Losing Track of Time on TikTok? An Experimental Study of Short Video Users' Time Distortion

Short video platforms have become globally popular due to immersive, personalized, auto-scrolling content and short durations, which may foster problematic use and influence time perception. Prior work suggests a duration effect in time estimation (short tasks overestimated, long tasks underestimated) across activities such as gaming and reading, but evidence for short video contexts is limited. Differences in time perception across activities (gaming, pornography, TV series) have been observed, yet comparative data for short videos versus other tasks are scarce. Building on findings that frequent short video use can predict overestimation for a 15-min task, this study examines whether task duration (short vs. long) and task type (watching short videos vs. reading) affect time perception, and whether individual differences (problematic short video watching, weekly usage—estimated and actual) relate to time distortion. Research questions: (1) How do users estimate time for tasks of different durations and for weekly short video use? (2) What is the relationship between problematic short video watching (PSVW) and time distortion in tasks and weekly use? (3) What is the relationship between weekly short video use (estimated and actual) and time distortion in tasks and weekly use?

Problematic short video watching (PSVW) is characterized by excessive short video consumption leading to functional impairments, aligned with broader online behavioral addiction frameworks (e.g., components model; I-PACE). The I-PACE model posits addictive behaviors evolve via reduced inhibitory control and compensation-driven engagement. Time perception involves subjective judgment of duration, pace, and order, influenced by internal (emotions, cognition) and external (task difficulty, stimulus order, music) factors, and can be studied through prospective vs. retrospective paradigms. Attentional models (processing-time, attentional gate, resource allocation) propose limited attentional resources are shared between temporal and non-temporal information, with engaging activities reducing attention to time and accelerating perceived passage. Memory models suggest duration judgments rely on stored event markers; segment-rich tasks lead to overestimation. Internal clock models (pacemaker-accumulator-comparator) describe timing mechanisms. Prior findings on online behaviors and time distortion are mixed: gamers may overestimate or underestimate depending on expertise and context; pornography/TV series often feel faster; social media addiction correlates with time distortion. Short video platforms differ from gaming and other social media in format and interaction, warranting specific study. A recent study found daily short video frequency predicted overestimation of 15-min tasks and that reading was estimated more accurately than short video watching, but only one duration was tested. This study addresses gaps by examining duration effects, task-type differences, and links to PSVW and weekly usage.

Design: 2 (time duration: long 16 min 9 s vs. short 5 min 23 s; between-subjects) × 2 (task type: watching short videos vs. reading public articles; within-subjects) mixed experimental design. Durations were chosen to approximate 15 min and one-third of that, with non-round values to prevent guessing. Participants: N = 56 college students from a comprehensive university in Jiangsu, China (mean age = 20.27, SD = 2.23; 17–25 years; 55.4% female). Inclusion required recent short video app use (e.g., Douyin/TikTok, Kuaishou). A priori power analysis (G*Power 3.1) indicated n ≈ 52 for α = 0.05, power = 0.80, large effect size d = 0.4. Procedure: Participants were randomly assigned to long or short duration conditions and performed both tasks (short video watching and reading) for the same duration, with counterbalanced task order. Short video task used participants’ own accounts to view personalized feeds; reading task used credible, engaging WeChat science accounts (e.g., Guokr, DingXiang Doctor), with moderately long, high-view articles selected by authors. After both tasks, participants retrospectively estimated the duration range (minimum and maximum seconds) for each task; the mean of these bounds was used as the estimated duration. Then, participants completed questionnaires including demographics, PSVW scale, estimated weekly short video use, and extracted actual weekly use via phone screen-time features (iOS Screen Time or Android Digital Wellbeing), selecting the Short-Video Apps category (Douyin/TikTok, Kuaishou/Kwai). Participants confirmed they did not use electronic tools to estimate time during tasks. Measures: (1) TE/TA for tasks: Estimated duration (TE) computed as mean of lower and upper bounds; TE/TA = TE / actual duration (5:23 or 16:09). TE/TA = 1 denotes accuracy; <1 underestimation; >1 overestimation. (2) TE/TA for weekly use: ratio of self-reported weekly short video time to phone-recorded actual weekly time. (3) PSVW: Problematic Short Video Usage Assessment Questionnaire for Adolescents (20 items, 5-point Likert; total 20–100). Higher scores indicate higher PSVW; reliability Cronbach’s α = 0.838. Analyses: Computed TE/TA ratios; descriptive statistics; correlations; one-sample t-tests against 1 to assess distortion; repeated-measures ANOVA for main effects of duration and task type and interaction; group-based ANOVAs (PSVW high/moderate/low; weekly use estimated/actual high/moderate/low using 27% cutoffs); chi-square tests on categorical over/underestimation.

Descriptives and correlations (n = 56):

• PSVW mean = 64.946 (SD = 10.788). TE/TA short video: M = 1.091 (SD = 0.402). TE/TA reading: M = 1.019 (SD = 0.456). Estimated weekly use: M = 937.16 min (SD = 677.20). Actual weekly use: M = 778.55 min (SD = 553.85). TE/TA weekly: M = 1.535 (SD = 1.212).
• Correlations: TE/TA short video correlated with TE/TA reading (r ≈ 0.395, p < 0.01). Estimated weekly use correlated with actual weekly use (r = 0.596, p < 0.01). Actual weekly use negatively correlated with TE/TA weekly (r ≈ −0.398, p < 0.01). Time distortion in tasks and weekly use:
• One-sample t-tests versus 1: • Short video task, 5:23: TE/TA = 1.206 ± 0.443; t(27) = 2.458, p = 0.021 (overestimation). • Short video task, 16:09: TE/TA = 0.977 ± 0.326; t(27) = −0.377, p = 0.709 (ns). • Reading task, 5:23: TE/TA = 1.144 ± 0.514; t(27) = 1.487, p = 0.149 (ns). • Reading task, 16:09: TE/TA = 0.893 ± 0.357; t(27) = −1.587, p = 0.124 (ns). • Weekly use TE/TA: 1.535 ± 1.212; t = 3.301, p = 0.002 (overestimation).
• Repeated-measures ANOVA (2 durations × 2 task types): main effect of duration significant, F(1,54) = 6.950, p = 0.011, ηp² = 0.114 (shorter duration yields higher TE/TA). Task type not significant, F(1,54) = 1.287, p = 0.262 (η² ≈ 0.024). Interaction not significant, F(1,54) = 0.031, p = 0.860. PSVW and time distortion:
• Grouping by PSVW (high n = 15, moderate n = 26, low n = 15): Weekly TE/TA overestimation evident in moderate group (M = 1.710, SD = 1.380; t(25) = 2.623, p = 0.015) and marginal in low group (M = 1.534, SD = 1.010; t(14) = 2.046, p = 0.060). No significant group differences for task TE/TA.
• Two-way ANOVA (duration × PSVW group): main effect of duration significant for short video TE/TA, F(1,50) = 5.110, p = 0.028, ηp² = 0.093, and for reading TE/TA, F(1,50) = 4.465, p = 0.040, ηp² = 0.082. Group and interaction effects ns.
• Repeated-measures ANOVA (task × PSVW group): task type, group, and interaction all ns (F(1,53) = 0.987, p = 0.325; F(2,53) = 0.219, p > 0.05; interaction F(2,53) = 0.070, p = 0.932).
• Chi-square: no significant over/underestimation trends by PSVW group for watching (χ²(2) = 0.596, p = 0.742) or reading (χ²(2) = 0.647, p = 0.724); weekly use direction ns (χ²(2) = 1.387, p = 0.500). Weekly use and time distortion:
• Estimated weekly use groups: Weekly TE/TA overestimation in high-frequency group (M = 1.804 ± 0.874; t(14) = 3.562, p = 0.003) and moderate group (M = 1.692 ± 1.534; t(25) = 2.300, p = 0.030). Moderate group also overestimated short video task duration (TE/TA M = 1.206 ± 0.455; t(25) = 2.307, p = 0.030).
• Actual weekly use groups: Weekly TE/TA overestimation significant in low-frequency group (M = 2.621 ± 1.701; t = 3.693, p = 0.002); high and moderate groups ns.
• Two-way ANOVA (duration × estimated weekly-use group): duration main effect significant for reading TE/TA, F(1,50) = 4.246, p = 0.045, ηp² = 0.078; duration effect for short video TE/TA ns, F(1,50) = 3.093, p = 0.085. Group and interaction ns in this model. Repeated-measures ANOVA (task × estimated weekly-use group) showed a significant interaction, F(2,53) = 3.404, p = 0.041, ηp² = 0.114: the moderate-frequency group had higher TE/TA for watching short videos than for reading.
• Chi-square: significant differences in over/underestimation tendencies across estimated weekly-use groups for the short video task (χ²(2) = 6.049, p = 0.049), with the moderate-frequency group tending to overestimate. No significant differences across actual weekly-use groups for the short video task (χ²(2) = 0.678, p = 0.712). Reading task tendencies ns for both actual (χ²(2) = 0.107, p = 0.948) and estimated (χ²(2) = 0.647, p = 0.724). Direction of weekly-use distortion significant across actual groups (χ²(2) = 7.267, p = 0.026) and estimated groups (χ²(2) = 6.885, p = 0.032).

Findings support a duration effect in short video contexts: shorter sessions (≈5 min) are overestimated, while longer sessions (≈16 min) show no significant distortion, aligning with theories of memory segmentation and attentional allocation. The absence of a task-type main effect suggests comparable time perception mechanisms for watching short videos and reading within this experimental context, indicating that time distortion may generalize across common daily activities when task lengths are controlled. Individual differences showed limited impact: PSVW did not significantly modulate task-related time distortion, though moderate and low PSVW groups tended to overestimate weekly use. Self-reported weekly use related to overestimation of weekly engagement and, for moderate-frequency estimators, to overestimation during the short video task, whereas actual weekly use was negatively related to weekly TE/TA ratios, with low actual users overestimating their weekly time. These results nuance prior evidence by emphasizing the role of task duration and by separating estimated versus actual usage patterns. Theoretically, the results can be integrated with attentional models (reduced temporal monitoring under engaging content), memory-based models (greater event segmentation inflating retrospective duration), and the I-PACE model (time distortion as a potential cognitive-affective marker within addictive behavior processes). Practically, the findings suggest design and self-regulation strategies (e.g., autoplay refinements, usage reminders) to help users calibrate time perception and mitigate overengagement.

This study provides experimental evidence that short video users tend to overestimate brief viewing sessions and their weekly engagement, with no robust differences between video watching and reading tasks when duration is controlled. Task duration exerts a significant influence on time perception, whereas PSVW shows minimal impact on task-based time distortion. Distinctions between estimated and actual weekly use highlight that perceived high/moderate users overestimate weekly time, and low actual users also overestimate their weekly use. Contributions include clarifying duration effects in short video contexts, disentangling task-type influences, and situating time distortion within models of online behavioral engagement and addiction. Future research should: (1) replicate with larger, more diverse samples across ages; (2) compare time distortion across varied video behaviors (short videos, binge-watching, live streaming, documentaries) and action-demanding activities (gaming, gambling) using authentic tasks; (3) pretest and refine reading materials; (4) contrast prospective versus retrospective timing paradigms and examine neural underpinnings (e.g., fMRI); (5) include additional duration groups to map nonlinearity in duration effects; and (6) incorporate emotional measures as covariates.

• Ecological design used participants’ own smartphones; although participants reported not using devices to track time, incidental exposure to time displays could have occurred.
• University student sample limits generalizability to other populations (e.g., adolescents, older adults).
• Grouping by cutoffs (27% high/low) may introduce artificial boundaries.
• The decision not to block smartphone time displays (to avoid revealing the study aim) poses a potential confound that future work should manage.
• Emotional states, known to influence time perception, were not measured; inclusion as covariates is recommended in future studies.