The overall model fit was modest (R² = 0.004, F(2, 5517) = 11.1, p < 274 .001). 275 Finally, we assessed how the current trial orientation was represented in EEG activity using 276 an inverted encoding model (IEM; see Methods). Orientation decoding was significant from 120 ms 277 to 485 ms post-stimulus (Figure 3D-E). Importantly, decoding accuracy was higher for R0 –R1 than 278 R1–R1 trials in the time window from 215 ms to 260 ms (Figure 3E), accompanied by increased 279 amplitude of the reconstructed channel response functions (Figure 3F), indicating stronger 280 orientation-selective responses following no-response trials. Together, these results suggest that after 281 no-response trials, EEG activity and the strength of orientation representations are enhanced. 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 Figure 3. Experiment 2: EEG results. A) Global Field Power (GFP) time -locked to the current stimulus by trial type, 302 averaged across participants. GFP was higher in the R0 –R1 condition (gray) compared to the R1 –R1 condition (green) 303 within a time window from 155 to 310 ms. Shaded regions indicate SEM corrected for repeated measures. The light gray 304 trace shows the GFP difference between conditions, with the significant time window marked by a black horizontal line 305 (see Results). B) Scalp topography of the GFP difference between trial types, with electrodes showing peak positive or 306 negative differences (top 25% percentile) highlighted by white circles. C) Linear model coefficients (error bars are the 307 95% CI) predicting bias in adjustment errors (positive = attraction toward the previous stimulus; negative = repulsion) as 308 a function of trial type and trial-by-trial GFP fluctuations within the significant time window identified in (A). Asterisks 309 denote significant coefficients (p < .05). D) Inverted Encoding Model (IEM) results showing reconstructed channel 310 responses over time (x = time relative to stimulus onset; y = IEM channels), separately for R0–R1 and R1–R1 trials, and 311 their difference (bottom row). The gray shaded rectangle indicates the GFP difference window reported in (A). E) 312 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 10 Decoding scores (derived from the slope of the channel tuning function; see Methods) for the two trial types, using the 313 same color coding as in (A). Decoding of the current orientation was significant from 125 ms onward (orange line) and 314 differed between trial types in a narrower window from 215 to 260 ms (black line). Shaded regions are corrected SEM. 315 F) During the time window showing a significant trial -type effect on decoding, channel tuning functions (CTFs) were 316 amplified in the R0 –R1 condition (left; shaded regions = corrected SEM) , indicating steeper slopes and enhanced 317 orientation representations (right; dots represent individual participants; error bars show 95% CI corrected for repeated 318 measures). 319 320

321 In two experiments, we investigated how trial -by-trial response expectations influence serial 322 dependence and EEG activity. In a behavioral study (Experiment 1), we manipulated the proportion 323 of response trials in separate blocks and found that serial dependence was modulated primarily after 324 rare no-response trials (i.e., in the 75% response block) . In the EEG study (Experiment 2), we 325 measured changes in EEG signal strength and orientation decoding in the current trial, depending on 326 the previous trial response requirement. Again, we observed a modulation of serial dependence in 327 behavior, accompanied by distinct changes in EEG activity and its orientation -selective content. 328 Together, these results suggest that no -response trials trigger a re -engagement with the task and 329 enhanced processing of the current stimulus. 330 In Experiment 1, the effect was evident in the condition where no -response trials were rare 331 (the 75% response condition), manifesting as a reduction in attractive serial dependence and an 332 increase in repulsive bias after no -response trials. The attractive and repulsive components of serial 333 dependence have previously been linked to effects of prior history at different processing stages: 334 repulsion has been linked to adaptation-like aftereffects at sensory levels, whereas attraction has been 335 related to the persistence of representations at higher levels such as decision -making or working 336 memory 18,25,26,35,36. Within this framework, the dominance of one component over the other would 337 depend on factors such as the strength of the previous stimulus (e.g., duration, contrast) and the extent 338 to which it was attended, maintained in memory, or reported. 339 Although this hierarchical distinction remains debated see 3 for a review , our results are broadly 340 consistent with the idea that prior responses modulate the balance between these opposing forces. 341 One possibility is that withholding a response on the previous trial, and the explicit cue indicating 342 that no response is required, leads to a “reset” or active removal of the most recent stimulus trace 15,37. 343 In a hierarchical view, this would reduce the influence of prior events at higher-level processing stages 344 (attractive biases) and consequently allow low -level adaptation to dominate, producing repulsive 345 effects 15,18,24. Alternatively, active removal of no longer relevant information from working memory, 346 as well as stimulus devaluation processes following response withholding, could themselves induce 347 repulsive biases without invoking sensory adaptation mechanisms 15,18,37. 348 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 11 However, these interpretations alone cannot explain why the effects of previous trial type, at 349 least in Experiment 1, emerged only when no -response trials were rare. Consistent with our 350 hypothesis, we propose that these effects reflect expectations to respond on the current trial following 351 a no-response trial. When response trials are frequent, such expectations are stronger, leading to a re-352 engagement with the task and current stimulus. This, in turn, enhances stimulus processing, reduces 353 the influence of prior events, and increases the tendency to discriminate the current stimulus from 354 previous ones—manifesting as stronger repulsive biases. One plausible mechanism underlying this 355 effect is a refocusing of attention toward the current sensory input. This interpretation aligns with 356 evidence that attractive serial dependence tends to occur under conditions of reduced sensory 357 processing or higher uncertainty, for instance when stimuli are weak, brief, or task resources are 358 limited 6,15,38. Our findings demonstrate that such modulations can arise entirely from internal states 359 of the observer, driven by expectations about upcoming task demands, even when the physical 360 properties of the stimuli are constant see also 39,40. 361 In Experiment 2, the EEG results provided a further, complementary characterization of this 362 phenomenon. We observed increased overall EEG signal strength and higher orientation decoding 363 scores for the current stimulus following no -response trials. These findings, evident in the GFP and 364 IEM analyses respectively, indicate that orientation -selective responses were amplified after no -365 response trials. The time window of the GFP effect ( 155-310 ms post -stimulus) and its scalp 366 distribution correspond to typical EEG components modulated by attention, including the P1 and 367 visual N1 . Likewise, the IEM results within this period revealed increased amplitude of the 368 reconstructed channel response functions after no -response trials, consistent with attentional gain 369 modulation 41. Together, these results support the interpretation that following no -response trials, 370 enhanced attentional engagement and increased allocation of processing resources strengthen the 371 neural representation of the current stimulus. 372 It is worth noting that , behaviorally, these effects emerged under different conditions in the 373 two experiments: when no-response trials were rare (Experiment 1, 75% response) and when response 374 and no-response trials were equally likely (Experiment 2, 50%). Although the precise nature of this 375 difference remains to be determined, the consistency of the pattern suggests a robust phenomenon see 376 also Figure 3 in 18 . It is possible that broader contextual factors , such as the overall structure of the 377 experiment (multiple response-ratio blocks in Experiment 1 vs. a fixed 50% design in Experiment 2), 378 might have contributed to participants’ expectations, pointing to an additional higher-level contextual 379 modulation of serial dependence by experimental design itself. 380 We propose that expectations naturally arising in contexts with intermixed response and no -381 response trials may be what induce these fluctuations in attention and task engagement, which in turn 382 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 12 modulate the balance between attractive and repulsive serial dependence. This perspective departs 383 from most prior studies, which have treated previous responses either as motor confounds 1,27 or as 384 key contributing factors in serial dependence 18,42, without identifying the processes responsible for 385 trial-by-trial modulations. 386 A central implication is that attractive serial dependence—in some frameworks interpreted as 387 evidence for a dedicated mechanism stabilizing perception 1,43—does not result from a fixed and hard-388 wired process. Rather, it is shaped by top -down factors such as dynamic fluctuations in task 389 engagement and expectation, beyond a fixed mechanism. In this sense, our findings support the view 390 that attractive serial dependence emerges under specific internal states that favor the persistence and 391 interference of previous task content with current decisions 39,40. These states alternate with states that 392 place greater weight on current sensory input. Our manipulations may have specifically targeted this 393 transition: from a state promoting attraction to prior stimuli (during continuous responding) to one 394 enhancing current-stimulus processing and reducing interference from prior events (following a no -395 response trial, when a response is expected next). To our knowledge, we provide the first evidence of 396 neural correlates associated with this putative state transition, marked by increased EEG activity and 397 amplified orientation tuning for the current stimulus. Previous decoding studies have primarily 398 focused on decoding residual traces of past stimuli or the bias they induce in current neural 399 representations 22,26,35,44, but none have reported changes in current -stimulus representations linked 400 to trial -by-trial fluctuations in serial dependence. The demonstration that serial dependence is 401 dynamically modulated by internal states—with identifiable EEG signatures of these fluctuations —402 offers a novel framework for understanding inter-individual variability in serial dependence, as well 403 as deviations observed in clinical populations 28,45,46. 404 In sum, we show that trial-by-trial response expectations represent a key source of modulation 405 in serial dependence, with no-response trials triggering re -engagement with the task and enhanced 406 processing of the subsequent stimulus, ultimately reweighting the balance between attractive and 407 repulsive influences from recent history. 408 409

410 Experiment 1 411 Participants 412 The study was conducted at the École polytechnique fédérale de Lausanne (EPFL, Lausanne, 413 Switzerland). Twenty-four healthy participants (11 females, age range: 21 -37 y) were recruited for 414 monetary reward (20 CHF/hour). All participants had normal or corrected-to-normal vision according 415 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 13 to the Freiburg Visual Acuity test 47 and received written informed consent before the experiment. 416 The study was approved by the local ethics committee in accordance with the Declaration of Helsinki 417 (World Medical Organization, 2013). 418 419 Apparatus 420 All experiments were run on a gamma-corrected VG248QE monitor (resolution: 1920 x 1080 pixels, 421 refresh rate: 120 Hz) in a darkened room. Stimuli were generated with custom -made scripts written 422 in MATLAB (R2013a) and the Psychophysics Toolbox and presented on a grey background (62.66 423 cd/m2). Participants sat at 57 cm from the computer screen, with their head on a chin rest. 424 425 Stimuli and procedure 426 Figure 1A depicts the sequence of events in an experimental trial. Each trial started with a fixation 427 spot shown for 1000 ms. After fixation, a low contrast (peak contrast of 25% Michelson) and low 428 spatial frequency (0.33 cycles per degree) Gabor (Gaussian envelope: 1.5°) was shown for 500 ms, 429 followed by a high -contrast noise mask (95% Michelson) for other 500 ms. The Gabor could have 430 any orientation in the 0 -179° circular space, randomly assigned on each trial. A blank interval (500 431 ms) preceded the appearance of the response tool or the stop signal. The response tool was made of 432 a gray circular frame (diameter of 4°) with two triangles (e.g., arrowheads) positioned on the outline, 433 at two equidistant extremities of an imaginary line. In stop trials, the triangles were not shown, and 434 the circular frame alone indicated the absence of a response. 435 In response trials, participants had to rotate the orientation of the imaginary line to match the 436 perceived orientation of the Gabor, by moving the mouse in the upward and downward directions, 437 confirming the response with a left click. In stop trials, they were instructed to withhold the response 438 and wait until the next trial. The waiting time before the next trial was calibrated to the running 439 average of individual adjustment times. 440 The main manipulation consisted in varying the proportion of response and stop trials in three 441 separate blocks. As shown in Figure 1C, the frequency of responding was either 25%, 50%, or 75%. 442 The order of blocks was counterbalanced across participants. 443 Before the experiment, all participants were provided written and verbal instructions. To 444 ensure that participants understood the task, they performed a brief practice before the experiment, 445 under the supervision of the experimenter. The experiment consisted of 3 blocks, for a total of 600 446 trials, lasting approximately 1 hour. 447 448 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 14 Behavioral data analysis 449 The analysis of behavioral serial dependence focused on adjustment errors, defined as the acute 450 angular difference between the reported orientation and the stimulus orientation on each trial. Prior 451 to the analysis, errors were cleaned through a multi-step procedure. 452 First, trials with absolute errors larger than 45° were classified as lapses and excluded 28. Trials 453 with adjustment response times longer than 10 seconds were also excluded. Second, the remaining 454 errors were demeaned and corrected for systematic orientation -dependent biases using a sinusoidal 455 residualization approach 18,48. Specifically, a sum of six sine and cosine functions was fitted to the 456 error as a function of the current orientation using the MATLAB fit.m function with the model 457 specification “sin6”, and the resulting fit was subtracted from the errors. Next, errors identified as 458 statistical outliers—defined as values more than 1.5 interquartile ranges above the upper quartile or 459 below the lower quartile —were removed using the isoutlier function in MATLAB with the 460 “quartiles” method. Across all steps, less than 5% of response trials were excluded from the analysis. 461 The remaining cleaned data were used to assess serial dependence effects. 462 To quantify serial dependence and compare its pattern across conditions, we modeled the 463 relationship between current -trial errors and Δ (the relative orientation between the current and 464 previous stimuli), using a parametric model based on the first derivative of a Gaussian (δoG) 1: 465 466 𝑒𝑟𝑟𝑜𝑟 = ∆𝛼𝑤𝑐𝑒!(#∆)! 467 [1] 468 where 𝑐 = √' ("#.% is a normalization constant, 𝑤 controls the inverse width of the curve, and α is the 469 amplitude parameter indicating the strength and direction of the bias. Positive values of α reflect an 470 attractive bias (i.e., errors shifted toward the previous stimulus), while negative values indicate a 471 repulsive bias (i.e., errors shifted away). 472 Previous studies have shown that serial dependence can exhibit both attractive and repulsive 473 components, with attraction typically occurring for small Δ values and repulsion for larger Δs 32,33. 474 Moreover, manipulations of prior response requirements have revealed increased repulsion at large 475 Δ following non -response (R0) trials see Experiment 3 in 18 . Because a single δoG cannot capture both 476 attractive and repulsive components simultaneously, we first visually inspected whether the data, 477 overall across conditions, displayed a biphasic pattern consistent with both effects. After confirming 478 this pattern, a “double δoG” model —comprising the weighted sum of two δoG functions with free 479 amplitude and width parameters—was fit to the error data 32. 480 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 15 To assess whether the shape and strength of serial dependence varied with trial history, we 481 implemented a condition-dependent model including interaction terms between Δ and the response 482 condition (R0-R1 vs. R1-R1). This allowed both the amplitude and width parameters of the two δoG 483 components to vary as a function of condition. 484 Model parameters were estimated using constrained nonlinear least squares optimization 485 (fmincon in MATLAB) applied to the pooled single-trial data across participants. Model comparison 486 was performed using the Bayesian Information Criterion (BIC), which penalizes model complexity. 487 Differences in BIC (ΔBIC) greater than 2 are considered positive evidence against the model with the 488 higher BIC 49. We specifically tested whether a model allowing different parameters across conditions 489 provided a better fit, indicating that the presence or absence of a response on the preceding trial 490 significantly altered the serial dependence pattern observed on the current trial. 491 492 Experiment 2 493 Participants 494 Experiment 2 was conducted at the University of Fribourg (Fribourg, Switzerland). Twenty -one 495 healthy participants (17 females, age range: 18-31 y) were recruited from the local student population. 496 All participants had normal or corrected-to-normal vision, verified prior to the experiment using the 497 Freiburg Acuity Test, for which a minimum value of 1 was required with both eyes open 47. Written 498 informed consent was obtained from all participants prior to participation. The experimental 499 procedures were approved by the regional ethics board (CER -VD, Protocol Nr. 2016 -00060) and 500 were conducted in accordance with the Declaration of Helsinki. 501 502 Apparatus 503 Participants were seated in a dark, electrically shielded room, at a distance of 80 cm from the display 504 monitor. EEG was recorded using a 128-channel BioSemi ActiveTwo system (BioSemi, Amsterdam, 505 The Netherlands). Stimuli were generated using MATLAB (MathWorks, Natick, MA) and the 506 Psychophysics Toolbox 50 and presented on a VIEWPixx/3D display system (1920 × 1080 pixels 507 resolution; 120 Hz refresh rate; VPixx Technologies, Canada). 508 509 Stimuli and procedure 510 An example trial sequence and the two main conditions are illustrated in Figure 2A. Each trial began 511 with a fixation point displayed at the center of the screen for 500 ms. This was followed by a Gabor 512 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 16 patch presented for 500 ms, which was then masked by a noise pattern lasting 200 ms. A blank screen 513 was shown for 1000 ms before the response stage. 514 In R1 trials, participants were shown a randomly oriented white bar and asked to rotate it to 515 match the perceived orientation of the preceding Gabor stimulus using the left and right arrow keys 516 on a keyboard. The final response was confirmed by pressing the spacebar. In R0 trials (50% of trials, 517 randomly assigned), a central gray oval appeared in place of the response tool, serving as an explicit 518 no-response cue. Participants were instructed to await the next trial without providing any response. 519 The duration of the no-response cue was initially fixed at 1.5 s, based on average adjustment 520 durations reported in similar tasks 1,18, but was subsequently adapted after the first 50 trials to reflect 521 the running average of the participant’s actual adjustment times. The inter-trial interval (blank screen) 522 was randomly jittered between 1.1 and 1.3 s in 20 ms increments. 523 Gabor stimuli were sinusoidal gratings (spatial frequency: 1 cycle/°; Michelson contrast: 0.25) 524 windowed by a Gaussian envelope with a sigma of 1°. Stimuli were presented centrally on a gray 525 background. The mask consisted of full -contrast Gaussian-filtered white noise. On each trial, the 526 orientation of the Gabor was selected at random from 0° to 170° in 10° increments, with the constraint 527 that the orientation difference relative to the preceding trial (Δ: previous minus current orientation) 528 did not exceed ±50°. This constraint resulted in 11 possible Δ values ranging from –50° to +50° in 529 10° steps. 530 531 Behavioral data analysis 532 The analysis of behavioral serial dependence followed the procedure and steps described for 533 Experiment 1. Across all steps, less than 6% of response trials were excluded from the analysis. 534 535 EEG recording and preprocessing 536 EEG data were recorded at a sampling rate of 2048 Hz. Signal quality was monitored online by 537 ensuring that offsets between active electrodes and the Common Mode Sense – Driven Right Leg 538 (CMS–DRL) feedback loop remained within ±20 mV. 539 Offline preprocessing was conducted in EEGLAB 2019.1 51. EEG data were first 540 downsampled to 200 Hz using pop_resample.m, with an anti -aliasing filter cutoff of 0.8 and a 541 transition bandwidth of 0.4 (normalized units in π rad/sample). Data were locally detrended using the 542 PREP pipeline plugin 52. Line noise at 50 Hz and its harmonics was removed using pop_cleanline.m. 543 Data were then epoched from -1000 to 2000 ms relative to stimulus onset. Bad channels and 544 epochs were identified by visual inspection and excluded prior to further preprocessing. 545 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 17 Remaining physiological artifacts were identified using independent component analysis 546 (ICA) with the Infomax algorithm as implemented in EEGLAB’s runica function. To mitigate rank 547 deficiency and reduce computational load, the data were first reduced in dimensionality using 548 principal component analysis (PCA) , retaining components that explained more than 98% of the 549 variance. 550 Artifact components were classified using the MARA (Multiple Artifact Rejection 551 Algorithm), and those explaining more than 90% of the total variance were inspected visually and 552 removed. Subsequently, bad channels were interpolated using the nearest -neighbor spline method, 553 and data were re-referenced to the average reference. On average, 49.76±33.19 (7.54%) epochs were 554 marked as outliers and excluded, 20.14±7.66 (15.74%) channels interpolated, and 14.43±6.72 555 (13.33%) ICA removed. 556 557 EEG analysis 558 Global Field Power (GFP) 559 To compare EEG evoked activity as a function of the previous trial condition (R0 –R1 vs. R1–R1), 560 we computed the Global Field Power (GFP) across electrodes. GFP is a reference -independent 561 measure of the overall strength of scalp EEG activity at each time point, defined as the standard 562 deviation of the voltage potentials across all electrodes 53. The analysis was restricted to the time 563 window from -100 to 500 ms after stimulus onset. 564 Here, GFP time courses were computed separately for each participant and condition, 565 following baseline normalization using the 100 ms pre -stimulus interval. This normalization step 566 controlled for potential differences in overall scalp amplitude due to residual motor -related activity 567 following R1–R1 trials. Only behaviorally valid trials were included in the analysis. In addition to 568 GFP, single-trial EEG data and channel-wise ERP maps were extracted for subsequent analyses. 569 To test for time-resolved differences in GFP between trial types, we employed non-parametric 570 cluster-based permutation testing with 1000 sign-flip permutations (flipping the GFP difference sign 571 within each participant). Cluster-level statistics—computed as the maximum summed t-values across 572 temporally contiguous points —were compared to a null distribution to determine significance , 573 providing intrinsic correction for multiple comparisons 54. Temporal clusters with p < .05 (two-tailed) 574 were considered statistically significant. 575 To identify the topographic distribution of GFP differences, we computed the average scalp 576 map of the R0–R1 vs. R1–R1 GFP difference, averaged over the time window corresponding to the 577 significant temporal cluster. 578 579 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 18 GFP-Behavior relationship 580 To assess whether trial-by-trial fluctuations in global field power (GFP) predicted behavioral biases, 581 we extracted single-trial GFP values within the significant time window of the effect (by averaging 582 over time and computing the standard deviation across electrodes for each trial). As a measure of 583 behavioral bias, we defined an individual-level global bias index as the product of signed error and Δ 584 (i.e., error × sign of orientation difference), excluding trials where Δ = 0°. This index —akin to the 585 error-folding procedures employed in previous studies 8—captures the direction and magnitude of 586 systematic deviations in response errors toward (positive bias) or away from (negative bias) the 587 previous-trial orientation. It thus provides a single -trial measure of serial dependence for each 588 participant and condition. 589 We modeled the relationship between GFP and the bias index using a linear regression 590 predicting trial-by-trial bias from GFP, including an additional predictor coding for the condition 591 (R0–R1 vs. R1 –R1, coded as a binary variable: 0 –1) and an intercept. This model allowed us to 592 estimate regression coefficients quantifying the relationship between GFP fluctuations and the 593 magnitude of serial dependence, while accounting for differences between experimental conditions. 594 595 Inverted Encoding Model (IEM) 596 To assess whether the quality of orientation representations in EEG activity was modulated by trial 597 history (e.g., previous response condition), we applied an Inverted Encoding Model IEM; 55,56 from -598 100 to 500ms. 599 The IEM assumes that stimulus orientation is encoded by a set of hypothetical neural channels 600 with idealized tuning functions. We used 18 orientation channels, each spaced 10° apart, covering 601 orientations from 0° to 170°. Each channel's tuning curve was modeled as a half-wave rectified cosine 602 raised to the 18th power. 603 For each trial, we generated a predicted pattern of channel responses based on the presented 604 stimulus orientation, forming a channel response matrix C of size [k × n] (with k = 18 channels and 605 n = number of trials). The observed EEG data were modeled as: 606 607 𝐸𝐸𝐺 = 𝑊 ⋅ 𝐶 + 𝑁 608 [2] 609 where EEG is the electrode × trial matrix [m x n] (with m = 128 electrodes), W is the weight matrix 610 [m x k] mapping channel responses to scalp sensors, and N is residual noise. Weights were estimated 611 using least-squares regression on a training set: 612 613 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 19 𝑊 = 𝐸𝐸𝐺!"#$% & ⋅ 𝐶!"#$% ' 614 [3] 615 where 𝐶!"#$% ' is the pseudoinverse of the training channel response matrix. Orientation -selective 616 channel responses for the test data were reconstructed as: 617 618 𝐶!()! = 𝑊& ⋅ 𝐸𝐸𝐺!()!. 619 [4] 620 The IEM was applied using a sliding window of 10 samples (corresponding to 50 ms), using 621 a 3-fold cross-validation procedure with 300 iterations per subject and condition. The decoder was 622 trained separately for each experimental condition (R0-R1 and R1-R1). 623 For each time point, the reconstructed channel tuning functions (CTFs) were circularly 624 realigned so that the channel tuned to the presented stimulus orientation was centered at 0°. This 625 yielded a [time x channel] matrix per subject. 626 Decoding performance was quantified as the slope of a linear fit between the reconstructed 627 CTF and an idealized basis function centered at 0°. This decoding score (CTF slope) reflects how 628 well the reconstructed signal matches the expected orientation tuning. Trials excluded from the 629 behavioral analysis (e.g., outliers in Behavioral data analysis ) were also excluded from the IEM 630 analysis. 631 632 IEM surrogate distribution and null hypothesis testing 633 To assess the statistical significance of decoding performance, we constructed a surrogate distribution 634 of CTF slopes under the null hypothesis of no orientation -specific information. Surrogates were 635 generated by shuffling the orientation labels during the circular realignment step, thereby disrupting 636 the systematic alignment between reconstructed CTFs and true orientations. This procedure was 637 repeated 300 times per subject, yielding a null distribution of 300 surrogate slopes per time point. 638 To test whether decoding performance (CTF slope) differed from zero and whether it varied 639 by condition (R0 -R1 vs. R1 -R1), we performed a regression analysis at each time point. For each 640 subject, CTF slope values across both conditions were concatenated into a single response vector. A 641 binary predictor variable coded the experimental condition (0 for R0-R1 and 1 for R1-R1). Regression 642 coefficients (intercept and condition effect) and their associated t-statistic were estimated using least-643 squares regression. 644 To evaluate the group -level significance of these effects, we applied the same regression 645 procedure to the surrogate slope distributions, producing a null distribution of t-statistics associated 646 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 20 with the regression coefficients (intercept and condition) per time point. To correct for multiple 647 comparisons across time points, we used a cluster -based permutation approach. For each surrogate 648 iteration, we identified time points where the empirical t -statistics for the intercept and condition 649 effect exceeded the two-tailed critical value (|t| > tα/2, with α = 0.05 and degrees of freedom df = 20). 650 We then computed the sum of supra -threshold t-values within each cluster to derive a cluster -level 651 statistic and retained the maximum absolute cluster sum from each iteration to form a null 652 distribution. Cluster-level statistics from the real data were calculated using the same procedure. P -653 values for observed clusters were obtained by comparing these statistics to the corresponding null 654 distributions for positive and negative clusters. Significant clusters in the intercept term identified 655 periods during which the current orientation was reliably decoded from EEG activity. A significant 656 condition effect indicated temporal windows where the quality of orientation representations differed 657 between the R0 -R1 and R1 -R1 conditions. Specifically, a negative condition coefficient reflected 658 reduced decoding fidelity in the R1-R1 condition relative to R0-R1. 659 660 Acknowledgments 661 The authors thank Gizay Ceylan for assistance with the setup and data collection of Experiment 1, 662 Laura Cohen for data collection, and Laura Cohen and Cemre Yilmaz for the initial preprocessing 663 and data handling of Experiment 2. We also thank Gijs Plomp for providing the resources to run 664 Experiment 2. 665 666 Funding 667 Data for Experiment 2 were collected while D. Pascucci was a postdoctoral researcher in the lab of 668 G. Plomp (University of Fribourg) with support from the Swiss National Science Foundation (grants 669 PZ00P3_131731 and PP00P1_157420). Data collection for Experiment 1 and the final analyses were 670 supported by the Swiss National Science Foundation (grants PZ00P1_179988, PZ00P1_179988/2, 671 and TMSGI1_218247). 672 673 Author contributions 674 Experiment design: D.P. 675 Data analysis: D.P., J.L 676 Writing manuscript: D.P., J.L 677 Illustrations: D.P. 678 Editing manuscript: D.P., J.L 679 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 21 680 Competing interests 681 The authors have no conflicts of interest to declare. 682 683 Data availability 684 Data and code will be made available before publication. 685 686

687 1. Fischer, J. & Whitney, D. Serial dependence in visual perception. Nat. Neurosci. 17, 738–743 688 (2014). 689 2. Manassi, M., Murai, Y. & Whitney, D. Serial dependence in visual perception: A meta -analysis 690 and review. J. Vis. 23, 18 (2023). 691 3. Pascucci, D. et al. Serial dependence in visual perception: A review. J. Vis. 23, 9 (2023). 692 4. Ceylan, G., Herzog, M. H. & Pascucci, D. Serial dependence does not originate from low -level 693 visual processing. Cognition 212, 104709 (2021). 694 5. Cicchini, G. M., Mikellidou, K. & Burr, D. The functional role of serial dependence. Proc. R. 695 Soc. B 285, 20181722 (2018). 696 6. Gallagher, G. K. & Benton, C. P. Stimulus uncertainty predicts serial dependence in orientation 697 judgements. J. Vis. 22, 6–6 (2022). 698 7. Tanrikulu, Ö. D., Pascucci, D. & Kristjánsson, Á. Stronger serial dependence in the depth plane 699 than the fronto-parallel plane between realistic objects: Evidence from virtual reality. J. Vis. 23, 700 20 (2023). 701 8. Barbosa, J. & Compte, A. Build-up of serial dependence in color working memory. Sci. Rep. 10, 702 10959 (2020). 703 9. Fischer, C. et al. Context information supports serial dependence of multiple visual objects across 704 memory episodes. Nat. Commun. 11, 1–11 (2020). 705 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 22 10. Kim, S., Burr, D. & Alais, D. Attraction to the recent past in aesthetic judgments: A positive serial 706 dependence for rating artwork. J. Vis. 19, 19–19 (2019). 707 11. Liberman, A. & Whitney, D. The serial dependence of perceived emotional expression. J. Vis. 708 15, 929–929 (2015). 709 12. Kiyonaga, A., Scimeca, J. M., Bliss, D. P. & Whitney, D. Serial Dependence across Perception, 710 Attention, and Memory. Trends Cogn. Sci. (2017). 711 13. Lõoke, M., Guérineau, C., Broseghini, A., Mongillo, P. & Marinelli, L. Visual continuum in non-712 human animals: serial dependence revealed in dogs. Proc. R. Soc. B Biol. Sci. 291, 20240051 713 (2024). 714 14. Papadimitriou, C., Ferdoash, A. & Snyder, L. H. Ghosts in the machine: memory interference 715 from the previous trial. J. Neurophysiol. 113, 567–577 (2014). 716 15. Ceylan, G. & Pascucci, D. Attractive and repulsive serial dependence: The role of task relevance, 717 the passage of time, and the number of stimuli. J. Vis. 23, 8 (2023). 718 16. Guan, S. & Goettker, A. Individual differences reveal similarities in serial dependence effects 719 across perceptual tasks, but not to oculomotor tasks. J. Vis. 24, 2–2 (2024). 720 17. Kim, S. & Alais, D. Individual differences in serial dependence manifest when sensory 721 uncertainty is high. Vision Res. 188, 274–282 (2021). 722 18. Pascucci, D. et al. Laws of concatenated perception: Vision goes for novelty, decisions for 723 perseverance. PLOS Biol. 17, e3000144 (2019). 724 19. Alais, D., Leung, J. & Van der Burg, E. Linear summation of repulsive and attractive serial 725 dependencies: orientation and motion dependencies sum in motion perception. J. Neurosci. 37, 726 4381–4390 (2017). 727 20. Blondé, P., Kristjánsson, Á. & Pascucci, D. Tuning perception and decisions to temporal context. 728 iScience 108008 (2023) doi:10.1016/j.isci.2023.108008. 729 21. Fornaciai, M. & Park, J. Spontaneous repulsive adaptation in the absence of attractive serial 730 dependence. J. Vis. 19, 21–21 (2019). 731 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 23 22. Luo, M., Zhang, H., Fang, F. & Luo, H. Reactivation of previous decisions repulsively biases 732 sensory encoding but attractively biases decision-making. PLOS Biol. 23, e3003150 (2025). 733 23. Moon, J. & Kwon, O. -S. Attractive and repulsive effects of sensory history concurrently shape 734 visual perception. BMC Biol. 20, 247 (2022). 735 24. Pascucci, D. & Plomp, G. Serial dependence and representational momentum in single -trial 736 perceptual decisions. Sci. Rep. 11, 9910 (2021). 737 25. Fritsche, M., Spaak, E. & de Lange, F. P. A Bayesian and efficient observer model explains 738 concurrent attractive and repulsive history biases in visual perception. eLife 9, e55389 (2020). 739 26. Sheehan, T. C. & Serences, J. T. Attractive serial dependence overcomes repulsive neuronal 740 adaptation. PLoS Biol. 20, e3001711 (2022). 741 27. Manassi, M., Liberman, A., Kosovicheva, A., Zhang, K. & Whitney, D. Serial dependence in 742 position occurs at the time of perception. Psychon. Bull. Rev. 1–9 (2018). 743 28. Pascucci, D. et al. Intact Serial Dependence in Schizophrenia: Evidence from an Orientation 744 Adjustment Task. Schizophr. Bull. sbae106 (2024). 745 29. Cicchini, G. M., Mikellidou, K. & Burr, D. Serial dependencies act directly on perception. J. Vis. 746 17, 6–6 (2017). 747 30. Murai, Y. & Whitney, D. Serial dependence revealed in history-dependent perceptual templates. 748 Curr. Biol. (2021). 749 31. Pascucci, D., Mastropasqua, T. & Turatto, M. Permeability of priming of pop out to expectations. 750 J. Vis. 12, 21–21 (2012). 751 32. Houborg, C., Pascucci, D., Tanrıkulu, Ö. D. & Kristjánsson, Á. The effects of visual distractors 752 on serial dependence. J. Vis. 23, 1 (2023). 753 33. Fritsche, M. & de Lange, F. P. The role of feature-based attention in visual serial dependence. J. 754 Vis. 19, 21 (2019). 755 34. Croson, R. & Sundali, J. The gambler’s fallacy and the hot hand: Empirical data from casinos. J. 756 Risk Uncertain. 30, 195–209 (2005). 757 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 24 35. Fischer, C., Kaiser, J. & Bledowski, C. A direct neural signature of serial dependence in working 758 memory. eLife 13, (2024). 759 36. Fritsche, M., Mostert, P. & de Lange, F. P. Opposite effects of recent history on perception and 760 decision. Curr. Biol. 27, 590–595 (2017). 761 37. Shan, J. & Postle, B. R. The Influence of Active Removal from Working Memory on Serial 762 Dependence. 5, 31 (2022). 763 38. Markov, Y. A., Tiurina, N. A. & Pascucci, D. Serial dependence: A matter of memory load. 764 Heliyon 10, e33977 (2024). 765 39. Ozkirli, A. & Pascucci, D. It’s not the spoon that bends: Internal states of the observer determine 766 serial dependence. 2023.10.19.563128 Preprint at https://doi.org/10.1101/2023.10.19.563128 767 (2024). 768 40. Weilnhammer, V., Murai, Y. & Whitney, D. Dynamic predictive templates in perception. Curr. 769 Biol. 34, 4301–4306 (2024). 770 41. Reynolds, J. H. & Chelazzi, L. ATTENTIONAL MODULATION OF VISUAL PROCESSING. 771 Annu. Rev. Neurosci. 27, 611–647 (2004). 772 42. Sadil, P., Cowell, R. A. & Huber, D. E. The push–pull of serial dependence effects: Attraction to 773 the prior response and repulsion from the prior stimulus. Psychon. Bull. Rev. 1–15 (2023). 774 43. Collins, T. The perceptual continuity field is retinotopic. Sci. Rep. 9, 1–6 (2019). 775 44. Bae, G. -Y. & Luck, S. J. Reactivation of previous experiences in a working memory task. 776 Psychol. Sci. 30, 587–595 (2019). 777 45. Barbosa, J. et al. Interplay between persistent activity and activity -silent dynamics in the 778 prefrontal cortex underlies serial biases in working memory. Nat. Neurosci. 23, 1016 –1024 779 (2020). 780 46. Bansal, S. et al. Qualitatively Different Delay-Dependent Working Memory Distortions in People 781 With Schizophrenia and Healthy Control Participants. Biol. Psychiatry Cogn. Neurosci. 782 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint Response expectations and serial dependence 25 Neuroimaging https://doi.org/10.1016/j.bpsc.2023.07.004 (2023) 783 doi:10.1016/j.bpsc.2023.07.004. 784 47. Bach, M. The Freiburg Visual Acuity Test-automatic measurement of visual acuity. Optom. Vis. 785 Sci. 73, 49–53 (1996). 786 48. Houborg, C., Kristjánsson, Á., Tanrıkulu, Ö. D. & Pascucci, D. The role of secondary features in 787 serial dependence. J. Vis. 23, 21 (2023). 788 49. Raftery, A. E. Bayesian model selection in social research. Sociol. Methodol. 111–163 (1995). 789 50. Brainard, D. H. The psychophysics toolbox. Spat. Vis. 10, 433–436 (1997). 790 51. Delorme, A. & Makeig, S. EEGLAB: an open source toolbox for analysis of single -trial EEG 791 dynamics including independent component analysis. J. Neurosci. Methods 134, 9–21 (2004). 792 52. Bigdely-Shamlo, N., Mullen, T., Kothe, C., Su, K. -M. & Robbins, K. A. The PREP pipeline: 793 standardized preprocessing for large-scale EEG analysis. Front. Neuroinformatics 9, 16 (2015). 794 53. Lehmann, D. & Skrandies, W. Reference -free identification of components of checkerboard -795 evoked multichannel potential fields. Electroencephalogr. Clin. Neurophysiol. 48, 609 –621 796 (1980). 797 54. Maris, E. & Oostenveld, R. Nonparametric statistical testing of EEG-and MEG-data. J. Neurosci. 798

164, 177–190 (2007). 799 55. Brouwer, G. J. & Heeger, D. J. Decoding and reconstructing color from responses in human visual 800 cortex. J. Neurosci. 29, 13992–14003 (2009). 801 56. Sprague, T. C. et al. Inverted Encoding Models Assay Population -Level Stimulus 802 Representations, Not Single-Unit Neural Tuning. eNeuro 5, ENEURO.0098-18.2018 (2018). 803 804 805 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted October 22, 2025. ; https://doi.org/10.1101/2025.10.22.683912doi: bioRxiv preprint