Does Presentation Duration Modulate the Effect of Task on Perceptual Organization?

Abstract

Visual symmetry activates the extrastriate cortex. This symmetry activation generates an Event Related Potential (ERP) named the Sustained Posterior Negativity (SPN). The SPN is larger when participants engage in regularity discrimination tasks than luminance discrimination tasks. Two recent studies suggest that the type of task matters more when the stimulus is presented briefly. We tested this claim with a new sample of 52 participants in a within-subjects design. As predicted, the SPNs were larger in Regularity tasks than Luminance tasks. However, contrary to predictions, the effect of the task was very similar in long and short presentation duration blocks. It remains unclear what factors modulate the task effect. However, presentation duration is not as important as previous results suggest.

1. Introduction

Reflectional symmetry (henceforth, symmetry) is detected rapidly and efficiently by human observers [1]. Many animals use symmetry to guide adaptive behavior. For instance, mammals, birds, fish and insects have all been shown to use symmetry in mate and/or food selection [2,3], and humans use symmetry to guide judgments of facial attractiveness [4]. Furthermore, symmetrical regions are usually perceived as figures, and asymmetrical regions are usually perceived as ground [5]. Symmetry is an important cue in perceptual organization. This could be why human infants seem innately sensitive to reflectional symmetry [6].

Symmetry activates a network of brain regions in the extrastriate cortex, with the strongest response in the shape-sensitive lateral occipital complex [7,8,9,10,11,12,13]. Symmetrical and random stimuli both produce Event Related Potentials (ERPs) at posterior electrodes. Amplitude is lower for symmetrical stimuli after the N1 component (Figure 1). This ‘Sustained Posterior Negativity’ (SPN) was first seen in steady-state visual-evoked potential studies [14] but is most often reported in ERP studies (starting with [15]).

Figure 1 from [16] is reproduced again as Figure 1 here. Figure 1A shows that SPN amplitude scales with the proportion of symmetrical dots in mixed symmetry and noise displays [17]. The more symmetry there is in the image, the larger the SPN. In other words, the SPN falls further below zero when there is more symmetry in the image. Figure 1B shows 227 grand average SPNs from a public database called the Complete Liverpool SPN catalog [18]. There are two effects in this scatterplot. First, SPN amplitude increases with ‘W’, a theoretical measure of symmetry salience [19,20]. Second, the SPN is larger when regularity is task-relevant (blue dots) than when it is not task-relevant (pink dots). The average magnitude of this task effect is around 0.4 microvolts (compare intercept of regression lines in Figure 1B). However, the size of the task effect is variable and may be systematically influenced by factors that have not been studied.

Two new findings suggest that stimulus presentation duration may explain some variance in the magnitude of the task effect (Figure 2). Figure 2A shows results from [21], in which duration was 1000 ms. Here, SPNs were only slightly larger in the Regularity task than the Luminance task. Figure 2B shows results from [16], in which duration was just 300 ms. Here, SPNs were much larger in the Regularity task than the Luminance task. Stimulus presentation duration was the main systematic difference between these studies, which were matched in terms of other stimulus properties and sample size.

To compare these results, we measured grand average SPN amplitudes during the 250–400 ms window from electrodes PO7, O1, O2 and PO8. In [21], the SPN amplitude was −2.57 microvolts in the Regularity task and −2.19 microvolts in the Luminance task (a difference of 0.39 microvolts). This is in line with the previous experiments, where presentation duration was also at least 1000 ms (Figure 1B). In [16], the SPN amplitude was −2.42 microvolts in the Regularity task and just −0.84 microvolts in the Luminance task (a difference of 1.59 microvolts). This task effect is unusually large, and the duration was unusually short.

These data sets from [16,21] have not been analyzed together and published before. We thus ran a new univariate ANOVA [Task (Luminance, Regularity) X Duration (1000 ms, 300 ms)] that confirmed the interaction (F (1,156) = 6.93, p = 0.009, ηp² = 0.04). In [21], where the duration was 1000 ms, there was no significant effect of Task (t (78) = 1.00, p = 0.319, d_s = 0.23). Conversely, in [16], where the duration was 300 ms, there was a large effect of Task (t (78) = 6.42, p < 0.001, d_s = 1.44).

These results suggest that Luminance and Regularity discrimination compete for attentional resources, and this competition is stronger when time is scarce. This is intuitively plausible: when there is plenty of time to complete a task, there is also plenty of spare time to process task-irrelevant things. It is also broadly consistent with the cognitive literature. It is well known that moving selective attention takes time [22,23]. In a Luminance task where the stimulus is presented for 1000 ms, there would be sufficient time for many spontaneous shifts of attention between the task-relevant luminance and task-irrelevant regularity. These luminance–regularity attentional shifts might happen at different points in different trials. ERPs averaged over trials would thus include many intervals where regularity was attended. Conversely, in a Luminance task where the stimulus is presented for 300 ms, participants may prioritize the relevant luminance information for the brief period it is available and never spontaneously attend to the irrelevant regularity information.

Current Study

We tested the hypothesis that stimulus presentation duration influences the magnitude of the task effect. All participants completed four experimental blocks [Task (Regularity, Luminance) X Duration (300 ms, 1000 ms)]. These blocks are referred to as RegShort, RegLong, LumShort and LumLong. As shown in Figure 3, all blocks involved the same stimulus types [Regularity (symmetry, asymmetry) X Luminance (light, dark)]. Stimuli were the same as those in [16,21]. While there would be advantages to including more than two levels of duration to uncover potential non-linear effects, we began by examining the 300 vs. 1000 ms difference apparent in Figure 2.

We predicted an SPN in all four blocks. We also predicted that mean SPNs would be larger in Regularity tasks than in Luminance tasks. Most importantly, we predicted that this task effect would be larger in the short duration blocks than the long duration blocks. The hypothesis and analysis plan were pre-registered on aspredicted.org.

2. Methods

2.1. Participants

A total of 52 participants (mean age = 21.1, age range = 18 to 61, 10 males, 4 left-handed) were recruited. All had normal or corrected vision. All subjects gave their informed consent for inclusion before participating in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the University of Liverpool Ethics Committee of (Approval Code: Ref 13550, approved 31 January 2024).

2.2. Power Analysis

This project had 52 participants in a within-subject design. This sample size was pre-registered and justified in two ways. First, we needed the study to detect pairwise differences between SPN means with two-tailed paired samples t tests. One theoretical commentary on statistical power [24] recommends a default sample of 52, assuming that a minimally interesting Cohen’s d_z is 0.4 and accepting the conventions of alpha = 0.05 and power = 0.8. Cohen’s d_z of 0.4 is also an approximate effect size for a 0.42 microvolt SPN modulation, and task effects can be expected to lie in this range [17].

Our core prediction was that we would find a Task X Duration interaction with repeated measures ANOVA. We confirmed that N = 52 was adequate for this with a power simulation. This was based on the means and standard deviations associated with the SPNs in Figure 1 [−2.45 (2.08), −2.42 (1.14), −2.04 (1.66) and −0.84 (1.07)]. We conservatively assumed that the correlation between SPN amplitudes in each block would be 0.4 based on other recent SPN data from a within-subjects blocked design (Project 44 from SPN catalog). The crucial Task X Duration interaction was significant in 92.85% of 2000 simulated experiments with 52 participants (p < 0.05). Statistical power is thus approximately 0.93, according to this analysis. This power simulation can be recreated on https://shiny.ieis.tue.nl/anova_power/ (accessed 20 April 2024).

2.3. Apparatus

The EEG experiment used the BioSemi Active-Two system (Amsterdam, The Netherlands). EEG data were recorded continuously from 64 scalp electrodes arranged according to the extended international 10–20 system. Bipolar VEOG and HEOG external channels were used to monitor for ocular artifacts but not included in ERP analysis. The participant’s head position was stabilized with a chin rest positioned 57 cm from a 51 × 29 cm (1920 × 1080 pixel) HP E233 LED backlit monitor with a 60 Hz refresh rate. The experiment was conducted in a darkened and electrically shielded room. Experimental programming was completed in Python using open-source PsychoPy3 software [25].

2.4. Stimuli

The stimulus generation algorithm was the same as that used in [16]. Every trial employed arrangements of 40 Gaussian-filtered dot elements in a square frame in the center of the screen (Figure 3). As explained in [16], an algorithm was used to arrange the dots in a different way on every trial so that participants never saw the same stimuli twice (unless by remote chance). The square was 6.6 × 6.6 cm (approximately 6.6 × 6.6 degrees of visual angle) and had an implicit grid of 12 × 12 cells. Within this, a central 10 × 10 grid contained small dots, with each approximately 0.25 degrees of visual angle in diameter. Each quadrant contained 10 dots, occupying 40% of the available 25 cells. The distribution of dots in the first quadrant was chosen randomly. Within each occupied cell, dot location was jittered on the X and Y dimensions by approximately 0.1 degrees of visual angle, so it was rarely located in the center of the cell. Intra-cell jittering prevented the appearance of multi-element straight lines spanning several cells. Without this, even asymmetrical patterns had rows and/or columns of aligned elements. For symmetrical patterns, the first quadrant was reflected three times, giving 2-fold vertical and horizontal reflection with a W load of 0.75. In asymmetrical patterns, all four quadrants were independently generated. Symmetrical and asymmetrical stimuli were indistinguishable based on information in a single quadrant. At center, dots were either dark gray (4 cd/m²) or black (approximately 0.1 cd/m²). The background was mid-gray (approximately 37 cd/m²). Previous research shows that these luminance and regularity differences are easily perceived [16].

2.5. Design

There were four basic types of stimuli: [Regularity (symmetry, asymmetry) X Luminance (dark, light)]. The stimuli were shown in a randomized order in each of the four blocks [Task (Regularity, Luminance) X Duration (300 ms, 1000 ms)]. These blocks are referred to as RegShort, RegLong, LumShort and LumLong. In each block, there were 192 trials (48 in each of the four stimulus conditions). There were thus 768 trials in total.

2.6. Procedure

Each trial began with 1500 ms baseline. Stimulus presentation was either 300 ms (short duration) or 1000 ms (long duration). Following stimulus offset, participants classified patterns as ‘Symmetry’ or ‘Random’ (Regularity task) or as ‘Gray’ or ‘Black’ (Luminance task). Responses were entered using the left (A) and right (L) keys. Response mapping was randomized to prevent lateralized motor preparation during the presentation interval. The word ‘wrong’ appeared in red for 1500 ms if participants entered an incorrect response.

A practice block of 16 trials preceded each block. The same distribution of trial types was used in the practice blocks.

Half of the participants completed the Regularity task first; half completed the Luminance task first. Half of the participants completed the short duration condition first; half completed the long duration condition first. In total, we planned for 24 possible block orders. We aimed to use the remaining 18 orders twice and the remaining 4 orders thrice. However, this counterbalancing was incorrectly applied to one replacement participant.

2.7. EEG Processing

The processing pipeline was the same as that in [16]. EEG data were analyzed using eeglab 2022.1 functions in Matlab 2023a. All raw data and analysis scripts are available in the Project 45 folders on the SPN catalog.

EEG datasets were first re-referenced to the scalp average, low-pass filtered at 25 Hz, and downsampled to 256 Hz. Continuous data were then segmented into −500 to + 500 ms epochs. Noisy channels were identified and zeroed during artifact rejection using Independent Components Analysis (ICA). Components capturing large artifacts were identified and removed using the Adjust procedure. After this, noisy channels were replaced using spherical interpolation. The number of ICA components removed was similar across the four blocks (LumLong, M = 6.48, min = 1, max = 16; LumShort, M = 6.29, min = 1, max = 34; RegLong, M = 5.60, min = 1, max = 23, RegShort, M = 6.08, min = 1, max = 19). Trials where amplitude exceeded +/− 100 microvolts at any scalp electrode were excluded. Trial inclusion rate was similar across all four blocks (LumLong, M = 96%, min = 63%, max = 100%; LumShort, M = 98%, min = 86%, max = 100%; RegLong, M = 96%, min = 52%, max = 100%; RegShort, M = 98%, min = 88%, max = 100%). Data were then averaged over trials in each condition and for each participant.

2.8. Analysis

The SPN was defined as the symmetry–asymmetry amplitude difference averaged across the posterior electrode cluster [PO7, O1, O2 and PO8] and across the 250–400 ms time window. This spatiotemporal cluster was pre-registered. SPN amplitudes were then analyzed in a 2 × 2 repeated measures ANOVA [Task (Regularity, Luminance) X Duration (300, 1000 ms)]. The presence of the four SPNs was confirmed with four two-tailed one-sample t-tests. None of the four SPNs significantly violated assumption of normality according to the Shapiro –Wilk test (p > 0.264).

3. Results

3.1. Behavioral Results

In the Luminance tasks, participants responded correctly on most trials (mean = 97.72%, worst participant = 90.10%, best participant = 100%). This was also true of the Regularity tasks (mean 97.89%, worst participant = 92.19%, best participant = 100%).

3.2. EEG Results

The ERP waves are shown in Figure 4, topoplots are shown in Figure 5, and the mean SPNs are shown in Figure 6. As predicted, there was an SPN in all four blocks (LumLong M = −1.35, t (51) = −8.87, p < 0.001, d_z = −1.23; LumShort M = −1.47, t (51) = −7.86, p < 0.001, d_z = 1.09; RegLong M −2.15, t (51) = 11.32, p < 0.001, d_z = −1.57; RegShort = −2.24, t (51) = −12.36, p < 0.001, d_z = −1.71). These t tests were not corrected for multiple comparisons but would all be significant if we had used a more stringent alpha level of 0.05/4 = 0.0125.

As also predicted, the SPN was larger in the Regularity tasks than Luminance tasks. However, contrary to our predictions, the effect of Task was very similar in the long and short duration blocks.

These impressions were supported by a 2 × 2 repeated measures ANOVA. This revealed a significant main effect of Task (F (1,51) = 21.82, p < 0.001, ηp² = 0.30). There was, however, no significant main effect of Duration (F (1,51) = 0.40, p = 0.530, ηp² = 0.01). There was also no significant Task X Duration interaction (F (1,51) = 0.02, p = 0.884, ηp² = 0.00).

Additional analyses found no significant effects involving the between-subject’s factor Block order (largest effect (F (23,28) = 1.020, p = 0.475, ηp² = 0.456). Furthermore, we ran an exploratory univariate ANOVA on SPNs from the first block only. This also found no effect of Duration (F (1,48) < 0.001, p = 0.976, ηp² = 0.00) and no Task X Duration interaction (F (1,48) < 0.001, p = 0.987, ηp² = 0.00). These analyses suggest that there was little role for learning effects.

3.3. Bayesian Analysis

The standard frequentist ANOVA can provide support for the presence of effects, but it cannot support the absence of effects. We therefore used a complementary Bayesian repeated measures ANOVA. BF10 > 3 provides moderate evidence for the presence of an effect, and BF10 < 1/3 provides moderate evidence for the absence of an effect. This analysis suggested the presence of a main effect of Task (BF10 = 892.496), the absence of a main effect of Duration (BF10 = 0.246), and the absence of a Task X Duration interaction (BF10 = 0.212). This Bayesian ANOVA used uninformed priors, however, it would be reasonable to take a step back and first consider previous datasets shown in Figure 2 [16,21]. A univariate Bayesian ANOVA on these earlier datasets found that the BF10 associated with Task X Duration was 4.621. We can then use 4.621 as the prior when analyzing the Task X Duration interaction in the current study. Specifically, the prior odds in favor of H1 (4.621:1) can be multiplied by the new BF10 (0.212) to give the new posterior odds in favor of H1 (0.984:1). This two-step Bayesian analysis implies that we should currently be agnostic about the existence of a Task X Duration interaction rather than concluding that there is no interaction. However, the switch from between- to within-subjects design is a complication here.

4. Discussion

The SPN is often larger in Regularity tasks than Luminance tasks. A comparison of two previous studies suggested that task effects are larger when stimulus presentation duration is shorter (Figure 2). However, contrary to expectations, our new experiment found that the task effect was very similar in long- and short-duration blocks.

Why was the expected Task X Duration interaction absent in this new study? The new study was matched with the previous studies in terms of stimuli and task difficulty. One difference was that we presented task and duration as blocks in a within-subjects design. However, there were no significant block order effects and no Task X Duration interaction when the first block was analyzed alone. This suggests that the switch to a within-subjects design was not critical.

Another consideration is that the results from [21] were ‘frontoparallel trials’ from an experiment with additional ‘perspective trials’. The frontoparallel trials used stimuli like those in the current work, while the perspective trials used the same stimuli from alternative viewpoints, distorting the retinal image. However, it is unclear why the interleaved perspective trials should reduce the task effect.

In the current Luminance task, the light and dark elements were readily distinguishable (mean correct response = 97.72%). Duration may become more important if the Luminance tasks were more difficult. However, [16] found that the SPN was larger in difficult Luminance tasks, so this is not a confident prediction.

Despite null results regarding duration, other claims regarding the SPN were supported. The presence of the SPN across all four blocks again shows that the SPN is produced automatically and during secondary tasks [16,17,26]. The expected task difference was present, although larger than in previous studies. In the current study, the average magnitude of the task effect was 0.78 microvolts. This is nearly double the SPN catalog average of 0.4 microvolts (Figure 1), although less than the 1.59 microvolt effect found in [16].

As always, claims about symmetry perception require the following caution: what is true of reflectional symmetry may not be true of rotational symmetry, translational symmetry, or Glass patterns. For instance, Glass patterns produce the same SPN as reflection when they are attended [27], but these SPNs might be more vulnerable to experimental manipulations of the task.

As mentioned above, reflectional symmetry is used in mate and food-selection. Furthermore, reflectional symmetry is a salient property of faces and bodies. Consequently, reflectional symmetry in the visual image signals that another animal has you in its awareness [28]. This is a special situation that may require special action. Other visual regularities do not have the same ecological significance and may not need to be detected preattentively.

5. Conclusions

We began by comparing previous SPN results [16,21]. These studies suggested that the tasks matter more when the duration is short. However, the modulating effect of duration was not present in this new study.

We finish with a broader point. Humanity requires cumulative science, where knowledge increases over time [29]. This project has some modest meta-scientific lessons. It would probably have been possible to publish the combined results from [16,21] without running the new experiment. This paper would simply have presented a new analysis of the two previously reported datasets. Figure 2 would have been the centerpiece of the Section 3, and then the discussion would have told an intuitively plausible story about spare time for irrelevant visual processing. However, that paper would have been premature. While it is useful to explore apparent patterns in existing datasets and show that they are significant, new experiments are required to ensure that these patterns are reliable.