Digital Lifestyles

Histograms

Boxplots

Correlation Matrix

Table 2 presents Pearson correlation coefficients among screen time, lifestyle balance indicators and mental health and performance outcomes.

Daily screen time was moderately negatively correlated with sleep duration (r = −0.59), suggesting that individuals with higher device use tended to report fewer hours of sleep. Screen time was also positively associated with stress levels (r = 0.32), indicating that greater digital engagement corresponded with higher perceived stress. Associations between screen time and focus (r = −0.03) and productivity (r = 0.09) were weak.

Sleep duration demonstrated a moderate negative correlation with stress (r = −0.29). Correlations between sleep and focus (r = 0.03) and productivity (r = −0.03) were negligible. Physical activity and study time showed minimal correlations with both screen time and outcome variables.

Importantly, none of the correlations between predictors were sufficiently large to raise immediate concerns regarding multicollinearity.

Notably, a patterned set of associations emerges among screen time, sleep duration and stress. Screen time is negatively related to sleep and positively related to stress, while sleep is inversely associated with stress. Together, these relationships form a pattern of interrelated associations consistent with a potential moderation framework, suggesting that sleep may condition the strength of the relationship between screen time and stress.

Overall, this pattern of associations provides empirical justification for further visual exploration and formal testing of moderation effects in subsequent analyses.

Heatmap

Figure 17 presents a heatmap visualization of the correlation matrix.

Warmer colors indicate positive correlations, whereas cooler colors represent negative associations. The heatmap visually reinforces the moderate inverse association between screen time and sleep, as well as the patterned relationships among screen time, sleep and stress.

Bivariate Relationships Among Screen Time, Sleep & Stress

Figure 18 shows a positive association between daily screen time and stress levels. As device use increases, reported stress appears to increase as well. The linear trend suggests that higher screen exposure may be associated with elevated psychological strain.

Figure 19 demonstrates a negative relationship between screen time and sleep duration. Individuals who report higher device use tend to report fewer hours of sleep. This pattern suggests that sleep may function as a behavioral mechanism linking screen exposure to stress.

Figure 20 indicates a negative association between sleep duration and stress. Participants reporting fewer hours of sleep tend to report higher stress levels. This inverse relationship further supports the possibility that sleep plays a moderating role in the association between screen time and stress.

Taken together, Figures 18–20 reveal a a configuration of correlations consistent with conditional effects: screen time is positively associated with stress and negatively associated with sleep, while sleep is negatively associated with stress. This pattern provides conceptual support for examining whether sleep moderates the relationship between screen time and stress.

Sleep as a Grouping Variable

To further explore this possibility, sleep duration was divided into high and low groups (median split) and plotted as a grouping variable.

Figure 21 displays the relationship between screen time and stress separately for high and low sleep groups. Visual differences in slope suggest that the strength of the association between screen time and stress may vary depending on sleep duration. This observation warrants formal moderation testing.

Data Interpretation

Since the variables were mean-centered prior to running the analysis, the results above do not represent the original data values. As a result, the conversion values back to the original data values are reflected below.

The association between screen time and stress was strongest among individuals sleeping approximately 6 hours per night (β = 0.29, p <0 .001) and weakest among those sleeping approximately 8.5 hours per night (β =0 .10, p < 0.001), indicating that adequate sleep attenuates the relationship between screen exposure and stress.

Furthermore, sleep duration ranged from 3-11 hours in the dataset. Johnson-Neyman analysis indicated that the association between screen time and stress was statistically significant for individuals sleeping less than approximately 9 hours per night. Thus, the analysis indicated that the effect of screen time on stress became non-significant only at very high levels of sleep duration (above ~9 hours per night), suggesting that sufficient sleep substantially buffers the association.

The association between screen time and stress was strongest among individuals reporting poor sleep quality (approximately 1.6 on the 1–5 scale; β = 0.43, p < 0.001) and weakest among those reporting high sleep quality (approximately 3.8; β = 0.17, p < 0.001), indicating that better perceived sleep quality attenuates the relationship between screen exposure and stress.

Johnson–Neyman analysis indicated that the association remained statistically significant across nearly the entire observed range of sleep quality, becoming non-significant only at very high levels (above approximately 4.55 on the 5-point scale).

Although both sleep duration and sleep quality attenuated the positive association between screen time and stress, sleep quality exhibited a more pronounced moderating effect. The reduction in slope across levels of sleep quality (Δβ = 0.26) exceeded that observed for sleep duration (Δβ = 0.19), suggesting that subjective sleep restoration may be more critical than total sleep time in mitigating the psychological impact of screen exposure. These findings imply that qualitative aspects of sleep may serve as a stronger resilience factor against technology-related stress than sleep quantity alone.

Data Interpretation

Since the variables were mean-centered prior to running the analysis, the results above do not represent the original data values. As a result, the conversion values back to the original data values are reflected below.

As shown in Table 11, the relationship between screen time and productivity strengthened as sleep duration increased. This pattern indicates that sleep duration amplifies the association between screen time and productivity. In other words, as sleep increases, the effects of screen time on productivity become stronger.

Johnson-Neyman analysis indicated that the relationship between screen time and productivity is statistically significant when individuals sleep more than ~5.04 hours per night. Given that the observed sleep range in the sample was 3-11 hours, the screen time-productivity association was significant for nearly all typical sleep levels except at extremely low sleep durations.

Overall, these findings suggest that sleep duration significantly moderates the relationship between screen time and productivity. Rather than weakening the association, higher sleep duration strengthens the positive effect of screen time on productivity. Individuals who sleep more appear to translate screen use into productive outcomes more effectively than those who sleep less.

Simple slopes analysis indicated that sleep quality significantly moderated the relationship between screen time and productivity. As depicted in Table 12, higher sleep quality amplifies the positive relationship between screen time and productivity. This suggests that individuals with better sleep quality exhibit greater productivity gains associated with increased screen time.

Johnson–Neyman analysis indicated a region of non-significance between -5.21 and 0.59 on the original sleep quality scale. However the observed sleep quality values in the sample ranged from 1.54 to 5.00 (being rated on a scale from 0-5), placing all observed values outside the non-significant region. This indicates that the relationship between screen time and productivity was statistically significant across the entire observed range of sleep quality. Consistent with the positive interaction coefficient, higher sleep quality strengthened the positive association between screen time and productivity.

Both sleep duration and sleep quality significantly moderated the relationship between screen time and productivity. However, sleep quality demonstrated a stronger moderating effect, as indicated by greater changes in simple slopes across levels of the moderator (as well as a by a larger interaction coefficient as presented earlier). While the association between screen time and productivity strengthens as sleep duration increased, this effect was most pronounced at higher levels of sleep quality. Additionally, Johnson-Neyman analyses indicated that screen time-productivity relationship was significant across the entire observed range of sleep quality, whereas for sleep duration the relationship was only non-significant at extremely low levels of sleep. Together, these findings suggest that sleep quality may play a more robust role than sleep duration in shaping how screen time relates to productivity.

Hierarchical Model Comparison Strategy

To evaluate whether interaction terms meaningfully improved model fit, hierarchical regression analyses were conducted. For each outcome variable, Model 1 included only the main effects of screen time and the respective sleep variable. Model 2 added the interaction term. Incremental variance explained (△R²) was calculated as the difference in R² between the two models. This approach isolates the unique contribution of moderation beyond additive effects.

△R² Results Across Outcomes

Cross-Outcome Comparison

Comparison of incremental variance estimates revealed clear differences in practical contribution of moderation effects across outcomes. Interaction terms accounted for the greatest additional variance in stress, with sleep quality producing the largest improvement in model fit (△R² = 0.0136), followed by sleep duration (△R² = 0.0094). Although modest in magnitude, these increases indicate that moderation meaningfully enhanced explanatory power for stress beyond additive effects alone.

In contrast, △R² values for focus were effectively zero (△R² = 0.0000 for sleep duration and △R² = 0.0001 for sleep quality), indicating that interaction terms provided no meaningful improvement in prediction. This finding suggests that the relationship between screen time and focus operates largely independently of sleep-based moderation.

For productivity, interaction effects contributed small but positive increments in explained variance (△R²= 0.0017 for sleep duration and △R²  = 0.0030 for sleep quality). While these effects were stronger than those observed for focus, they remained substantially smaller than those for stress, indicating comparatively weaker moderation.

Across all outcomes, sleep quality consistently produced larger △R² values than sleep duration, reinforcing prior standardized analyses suggesting that subjective sleep experience exerts a stronger moderating influence than objective sleep duration.

Overall, hierarchical model comparisons demonstrate that moderation effects were most substantively meaningful for stress, negligible for focus and modest for productivity. These findings suggest that sleep-based buffering mechanisms may be particularly relevant for stress-related outcomes rather than cognitive or performance-related domains.

Assumption Diagnostics

To ensure the validity of regression-based moderation analyses, standard diagnostic procedures were conducted. Assumptions of linearity, homoscedasticity, normality of residuals and absence of influential outliers were evaluated using graphical diagnostics and influence statistics. Diagnostics were examined for the full moderation model predicting stress with sleep quality, as this model demonstrated the strongest interaction effect.

Linearity

Linearity was evaluated using the Residuals vs. Fitted plot Figure 29). Visual inspection did not reveal systemic curvature or nonlinear patterns in the residuals. Residuals appeared randomly dispersed around zero, supporting the assumption of linearity between predictors and the outcome.

Homoscedasticity

Homoscedasticity was assessed using both visual inspection and a formal Breusch–Pagan test.

Visual examination of the Residuals vs. Fitted plot suggested a mild funnel pattern, indicating potential non-constant variance. To formally test this assumption, a studentized Breusch–Pagan test was conducted:

# Homoscedasticity: Breusch-Pagan test 
library(lmtest)
bptest(stress_int_sleepq, studentize = TRUE)

BP = 81.879, df = 3, p-value < 2.2e-16

The test was statistically significant, BP(3) = 81.879, p < 0.001, indicating evidence of heteroscedasticity.

Importantly, heteroscedasticity does not bias regression coefficient estimates, but it may affect the accuracy of standard errors and hypothesis tests. Given the large sample size (N = 3,500), minor deviations from homoscedasticity are unlikely to meaningfully affect parameter estimates. Nevertheless, to ensure robust inference, heteroscedasticity-consistent (HC3) standard errors were computed.

library(sandwich)
library(lmtest)

# Robust SE for stress x sleep quality model
coeftest(sm2, vcov = vcovHC(sm2, type = "HC3"))

The interaction between screen time and sleep quality remained statistically significant under robust standard errors (β = -0.1172, SE = 0.0148, p < 0.001), indicating that the moderation effect was not driven by heteroscedasticity. The pattern and significance of results were substantively unchanged.

Normality of Residuals

Normality was evaluated using a Q–Q plot of standardized residuals (Figure 29). Minor deviations were observed in the tails of the distribution. However, residuals closely followed the reference line overall.

Given the large sample size (N = 3,500), slight deviations from normality are not concerning due to the Central Limit Theorem, which ensures that sampling distributions of parameter estimates approximate normality in large samples. Therefore, the assumption of normality was considered adequately met.

Influential Observations

Cook’s Distance values were examined to identify influential observations. No cases exceeded conventional thresholds (Cook’s D > 1), indicating that no individual observations exerted disproportionate influence on model estimates.

Alternative Model Specifications

To evaluate whether the observed moderation effects were dependent on specific analytic decisions, several alternative model specifications were estimated. These analyses tested whether the interaction between screen time and sleep variables remained stable across scaling approaches, model forms and co-variate adjustments.

Uncentered Predictors

Because predictors were mean-centered prior to estimating the primary moderation models, an alternative specification using uncentered variables was estimated. Mean centering does not affect interaction coefficients or overall fit, but it can alter interpretation of lower-order terms. Estimating the model without centering therefore provides a formal test that the observed moderation effect is not an artifact of variable transformation.

As shown in Table 13, the interaction between screen time and sleep quality remained statistically significant across standardized, mean-centered and uncentered specifications. The interaction coefficient was identical in the centered and uncentered models, consistent with regression theory, indicating that mean-centering did not influence the moderation effect. Although the magnitude of the coefficient differed slightly in the standardized model due to scaling, the direction, statistical significance (p < 0.001), and the overall model fit (R² = .1175) remained unchanged. Together, these findings demonstrate that the moderation role of sleep quality in the relationship between screen time and stress is robust to alternative scaling and centering decisions.

Simultaneous Moderator Models

While prior moderation analyses focused mainly on examining each lifestyle variable independently, a more stringent test of moderation involves estimating a model that includes all moderators and their respective interaction terms simultaneously. This approach evaluates whether individual moderation effects remain statistically significant when controlling for overlap among moderators.

Because sleep duration and sleep quality are moderately correlated, examining them in isolation may overestimate their unique moderating contributions. Simultaneous modeling allows for assessment of each moderator’s unique interaction effect while accounting for shared variance.

# Stress simultaneous moderation model
stress_full <- lm(
  stress_level ~
    screen_c * sleep_c +
    screen_c * sleepq_c +
    screen_c * activity_c +
    screen_c * study_c,
  data = data
)

# Focus simultaneous moderation model
focus_full <- lm(
  focus_score ~
    screen_c * sleep_c +
    screen_c * sleepq_c +
    screen_c * activity_c +
    screen_c * study_c,
  data = data
)

# Productivity simultaneous moderation model
prod_full <- lm(
  productivity_score ~
    screen_c * sleep_c +
    screen_c * sleepq_c +
    screen_c * activity_c +
    screen_c * study_c,
  data = data
)

Cross-Outcome Synthesis of Moderation Effects

Across the three simultaneous moderation models, a clear pattern emerged. The model predicting stress was statistically significant and demonstrated meaningful moderation effects, particularly for sleep hours and sleep quality. In contrast, the model predicting focus was not significant, explaining less than 1% of the variance and showing no meaningful main or interaction effects. The productivity model was statistically significant, though with a small effect size, and identified sleep quality as a significant moderator of the relationship between screen time and productivity.

Sleep quality emerged as the most consistent moderator across outcomes. It significantly interacted with screen time in both the stress and productivity models, suggesting that the effects of screen time depend in part on the quality of an individual’s sleep. Specifically, higher sleep quality appeared to buffer the association between screen time and stress, while enhancing the positive association between screen time and productivity.

Notably, focus did not appear to be meaningfully influenced by screen time or the proposed moderators. This divergence suggests that screen time may exert stronger effects on emotional outcomes (stress) and behavioral performance outcomes (productivity) than on perceived cognitive focus.

Overall, these findings indicate that lifestyle balance, particularly sleep quality, plays a conditional role in shaping how screen time relates to well-being and performance outcomes. Together, these findings suggest that the impact of screen time is not uniformly detrimental or beneficial, but instead depends on broader lifestyle context.

Summary of Robustness Findings

Across multiple analytic approaches, the moderation effects demonstrated strong consistency and stability. Standardized models confirmed that interaction magnitudes were largest for stress, modest for productivity and negligible for focus. Hierarchical model comparisons further demonstrated that interaction terms meaningfully improved model fit primarily for stress, with smaller incremental contributions for productivity and virtually none for focus.

Diagnostic testing indicated no major violations of regression assumptions. Although mild heteroscedasticity was detected in the stress model, interaction effects remained statistically significant under heteroscedasticity-consistent (HC3) standard errors, confirming that results were not driven by non-constant variance. Residual distributions were approximately normal, and no influential outliers materially affected parameter estimates.

Importantly, alternative model specifications, including standardized, mean-centered and uncentered predictors, produced substantively identical conclusions. The interaction between screen time and sleep quality remained statistically significant across scaling approaches, and overall model fit was unchanged. These findings indicate that moderation effects were not artifact of variable transformation or analytic decision-making.

Taken together, the extended statistical analyses provide converging evidence that sleep quality plays a robust moderating role in the relationship between screen time and stress, and to a lesser extent productivity. In contrast no meaningful moderation effects were observed for focus. The consistency of findings across scaling methods, hierarchical comparisons, diagnostic checks and simultaneous moderator models strengthens confidence in the validity and interpretability of the observed effects.

Overall, the results suggest that the psychological impact of screen exposure is conditional rather than uniform, with subjective sleep experience serving as a key contextual factor shaping stress-related and performance-related outcomes.