Wearable EEG for Detecting Beta-Band Differences Between Paper and Screen Reading: A Proof of Concept

Abstract

Portable EEG systems may facilitate electrophysiological research in naturalistic settings, but their ability to detect subtle condition-related differences during everyday cognitive tasks remains underexplored. This study is a proof of concept which examines whether a seven-channel Bluetooth EEG system could detect spectral differences between paper- and screen-based reading in adults. Fifteen participants, aged 28–43 years, completed both reading conditions in a counterbalanced within-subject design. Reading time, comprehension accuracy, and mean spectral power in the alpha and beta bands were compared between media. Our initial hypothesis was that power in the beta waves would be reduced in screen reading because its association with attentional processes is established in the current literature. No statistically significant differences were detected in reading time or comprehension accuracy. Alpha-band power also did not differ between conditions. In contrast, and in line with our hypothesis, beta-band power was statistically lower during screen-based reading than during paper-based reading. These findings suggest that the reading medium could be associated with differences in scalp-recorded beta activity even when behavioral differences are small. Although limited by the sample size and a lack of comparison with a high-density laboratory EEG, the results offer a proof of concept that a wearable low-density EEG system could be a useful tool for investigating spectral differences during naturalistic reading tasks.

1. Introduction

Digital devices have become important media for reading, learning, and information consumption [1]. However, evidence regarding the cognitive and neurophysiological correlates of screen-based reading remains mixed, and has focused predominantly on children and adolescents [2,3]. Adults who acquired literacy before the widespread adoption of mobile digital devices represent a relevant population for examining whether reading medium is associated with differences in cognitive processing, as these adults developed their neural reading networks primarily scaffolded by traditional print-based literacy practices [4,5] but now are forced to adapt to the fragmented and rapid processing demands of digital interfaces [6,7].

Advances in portable neurotechnology have made it possible to record electrophysiological activity during tasks that closely resemble everyday behavior. Mobile EEG is particularly promising for this purpose because it can reduce preparation demands and dependence on specialized laboratory infrastructure. However, the use of sparse-channel mobile EEG to detect condition-related spectral differences during naturalistic reading remains insufficiently investigated.

Previous behavioral studies have reported differences between paper- and screen-based reading [8,9], although these effects vary according to text characteristics, task demands, time pressure, and reader experience. Neurophysiological measures may complement behavioral outcomes by identifying differences in cognitive processing that are not necessarily reflected in reading time or comprehension performance.

Mangen et al. [10] conducted one of the foundational studies supporting the screen inferiority hypothesis in reading comprehension, proposing that the materiality of paper plays a critical role in comprehension. The authors compared reading on paper with reading on a computer screen with scrolling and used a randomized experimental design; they found that students achieved better reading comprehension when reading on paper than on a screen. To account for these findings, the authors proposed that reading is an embodied activity in which the physical properties of the medium contribute to cognitive processing. Specifically, they argued that paper provides richer sensorimotor and spatial cues, including physically turning pages, tactile feedback, and a tangible sense of reading progress through the thickness of the remaining pages. Although these findings are consistent with the hypothesis that the materiality of paper facilitates reading comprehension, the proposed mechanisms remain largely theoretical and require further empirical investigation.

To move beyond behavioral metrics, which often rely on retrospective self-reports or post hoc comprehension tests [11], it is essential to employ neurophysiological tools that capture the dynamics of cognitive processing. EEG is particularly well suited for this purpose due to its high temporal resolution, which enables precise tracking of neural fluctuations during the rapid visual processing required for reading [12]. A recent study investigated EEG spectral differences in children [13]. In fact, spectral power in the alpha and beta frequency bands has been associated with many functions such as attention, cognitive engagement, sensorimotor processing, and task-related state changes [14,15]. However, these frequency bands are not specific markers of a single cognitive process, and their interpretation depends on the task and the recording context. Our initial hypothesis was that spectral power in the beta band would be diminished in screen reading, compared to paper reading, because its association with attentional processes is established in the current literature.

This proof-of-concept study had two objectives. First, we examined whether a seven-channel mobile and wearable EEG system could be used to study condition-related differences in scalp-recorded spectral power during a naturalistic reading task. Second, we compared behavioral performance and alpha- and beta-band power between paper reading and screen reading in adults.

2. Materials and Methods

2.1. Participants

A total of 20 volunteers (10 males), aged 28–43 years (mean = 32.5 and s.d. = 3.8), participated in this study. After preprocessing steps for EEG signals quality control, 5 subjects were excluded due to large numbers of epoch rejections due to head motion and muscular artifacts. Therefore, the final sample size was 15 participants. This age range was selected to include adults who acquired literacy before the widespread adoption of mobile digital devices but currently use digital media in everyday life. Prior to data collection, all subjects provided written informed consent. The experimental protocol was approved by the Ethics Committee of Universidade Federal do ABC and conducted in accordance with federal ethical and safety guidelines.

Participants were included based on the following criteria: native Portuguese speakers; recruited primarily from an academic setting; and possessing normal or corrected-to-normal visual acuity. Exclusion criteria consisted of histories of neurological, psychiatric, or vascular disorders that could compromise cognitive function or EEG data quality.

2.2. Instrumentation

We recorded the electrophysiological data using an X.on EEG system (X-trodes, Herzliya, Israel). Signals were acquired through seven passive electrodes with a saline-soaked sponge base, positioned at F3, F4, C3, Cz, C4, P3, and P4 according to the International 10–20 System, with a reference electrode placed on the earlobe (see Figure 1). The device features a flexible montage suitable for adult head circumferences ranging from 52 to 60 cm. The signal was amplified and digitized with a sampling rate of 125 Hz and 24-bit A/D conversion resolution. Data transmission was performed via Bluetooth Low Energy (BLE) to a portable recording unit (Samsung Galaxy Tab S6, Samsung, Suwon-si, Republic of Korea).

The mobile EEG system was selected because its sampling rate was adequate for the prespecified 1–30 Hz spectral analysis and because its portability and sparse montage were compatible with the naturalistic aims of the study. The seven-channel configuration was intended to support within-subject comparisons of scalp-recorded oscillatory power and was not intended for source localization or fine-grained anatomical inference.

2.3. Experimental Procedure

Texts were printed on standard A4 paper. Digital reading was performed on a 15.6-inch Samsung 300E5K laptop with a screen resolution of 1366 × 768 pixels. Screen brightness was standardized at 60% throughout the experiment. In the paper media, the pages were stacked over the notebook screen (turned off, to preserve the same perspective angle for both media), and page turns were performed by physically flipping the pages. On the laptop, digital page turns were performed by pressing the left or right arrow keys on the keyboard. It is important to mention that the page spatial layout was identical for both media types. The texts were organized into pages which maintained the same size on screen and paper, with the same font and font size (Times New Roman 12).

After EEG setup and calibration, participants underwent a 5 min rest period to acclimate to the environment. The task of interest was reading texts in Brazilian Portuguese. Two texts were used, each with an average reading time of approximately seven minutes. The texts were entitled “A bolsa aljofarada de uma senhora” by Tennessee Williams (text A) and “O gato” by Juan Onetti (text B). It is important to mention that the experimental protocol used a balanced intra-subject design to avoid bias. The reading media (paper or screen) and text order (A or B first) were balanced across the participants.

After reading the first text, there was a 5 min rest period, after which the second block began, and the participant read the other text. After reading each text, the participants answered a comprehension/recall test, consisting of 10 multiple-choice questions. The reading time for each text was recorded using a standard digital timer.

2.4. Signal Preprocessing and Spectral Analysis

EEG data processing was performed using Python 3 and the MNE-Python toolbox (v1.9.0). Raw data from the seven channels (F3, F4, C3, Cz, C4, P3, and P4) was converted into a Pandas DataFrame structure to facilitate manipulation.

The continuous EEG signal was segmented into non-overlapping 5 s epochs. To ensure data quality, an automated artifact-rejection protocol was implemented. Epochs exhibiting peak-to-peak amplitudes exceeding 100 µV or flat signals were excluded from the analysis.

Following artifact rejection, a band-pass filter (1–30 Hz) was applied to all channels to remove low-frequency drift and high-frequency noise. Spectral analysis was performed by calculating the Power Spectral Density (PSD, Welch method) for each valid epoch. The average PSD across epochs was computed for each channel, specifically extracting the mean (across frequencies) power in the alpha (8–12 Hz) and beta (13–30 Hz) frequency bands for both paper and screen reading conditions.

Given the small sample, we opted to conduct the following analytical strategy: first, alpha and beta power values were averaged across seven recorded channels, yielding one value for each frequency band, participant, and reading condition; then, post hoc exploratory channel-wise analyses without multiple comparison correction were subsequently conducted to describe how condition-related pattern was broadly distributed across the recorded electrodes. Given the sparse montage, these analyses should not be interpreted as evidence of anatomically localized effects. The second analysis was conducted only to further explore the result and its possible interpretation.

2.5. Statistical Analysis

Statistical analyses were conducted using JAMOVI (v2.6.19). Reading time, comprehension accuracy, and mean alpha- and beta-band power were compared between paper- and screen-based reading using two-tailed paired-samples t-tests. Because the sample was small, Wilcoxon signed-rank tests were additionally reported as sensitivity analyses to assess the robustness of the results to distributional assumptions and potential outliers. The significance threshold was set at α = 0.05. The absence of a statistically significant difference was interpreted as a lack of detectable evidence of a condition effect. Both parametric and nonparametric tests were applied in parallel to ensure robustness of the findings against potential violations of normality and the influence of outliers, given the modest sample size.

3. Results

First, of the initial twenty participants, five were not considered in subsequent analyses because they presented excessive numbers of EEG epochs rejected due to artifacts. Among the remaining fifteen subjects, the percentage of epochs rejected was comparable between Text A and Text B, averaging 34% (s.e. = 4%) and 35% (s.e. = 5%), respectively. Second, when comparing Text A with Text B, we found statistically significant differences both in reading time (paired-samples t-test t-stat = 15.1, df = 14, p < 0.001; Wilcoxon signed-rank test W = 120, p < 0.001) and number of correct responses (paired-samples t-test t-stat = 3.79, d.f. = 14, p = 0.003; Wilcoxon signed-rank test W = 72.5, p = 0.009) in the post-reading questionnaires. These differences indicate that the texts were not equivalent in difficulty or reading demands, and they reinforce the importance of the counterbalanced within-subject design.

No statistically significant difference in reading time was detected between paper- and screen-based reading (paired-samples t-test t-stat = 0.195, p = 0.85; Wilcoxon signed-rank test W = 66, p = 0.76; Figure 2). Similarly, no statistically significant difference in comprehension accuracy was detected between the two reading media (paired-samples t-test t-stat = −0.31, d.f. = 14, p = 0.76; Wilcoxon signed-rank test W = 34, p = 0.72, Figure 2).

In contrast to the behavioral findings, the spectral analysis showed a condition-related difference in scalp-recorded beta-band power (Figure 3). Alpha-band power did not differ significantly between paper- and screen-based reading (paired-samples t-test p = 0.47; Wilcoxon signed-rank test p = 0.10, see

Table 1. Mean scalp (across channels) comparisons of spectral power in the alpha and beta frequency bands during paper- and screen-based reading.

Band		Statistic	df	p	Mean Difference	SE Difference		Effect Size
Alpha	Student’s t	0.737	14.0	0.473	27.7	37.6	Cohen’s d	0.190
	Wilcoxon W	89.0		0.107	50.0	37.6	Rank biserial correlation	0.483
Beta Screen	Student’s t	2.764	14.0	0.015	36.7	13.3	Cohen’s d	0.714
	Wilcoxon W	103.0		0.012	26.1	13.3	Rank biserial correlation	0.717

for further details). Beta-band power was lower during screen-based than paper-based reading according to both the paired-samples t-test and the Wilcoxon sensitivity analysis. (paired-samples t-test p = 0.015; Wilcoxon signed-rank test p = 0.012, see Table 1). Exploratory channel-wise comparisons showed the same directional pattern at most recorded electrodes (p < 0.05, see Figure 4 and

Table 2. Post hoc channel-wise comparisons of spectral power in the alpha and beta frequency bands during paper- and screen-based reading. The reported p-values were not corrected for multiple comparisons and are provided for exploratory purposes only, complementing the primary analysis of mean scalp-recorded power.

Channel	Paper–Screen	Statistic	d.f.	p	Mean/Median Diff.	S.E.		Effect Size
F3	Student’s t	1.15	14.0	0.270	21.5	18.7	Cohen’s d	0.297
	Wilcoxon W	82.0		0.229	21.4		Rank biserial correlation	0.367
F4	Student’s t	1.34	14.0	0.202	26.9	20.1	Cohen’s d	0.346
	Wilcoxon W	96.0		0.041	33.7		Rank biserial correlation	0.600
C3	Student’s t	2.51	14.0	0.025	54.1	21.5	Cohen’s d	0.649
	Wilcoxon W	96.0		0.041	47.9		Rank biserial correlation	0.600
Cz	Student’s t	2.92	14.0	0.011	37.4	12.8	Cohen’s d	0.754
	Wilcoxon W	105.0		0.008	33.6		Rank biserial correlation	0.750
C4	Student’s t	3.406	14.0	0.004	36.72	10.8	Cohen’s d	0.8795
	Wilcoxon W	112.0		0.002	30.8		Rank biserial correlation	0.867
P3	Student’s t	3.209	14.0	0.006	97.0	30.2	Cohen’s d	0.828
	Wilcoxon W	114.0		<0.001	102.6		Rank biserial correlation	0.900
P4	Student’s t	3.058	14.0	0.009	72.6	23.8	Cohen’s d	0.7895
	Wilcoxon W	108.0		0.004	62.4		Rank biserial correlation	0.800

for further statistical information) except F3 and F4, suggesting that this finding is indeed global and not specific to a brain region.

4. Discussion

The main objective of this study was to investigate whether a wearable EEG system could detect condition-related spectral differences during paper- and screen-based reading. The main finding was lower scalp-recorded beta-band power during screen-based reading, in line with our initial hypothesis. These results indicate that reading medium was associated with different electrophysiological patterns, despite an absence of statistically detectable behavioral differences.

First, these findings should not be interpreted as evidence that a sparse mobile EEG is equivalent to high-density laboratory systems. Rather, they suggest that, for a narrowly defined within-subject spectral comparison, a seven-channel mobile platform can provide analytically useful electrophysiological information. This fit between instrument capability and research question supports the potential use of resource-efficient EEG in scalable and ecologically oriented studies, particularly when portability, preparation burden, and access to specialized infrastructure are relevant constraints. No statistically significant differences in reading time or comprehension accuracy were detected between paper- and screen-based reading. However, reading time and comprehension differed substantially between the two texts, indicating that content characteristics influenced behavioral performance. It is important to mention that the experimental design balanced both texts between paper and screen conditions. Therefore, the specific characteristics of text A or text B were equally distributed between the media groups.

Complementing the behavioral findings, the reduced beta-band power observed during screen-based reading is noteworthy from a neurophysiological perspective. The finding is in line with our initial hypothesis of beta waves’ association with sustained attention and executive function engagement [14]. This would also align with recent findings in the developmental literature [13] which showed a parallel higher theta/beta ratio in children reading on screen compared to paper, even though reading comprehension did not differ significantly. This lends support to the conjecture that digital environments may be related to cognitive demands that compete with the neural resources required for attentional engagement and cognitive efficiency [9,13]. However, this effect lacks specificity, as beta-band activity has been associated with a wide range of cognitive processes, making it difficult to demonstrate that the observed change is due to any particular cognitive function. Moreover, it is important to emphasize that these interpretations remain speculative and should be regarded as hypotheses rather than established conclusions. They require rigorous empirical testing and more in-depth investigation in future studies. Our intention here is merely to offer a few preliminary insights that may motivate and justify further research in this area.

Many adults have experienced the transition from print to digital media. One hypothesis known as screen inferiority [10] suggests poorer performance in screen reading regardless of the reader’s familiarity with the medium, as described in meta-analyses [9,16]. From a cognitive perspective, it is known that reading on paper can engage attentional networks and promote visual–motor integration [17]. The authors argue that reading relies on a distributed brain network rather than a single brain region, and that this network evolves with reading expertise. They further emphasize that environmental factors, particularly the home literacy environment, play a substantial role in shaping brain development. However, studies investigating whether the neural benefits associated with reading are equivalent across paper and digital media remain scarce, making this an important open question.

Moreover, the findings of Ackerman and Lauterman [11] suggest that reading on paper leads to better comprehension than reading on a screen when participants are under time pressure. Interestingly, and consistent with the findings of the present study, this advantage disappears when time constraints are removed. Although participants reported a preference for reading on paper, the authors concluded that preference played only a secondary role. Instead, they argued that the critical factor was metacognitive regulation, specifically, the ability to effectively monitor and allocate time during reading, rather than the screen technology itself. These findings raise an important question regarding how different reading media influence metacognitive calibration and the tendency toward overconfidence when reading on electronic devices. Complementarily, Jeong and Gweon [18] concluded that although reading on paper provided a more positive experience, reading performance was largely equivalent across paper, computers, and tablets.

This study is subject to many limitations. First, the modest sample size and recruitment primarily from an academic environment limit the generalizability of the findings. This may explain the lack of between-media statistical findings on behavioral data, due to lack of power. In addition, because our hardware did not include dedicated ocular or muscular recordings, residual contributions from eye movements, facial muscles, posture, or motor interaction with the reading medium cannot be fully excluded, particularly in the beta band. With only seven EEG channels, no EOG recordings, and relatively short recording durations, there is insufficient information to reliably estimate stable independent components using ICA. Effective source separation requires adequate spatial sampling and sufficiently long recordings to achieve robust decomposition. Moreover, individual reading habits, daily screen exposure, device familiarity, and preference for paper or digital media were not assessed. Because the participants were mostly from an academic environment, the high educational attainment may be associated with reading habits and digital familiarity that are substantially different from the general population. Finally, there were no comparisons with conventional high temporal sampling and high-density EEG to investigate the similarity of results and quantify how much information was being missed by the wearable system. High-density systems make source localization viable, and high temporal sampling allows accurate investigation of gamma waves.

In conclusion, the observed reduction in beta power is consistent with our initial hypothesis that attentional engagement may be lower during screen reading, compared to paper reading. However, our findings should be interpreted as converging evidence supporting this hypothesis, rather than as direct evidence that attention itself was reduced. More broadly, the results support the practical potential of seven-channel wearable EEG for detecting condition-related spectral differences during naturalistic cognitive tasks. The value of this approach lies in matching the technical capabilities of an accessible and portable system to a focused research question.