Screen-time exposure and modality-specific working memory capacity in adolescents

Participants

A total of 81 adolescents participated in the study, consisting of 42 females (mean age±standard deviation=14.9 ± 1.3 years) and 39 males (mean age±standard deviation=15.4 ± 1.5 years). The participants were recruited from three government schools under the Zilla Panchayath in the Dakshina Kannada District of Karnataka, India. All testing took place in a quiet room within the school premises, with ambient noise levels confirmed below 40 dB SPL (A-weighted) using a Class I sound level meter (SL-4036SD, Lutron Instruments, Taiwan).

All participants demonstrated normal hearing sensitivity, with binaural air-conduction pure-tone thresholds of 15 dB HL or better at octave frequencies from 250 Hz to 4 kHz, as measured by a calibrated diagnostic audiometer (Elite, Acoustics Digital, India) paired with TDH 39 P circum-aural headphones (Telephonics Corporation, USA). Additionally, outer hair cell function was confirmed through a pass in transient evoked otoacoustic emissions (TEOAEs) screening, conducted using an Otoport Lite OAE screener (Otodynamics Inc., USA). Further inclusion criteria required participants to score within normal limits on the Screening Checklist for Auditory Processing Disorders in Children (SCAP-C) [16], 17] and to exhibit 6/6 corrected or uncorrected visual acuity, as assessed using a Snellen chart at a 2-foot distance. Parents and adolescents were interviewed to exclude any history of otological, neurological, speech-language, or intellectual impairments to ensure participant eligibility. Informed written consent to participate in the study was obtained from the parents of each child before participation, and the Institutional Ethics Committee approved the study. Adherence to the Declaration of Helsinki principles was maintained throughout the study [18].

Procedure

Working memory assessment

Three tasks were employed to evaluate WM across auditory-verbal, visual-verbal, and visuospatial domains (Figure 1). The Auditory Reverse Digit Span (ARDS) task targeted phonological WM and auditory-verbal processing. Stimuli consisted of English spoken digits (1–9) recorded from a 23-year-old female native speaker at a consistent amplitude and intonation. The ARDS stimuli were presented binaurally through professional-grade Audio-Technica ATH-M50X headphones calibrated to 65 dB SPL. The reverse digit span task, programmed using Paradigm Software [20], consisted of five trials, each with a digit sequence ranging progressively from 3 to 8 items. Each digit was presented for 500 ms with an equivalent inter-stimulus interval, followed by a 10-s response window. A black fixation cross (+) appeared for 500 ms on a white background before each trial to orient attention. During the response window, the participants were instructed to carefully listen to the sequence of digits and type it in reverse order on a standard QWERTY keyboard.

Figure 1:

Schematic representation of the assessment procedure for corsi backward (CB, top row), auditory reverse digit span (ARDS, middle row), and visual reverse digit span (VRDS, bottom row).

The Visual Reverse Digit Span (VRDS) task paralleled the ARDS in structure but utilised the visual presentation of Indo-Arabic numerals (1–9) to assess visual-verbal WM. VRDS stimuli were displayed at the centre of a 21.5” LCD screen (1920 × 1,080 resolution) positioned at a 75 cm viewing distance (0° azimuth, eye level). Each 1-inch numeral appeared for 500 ms against a white background, with identical sequence lengths, trial numbers, and timing parameters as the ARDS. Participants again responded via keyboard, maintaining response modality consistency with the ARDS.

For visuospatial WM assessment, the Corsi Backward Task (CB) was implemented through PsychToolkit [21] and displayed via Google Chrome (Google Inc., USA). The task presented a 4 × 3 grid (8 × 6 inches) containing 12 equal-sized squares. On each trial, a pseudorandom sequence of squares (3–8 items) was highlighted for 500 ms each with 500-ms intervals. Participants reproduced sequences in reverse order by mouse-clicking the squares, with response accuracy and latency recorded. The display setup was similar to the VRDS task setup to maintain visual presentation consistency across modalities.

No practice trials were provided for testing to minimise potential practice effects and ensure that performance reflected participants’ baseline WM capacity.

Scoring

The Paradigm software automatically recorded all responses, and the output files were subsequently verified by two independent researchers (the study’s third and fourth authors) to ensure accurate scoring. For each WM task, the lowest string level at which participants reliably achieved at least three correct responses out of five trials (60 %) determined the WM span for that measure.

Statistical analysis

Statistical analysis was conducted using JASP software version 0.19.3 [22]. The data exhibited a non-normal distribution. Descriptive statistics were applied to determine the median and interquartile range for all the data. The Mann-Whitney U test compared performance between groups with low and high screen time across tasks. The Wilcoxon signed-rank test was utilised for pairwise comparison. Binary logistic regression analyses evaluated the predictive relationship between screen time groups and WM performance. The model fit was evaluated using metrics such as chi-square (χ2), Nagelkerke (R²), and Area Under the Curve (AUC). The low and high screen-time groups for AST and BST were defined using median splits of their respective overall scores.

Further, a post-hoc power analysis was conducted using G*Power 3.1 to determine the study’s ability to detect significant effects [23]. Power was estimated for key analyses, including Mann-Whitney U tests (group comparisons), Friedman tests (within-group differences), and binary logistic regression (predictive models), using the observed effect sizes and sample sizes. All power calculations assumed a two-tailed α-level of 0.05 and that true effect sizes, appropriate group sizes, and the statistical test’s underlying assumptions were met.

Screen time characteristics of the study participants

The study population had higher AST (Mdn=5.0, IQR=2.75–8.50, n=81) compared to BST (Mdn=0.50, IQR=0.00–1.25, n=81) (Figure 2). Wilcoxon signed-rank test findings suggested that the participants had significantly higher AST than BST (Z=−7.27, p<0.001, rrb=−0.95). To examine the impact of screen time on WM performance, the participants were divided into low and high AST and BST groups based on the 50th percentile values of 0.5 h for BST and 5.0 h for AST.

Figure 2:

Raincloud plot depicting active (AST) and background (BST) screen time characteristics (hours/week) in the study population.

Post division, the low AST group had a median screen time of 3.0 h/week (IQR=1.50–4.31, n=44), and the high AST group had a median screen time of 8.50 h/week (IQR=6.50–11.50, n=37). The low BST group had a median screen time of 0.25 h/week (IQR=0.00–0.50, n=55), and the high AST group had a median screen time of 1.50 h/week (IQR=1.50–3.18, n=26). A Mann-Whitney U test analysed the group differences in AST and BST. As expected, the low AST (U=1,430.00, p<0.001, rrb=1.00) and BST (U=1,628.00, p<0.001, rrb=1.00) groups had significantly lower screen time compared to the high AST and BST groups.

Screen time effect on working memory measures

Three WM measures were used in the present study. Mann-Whitney U test found no statistically significant group differences between high and low BST groups in CB (U=816.50, p=0.28, rrb=0.14), ARDS (U=890.50, p=0.79, rrb=−0.03) and VRDS (U=896.00, p=0.84, rrb=−0.03) performances, suggesting no significant impact of BST on WM abilities (Figure 3a–c).

Figure 3:

Raincloud plot representing working memory performance across different screen time conditions. The top row compares performance in low vs. high background screen time (BST) groups, with panel (a) showing corsi backward (CB), panel (b) displaying auditory reverse digit span (ARDS), and panel (c) illustrating visual reverse digit span (VRDS) for visual working memory. The bottom row examines low vs. high active screen time (AST) groups, where panel (d) corresponds to CB, panel (e) to ARDS, and panel (f) to VRDS. An asterisk (*) indicates a statistically significant difference at p=0.05.

In contrast, high AST group performed significantly better than low AST group in CB (U=1,254.50, p<0.001, rrb=0.54), ARDS (U=1,215.50, p<0.001, rrb=0.48) and VRDS (U=1,098.50, p=0.003, rrb=0.35) performances, indicating that active engagement with the screen correlated with visuo-spatial, and verbal WM (Figure 3d–f). Post-hoc power analyses revealed that the study was well-powered to detect medium-to-large effects in primary comparisons. Mann-Whitney U tests for AST group differences (CB: r=0.54, power=99 %; ARDS: r=0.48, power=96 %; VRDS: r=0.35, power=80 %).

Within-group comparisons of working memory performance

The Friedman test revealed significant differences in WM performance across tasks in all four experimental groups. In the high BST group, a significant overall effect was observed (χ² (2)=20.27, p<0.001, Kendall’s W=0.35), with Conover’s post-hoc comparisons showing CB outperforming both ARDS (T=3.27, p=0.002, r=0.69) and VRDS (T=5.45, p<0.001, r=0.80), and ARDS surpassing VRDS (T=2.18, p=0.03, r=0.71) (Figure 4a). The low BST group also exhibited a similar pattern (χ² (2) = 12.79, p=0.002, W=0.29), with CB significantly higher than ARDS (T=3.31, p=0.002, r=0.69) and VRDS (T=3.82, p<0.001, r=0.676), but no difference between ARDS and VRDS (T=0.52, p=0.61, r=0.10) (Figure 4b).

Figure 4:

Mean reverse span scores (±SE) are shown for three working memory tasks: Corsi block (CB), auditory reverse digit span (ARDS), and visual reverse digit span (VRDS), separately for participants with (a) high background screen time (BST), (b) low BST, (c) high active screen time (AST) and (d) low AST.

For AST, the high exposure group demonstrated the strongest effects (χ² (2)=20.03, p<0.001, W=0.42), with CB span significantly greater than ARDS (T=3.48, p= 0.001, r=0.88). Further, the VRDS performance was significantly poorer than ARDS (T=2.21, p= 0.03, r=0.66) (Figure 4c). The low AST group showed a similar hierarchy but with weaker differentiation (χ² (2) = 13.24, p=0.001, W=0.25): CB span was significantly higher than ARDS (T=3.15, p=0.003, r=0.64) and VRDS (T=3.86, p<0.001, r=0.69), while ARDS and VRDS did not differ (T=0.71, p=0.48, r=0.13) (Figure 4d). Friedman tests for within-group WM performance (all W≥0.25, power≥92 %) achieved adequate-to-high power (>80 %). However, comparisons involving BST groups were underpowered (24 % for r=0.14) due to smaller effect sizes and unequal group sizes (low BST: n=55; high BST: n=26).

Logistic regression analysis: predicting screen time level from working memory performance

Binary logistic regression analyses were conducted to evaluate whether WM span measures, CB, ARDS, and VRDS, could predict individuals’ levels of AST and BST. For both models, screen time was categorised into high and low levels. Figure 5a depicts the ROC curve (95 % CI) for the best-performing AST and BST models.

Figure 5:

Results of binary logistic regression analyses. Panel (a) displays the ROC curves for the best-performing active screen time (AST) and background screen time (BST) models. Panels (b) and (c) show conditional estimates (95 % CI) for corsi backward (CB) scores in the BST and AST groups, respectively, while panel (d) presents the same for auditory reverse digit span (ARDS) scores in the AST group.

The BST models did not show significant predictive value. Only CB entered the model, but the improvement over the intercept-only model was not statistically significant (χ² (1)=2.36, p=0.13, Nagelkerke R²=0.04) (Figure 5b). CB was also not a significant predictor (B=−0.38, SE=0.25, z=−1.51, p=0.13). Although the model’s overall accuracy was 71.1 %, the AUC was 0.57, suggesting limited discrimination between high and low BST groups.

In contrast, for AST, three models were evaluated using forward selection. Model 1 included only the intercept, Model 2 included the CB score, and Model 3 included both CB and ARDS. Model 2 significantly improved upon the intercept-only model (Δχ² (1)=22.39, p<0.001, Nagelkerke R²=0.32). Model 3, which included both CB and ARDS, provided a further significant improvement over Model 2 (Δχ² (1)=9.41, p=0.002) and explained 43.4 % of the variance (Nagelkerke R²=0.43). In Model 3, both CB (B=−1.11, SE=0.36, Z=−3.13, p=0.002) and ARDS (B=−1.18, SE=0.41, Z=−2.79, p=0.005) emerged as significant predictors of AST level (Figure 5c and d). These results indicate that poorer performance on visual and auditory WM tasks was associated with a greater likelihood of belonging to the low AST group. The model achieved an overall classification accuracy of 76.5 %, with an AUC of 0.83, indicating strong discriminative ability. Logistic regression models predicting AST levels (Nagelkerke R²=0.43) demonstrated excellent power (99 %).

Effect of high vs. low screen time on working memory tasks

The comparison between high and low screen time groups showed clear differences in WM performance based on how screens were used. For AST, participants in the high-exposure group performed much better on all WM tasks than those in the low-exposure group. The performance advantage observed in the high AST group suggests that engaging with digital screens may provide broad cognitive benefits that apply to various domain-specific WM. The findings are further corroborated by logistic regression analysis, which indicated that CB and ARDS are important predictors of high vs. low AST use.

Moderate screen time exposure improves auditory performance, while excessive exposure harms auditory processing and WM, as shown by Jain et al. [27] and supported in the present study. The relationship between screen time and cognition appears non-linear, with a trade-off emerging: gains in visuospatial WM may coincide with losses in other modalities. Consistent observations of this pattern across studies confirm that screen time’s effects on WM are modality-specific.

The superior performance across all WM tasks in the high AST group may reflect several underlying mechanisms. Firstly, the dynamic and interactive nature of screen-based activities promotes the development of generalised cognitive control abilities that transfer to WM tasks. Second, frequent switching between different types of digital content may enhance cognitive flexibility, allowing for more efficient allocation of attentional resources across modalities. This explanation is supported by Li et al. [28], who found that passive exposure to media resulted in poorer inhibitory control scores but not the active engagement. In addition, a scoping review from Shaleha & Roque [29] concluded that active screen exposure is associated with better cognitive outcomes, such as attention, memory, and executive functions. Although the results are from various age groups, they highlight the influence of active exposure on executive functions like inhibitory control. Together with our study results, these findings provide a reasonable basis to interpret the observable patterns.

In contrast to these AST-related effects, the analysis of BST revealed no significant performance differences between high and low-exposure groups for any WM measure. This null finding persists despite adequate statistical power to detect medium-sized effects, suggesting that passive screen exposure may have minimal immediate impact on WM performance at the exposure levels observed in this study, but, considering the small sample size in the current study and lesser BST duration among participants, the results may be inconclusive. Hence, future studies should consider this limitation and validate this finding. Further, the dissociation between AST and BST effects aligns with cognitive load theory [30], which posits that the nature of engagement with media is more consequential than mere exposure duration. While active use requires ongoing cognitive processing and adaptation, passive background exposure may not impose sufficient cognitive demands to enhance or impair WM functioning. This interpretation is supported by research showing that the cognitive impacts of screen time are strongly moderated by exposure type [26], 28], 29].

Within-group comparisons of Working memory performance

Our findings show that CB consistently outperformed ARDS and VRDS across all screen time groups. This aligns with research suggesting that screen-based activities enhance visuospatial WM [31], [32], [33]. The CB task mainly involves visuospatial WM, while the VRDS task depends exclusively on visual-verbal processing. These results differ from the study [26], which found a significant association between increased screen time and VRDS performance but saw no notable effects on ARDS or CB span tasks. This difference may be due to variations in the age groups studied in the two studies, suggesting that the effect of screen time on modality-specific WM could differ across age ranges due to variable content exposure in them. These findings emphasise the need to examine screen time effects across different age cohorts rather than generalising findings from one group to all.

Further, the consistent superiority of visuospatial WM (CB task) over auditory-verbal tasks (ARDS/VRDS tasks) may also reflect differential engagement of components within WM assembly (Baddeley, 2000) [34]. In reverse digit span tasks, the participation of the phonological loop is also required, and entails more demanding serial verbal processing. In these two tasks, the central executive component is also tapped; on the other hand, CB requires only the visuo-spatial sketchpad for chunking and manipulation of spatial sequences to give the response. This could be less demanding than the phonologically-loaded digit span task, which could lead to a performance decrement due to manipulation. In addition, the central executive is taxed by reverse digit span tasks of both types, which suffer greater performance decrements due to the demands on multiple components that are in action during processing and response production.

Using digits as stimuli in both ARDS and VRDS inherently engages linguistic and phonological processing systems. While digits represent a standardised and familiar set of verbal materials suitable for assessing phonological WM, their use may bias task performance toward verbal WM advantages compared to purely visuospatial tasks. Consequently, observed differences in WM capacity across modalities may partly reflect this linguistic advantage. Future studies could consider employing non-linguistic stimuli or matched difficulty levels to further dissociate modality-specific WM processes.

While performance on CB demonstrated consistent strength regardless of screen time exposure, statistical analyses revealed a crucial dissociation between auditory and visual-verbal WM performance. The Friedman tests, followed by Conover post-hoc comparisons, showed that adolescents in high AST and BST groups performed significantly better on the ARDS than the VRDS task; these findings support the auditory superiority hypothesis [35]. Interestingly, no significant differences emerged between auditory and visual-verbal WM performance in the low screen time groups, with the important caveat that this advantage only manifests under higher screen time exposure conditions. Furthermore, performance discrepancy on ARDS-VRDS can be explained by these two respective WM tasks engaging overlapping phonological loop resources, where auditory inputs convert to phonological codes and visual-verbal stimuli into a shared verbal format. This encoding process, supported by neural networks like the supramarginal gyrus [36] and arcuate fasciculus [37], reduces modality performance differences. However, high screen time disrupts this balance. Excessive screen use may overload the phonological loop, especially for the VRDS task, necessitating visual-to-verbal conversion, whereas ARDS benefits from direct auditory pathways. Screen multitasking can impair visual-verbal integration by splitting attentional resources, negatively affecting VRDS performance. Additionally, background screen noise and clutter may hinder dorsal stream processing, essential for visual-phonological conversion in VRDS, while ARDS remains protected through auditory input [38].

The observed AST-related enhancement of CB performance may further reflect adolescents’ adaptation to screen-rich environments. Adolescents with greater screen time exposure may develop cognitive adaptations that help transfer these experiences to tasks involving visuospatial manipulation, leading to better performance on the CB task than the VRDS task. This interpretation aligns with findings of a study [39], which reported that high-media multitaskers, particularly young adults, demonstrated superior performance in task-switching paradigms compared to their low-media multitasking counterparts. Their results suggest that exposure to dynamic and complex visual patterns may enhance cognitive flexibility, supporting improvements in visuospatial WM. Also, better performance on CB may reflect the distinct neural substrates of spatial processing, which rely more on parietal networks rather than the prefrontal regions implicated in verbal WM [40].

In the present study, it is important to acknowledge that the stimuli used in the ARDS and VRDS tasks consist of linguistic elements (i.e., digits), which differ from the non-linguistic stimuli used in the CB task. The use of digits may have contributed to differences in performance across modalities in the reverse digit span task. However, due to extensive exposure in daily life and academic settings, digits are likely to be highly familiar and automated in adolescents. Future research should consider this factor and employ comparable task designs to systematically assess auditory, visual, and visuospatial WM.

This study and prior research on digital media use suggest that screen exposure influences modality-specific WM in adolescents, a period marked by significant prefrontal cortex maturation [41]. While screen time has been linked to cognitive functions such as WM, attention, and processing speed, methodological constraints limit these findings to correlational interpretations. In this context, establishing causality requires robust experimental designs, such as controlled training or longitudinal studies. The findings further advocate for exercising improved variable control, implementing objective screen time measurement, and precise cognitive assessments in future research.

This study is among the few investigations in the Indian context examining the effects of screen time exposure on a specific cognitive process, like modality-specific WM. A key strength of the study is its classification of screen time exposure based on the nature of engagement, distinguishing between AST and BST. While the study was sufficiently powered to detect AST-related effects, the nonsignificant findings for BST should be interpreted cautiously due to low statistical power (24 %). This limitation likely stems from restricted variability in BST exposure (median=0.5 h/week) and the smaller high-BST subgroup (n=26). Future studies should recruit larger samples with balanced BST exposure levels to improve sensitivity for detecting subtle effects. Additional limitations include the reliance on self-reported screen time data, which may introduce measurement bias, and the restricted geographic sampling from only three schools in the Dakshina Kannada district, potentially limiting generalizability to broader populations. Another limitation of the study is that socio-economic status was not assessed in detail or analysed in the current study. Future research in this context could consider examining socio-economic status more systematically to investigate its association with screen exposure and influence on WM. The exclusive focus on adolescents also questions whether these patterns extend to other age groups. These methodological constraints highlight important directions for future research, including the need for larger, more diverse samples with balanced screen exposure groups, incorporation of objective screen time monitoring methods, and expansion to different demographic and regional populations.