See real interpretations, not screenshots

The variables were measured on a 5-point Likert scale ranging from 1 to 5. Descriptive statistics were computed for all four study variables. IQ_Mean obtained the highest mean score (M = 3.63, SD = 0.70), followed by LS_Mean (M = 3.57, SD = 0.73), TU_Mean (M = 3.48, SD = 0.74), and PI_Mean (M = 3.23, SD = 0.76), which recorded the lowest mean. Applying the interpretation thresholds for a 5-point scale (Low: 1.00–2.33; Moderate: 2.34–3.66; High: 3.67–5.00), all four variables fell within the moderate range.

Regarding the normality of the data, skewness values ranged from -0.305 (LS_Mean) to 0.027 (PI_Mean), and kurtosis values ranged from -0.615 (PI_Mean) to -0.009 (IQ_Mean). All skewness values were within the acceptable threshold of |Skew| < 2.0 (Gravetter & Wallnau, 2014), and all kurtosis values were within the acceptable threshold of |Kurt| < 7.0 (Byrne, 2010). These results indicated that the distributions of all variables approximated normality and were suitable for subsequent parametric analyses.

Table 1: Descriptive Statistics

Note. Variables are ordered from highest to lowest mean. Scale range: 1–5 (Low: 1.00–2.33; Moderate: 2.34–3.66; High: 3.67–5.00). All skewness and kurtosis values fall within acceptable thresholds (|Skew| < 2.0, Gravetter & Wallnau, 2014; |Kurt| < 7.0, Byrne, 2010), indicating approximate normality.

References

Byrne, B. M. (2010). Structural equation modeling with AMOS: Basic concepts, applications, and programming (2nd ed.). Routledge.

Gravetter, F. J., & Wallnau, L. B. (2014). Essentials of statistics for the behavioral sciences (8th ed.). Wadsworth.

A reliability analysis was conducted to assess the internal consistency of all scales used in the study. Using the guidelines proposed by George and Mallery (2003), Cronbach's alpha coefficients were interpreted as follows: ≥ .90 = excellent, .80–.89 = good, .70–.79 = acceptable, .60–.69 = questionable, .50–.59 = poor, and < .50 = unacceptable. The LS scale (5 items) yielded a Cronbach's alpha of α = .868, indicating good internal consistency. The IQ scale (5 items) demonstrated good reliability (α = .860), as did the TU scale (5 items; α = .871) and the PI scale (4 items; α = .820). Taken together, all four scales demonstrated good internal consistency, supporting their suitability for use in subsequent analyses.

Item-total correlation analyses were conducted to further evaluate scale quality. Following the criterion recommended by Field (2018), corrected item-total correlations should exceed .30 to indicate that an item is measuring the same underlying construct as the scale. For the LS scale, corrected item-total correlations ranged from .662 to .721, and removing any item would not have improved scale reliability. For the IQ scale, corrected item-total correlations ranged from .664 to .698, with no item deletion improving alpha. For the TU scale, corrected item-total correlations ranged from .680 to .723, and no item deletion would have resulted in a higher alpha coefficient. For the PI scale, corrected item-total correlations ranged from .632 to .647, and no item deletion would have improved reliability. All items across all four scales demonstrated acceptable item-total correlations well above the .30 threshold (Field, 2018), and the alpha-if-item-deleted values confirmed that each item was contributing positively to its respective scale. These findings collectively indicate that all scales possessed satisfactory internal consistency for research purposes.

Table 1: Item-Total Statistics

Note. Items with corrected item-total correlations < .30 may not be measuring the same construct (Field, 2018). All items in the present study exceeded this threshold.

References

Field, A. (2018). Discovering statistics using IBM SPSS Statistics (5th ed.). SAGE Publications.

George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference, 11.0 update (4th ed.). Allyn & Bacon.

Prior to conducting the primary analyses, a series of assumption tests were performed at a significance level of α = .05. Normality was assessed for all four variables using both the Shapiro-Wilk and Kolmogorov-Smirnov tests. Both tests indicated statistically significant departures from normality for all variables: LS_Mean, IQ_Mean, TU_Mean, and PI_Mean all returned p < .05 on both tests (see Table 1). However, inspection of skewness and kurtosis values revealed that all variables fell within the acceptable range of ±1, and visual examination of Q-Q plots suggested approximate normality in the distributions. Furthermore, given the large sample size (N = 400), the Shapiro-Wilk and Kolmogorov-Smirnov tests are known to be highly sensitive to even trivial deviations from normality. In accordance with the Central Limit Theorem, parametric tests remained appropriate and robust to these violations (Pallant, 2020; Field, 2018). The assumption of homogeneity of variance was assessed using Levene's test based on the mean. The result was non-significant, F(3, 396) = 0.623, p = .601, indicating that equal variances across Year groups for LS_Mean could be assumed. This assumption was met.

For the multiple regression model, three additional diagnostic assumptions were examined. Independence of residuals was assessed using the Durbin-Watson statistic, which yielded a value of 1.979, falling within the acceptable range of 1.5–2.5 and indicating no problematic autocorrelation among residuals. Multicollinearity was evaluated using Tolerance and Variance Inflation Factor (VIF) values for each predictor (see Table 2). All VIF values were well below the threshold of 5, and all Tolerance values were well above 0.10, indicating no multicollinearity concerns among the four predictors in the model. Outliers were assessed through inspection of standardized residuals. The maximum absolute standardized residual observed was 3.024, which is marginally beyond the ±3.0 guideline but remains within the more conservative ±3.29 threshold for extreme outliers (Field, 2018). Cook's Distance and leverage values were not available in the output; however, based on the standardized residual inspection, no severe outlier concern was identified. Overall, the assumptions for parametric testing and multiple regression were considered sufficiently met, and the planned analyses proceeded accordingly.

Table 1: Normality Test Results

Note. S-W = Shapiro-Wilk; K-S = Kolmogorov-Smirnov. df = 400 for all variables. All variables showed statistically significant departures from normality; however, parametric tests remained appropriate due to the large sample size (N = 400) and the Central Limit Theorem (Pallant, 2020; Field, 2018). Skewness and kurtosis values for all variables fell within ±1, and Q-Q plot inspection suggested approximate normality.

Table 2: Collinearity Statistics

Note. VIF values < 5 indicate no multicollinearity concern (Hair et al., 2019). Tolerance values > .10 further confirm the absence of multicollinearity among predictors.

References

Field, A. (2018). Discovering statistics using IBM SPSS Statistics (5th ed.). SAGE Publications.

Hair, J. F., Babin, B. J., Anderson, R. E., & Black, W. C. (2019). Multivariate data analysis (8th ed.). Cengage Learning.

Pallant, J. (2020). SPSS survival manual: A step-by-step guide to data analysis using IBM SPSS (7th ed.). McGraw-Hill Education.

H1

An independent samples t-test was conducted to examine whether LS_Mean differed between Gender group 1 and Gender group 2 (Gender group 3, n = 2, was excluded per syntax specification; analysis N = 398). Levene's test for equality of variances was non-significant, F(1, 396) = 0.31, p = .578, so equal variances were assumed. The test revealed a statistically significant difference, t(396) = −3.703, p < .001, 95% CI [−0.41, −0.13], with Gender group 1 (M = 3.42, SD = 0.74) scoring lower than Gender group 2 (M = 3.69, SD = 0.71). Cohen's d = 0.37, indicating a small effect size (Cohen, 1988). The hypothesis was supported.

Table 1: Group Statistics and Independent Samples t-Test Results for LS_Mean by Gender

Note. Equal variances assumed; Levene's F(1, 396) = 0.31, p = .578. Gender group 3 (n = 2) was excluded from the analysis. The 95% CI for the mean difference was [−0.41, −0.13]. Cohen's d reflects a small effect size (Cohen, 1988).

H2

A paired samples t-test was conducted to examine whether IQ_Mean and PI_Mean differed within the same participants (N = 400). This comparison was a within-subject contrast of two constructs in a cross-sectional dataset, not a pre/post design. IQ_Mean (M = 3.63, SD = 0.70) was significantly higher than PI_Mean (M = 3.23, SD = 0.76), t(399) = 8.670, p < .001, 95% CI [0.32, 0.50]. The paired mean difference was 0.41 (SD = 0.94, SE = 0.05), and the within-pair correlation was r = .177, p < .001. Cohen's dz = 0.43, indicating a small effect size (Cohen, 1988). The hypothesis was supported.

Table 2: Paired Samples t-Test Results for IQ_Mean and PI_Mean

Note. N = 400. The 95% CI for the mean difference was [0.32, 0.50]. Paired correlation r = .177, p < .001. Cohen's dz reflects a small effect size (Cohen, 1988).

H3

A one-way ANOVA was conducted to examine whether LS_Mean differed across Year of study groups (N = 400). Levene's test was non-significant, F(3, 396) = 0.623, p = .601, confirming homogeneity of variance. The ANOVA revealed a statistically significant overall effect, F(3, 396) = 3.006, p = .030, η² = .022, indicating a small effect size (Cohen, 1988). Group means ranged from 3.39 (Year 1) to 3.67 (Year 2), as shown in Table 3. Post-hoc comparisons using Tukey HSD (Table 4) indicated that only the difference between Year 1 (M = 3.39, SD = 0.69) and Year 2 (M = 3.67, SD = 0.72) was statistically significant (mean difference = −0.27, SE = 0.10, p = .046). All remaining pairwise comparisons were non-significant, with all p ≥ .077. The hypothesis was supported.

Table 3: Descriptive Statistics for LS_Mean by Year of Study

Note. N = 400. One-way ANOVA: F(3, 396) = 3.006, p = .030, η² = .022.

Table 4: Tukey HSD Post-Hoc Comparisons for LS_Mean by Year of Study

Note. Tukey HSD procedure. Only the Year 1 vs. Year 2 comparison reached statistical significance at α = .05.

H4

Pearson correlation analysis was conducted to examine associations among LS_Mean, IQ_Mean, TU_Mean, PI_Mean, and StudyHours (N = 400). The full correlation matrix is presented in Table 5. LS_Mean showed a significant positive correlation with IQ_Mean (r = .479, p < .001), TU_Mean (r = .390, p < .001), and PI_Mean (r = .243, p < .001). The correlation between LS_Mean and IQ_Mean represents a medium effect size, the correlation between LS_Mean and TU_Mean represents a medium effect size, and the correlation between LS_Mean and PI_Mean represents a small effect size (Cohen, 1988). IQ_Mean was also significantly positively correlated with TU_Mean (r = .260, p < .001) and PI_Mean (r = .177, p < .001), both representing small effect sizes (Cohen, 1988). The correlation between TU_Mean and PI_Mean was not significant (r = .066, p = .185). StudyHours showed no significant correlation with any of the other variables: LS_Mean (r = .044, p = .375), IQ_Mean (r = −.010, p = .835), TU_Mean (r = −.005, p = .922), or PI_Mean (r = .003, p = .957); all correlations involving StudyHours were negligible in magnitude (Cohen, 1988). Because TU_Mean and PI_Mean were not significantly correlated with each other, and StudyHours was not significantly correlated with any variable, the hypothesis is partially supported: significant positive correlations were found among LS_Mean, IQ_Mean, TU_Mean, and PI_Mean, but TU_Mean and PI_Mean did not correlate significantly with each other, and StudyHours was unrelated to all other variables.

Table 5: Pearson Correlation Matrix for LS_Mean, IQ_Mean, TU_Mean, PI_Mean, and StudyHours

Note. N = 400. Only the lower triangle is reported; the upper triangle contains the mirrored values. *** p < .001.

H5

Simple linear regression was conducted to examine whether IQ_Mean predicted LS_Mean (N = 400). The overall model was statistically significant, F(1, 398) = 118.343, p < .001, R = .479, R² = .229, adjusted R² = .227, indicating that IQ_Mean accounted for 22.9% of the variance in LS_Mean. The effect size f² = 0.297 represents a medium effect (Cohen, 1988). IQ_Mean was a significant positive predictor of LS_Mean (B = 0.50, SE = 0.05, β = .479, t(398) = 10.879, p < .001); a one-unit increase in IQ_Mean was associated with a 0.50-unit increase in LS_Mean. The hypothesis was supported.

Table 6: Simple Linear Regression Coefficients for the Prediction of LS_Mean by IQ_Mean

Note. N = 400. R = .479, R² = .229, adjusted R² = .227, F(1, 398) = 118.343, p < .001. f² = 0.297, indicating a medium effect size (Cohen, 1988).

H6

Multiple linear regression was conducted to examine whether IQ_Mean, TU_Mean, PI_Mean, and StudyHours jointly predicted LS_Mean (N = 400). The overall model was statistically significant, F(4, 395) = 48.878, p < .001, R = .575, R² = .331, adjusted R² = .324, accounting for 33.1% of the variance in LS_Mean. The effect size f² = 0.495 represents a large effect (Cohen, 1988). Regression coefficients are presented in Table 7. Three predictors were statistically significant: IQ_Mean (B = 0.39, SE = 0.05, β = .378, t(395) = 8.752, p < .001), TU_Mean (B = 0.28, SE = 0.04, β = .281, t(395) = 6.594, p < .001), and PI_Mean (B = 0.15, SE = 0.04, β = .157, t(395) = 3.762, p < .001). StudyHours was not a significant predictor (B = 0.01, SE = 0.01, β = .049, t(395) = 1.199, p = .231). Comparing standardized coefficients, IQ_Mean was the strongest predictor (|β| = .378), followed by TU_Mean (|β| = .281) and PI_Mean (|β| = .157). A one-unit increase in IQ_Mean was associated with a 0.39-unit increase in LS_Mean, a one-unit increase in TU_Mean with a 0.28-unit increase, and a one-unit increase in PI_Mean with a 0.15-unit increase, holding all other predictors constant. Multicollinearity diagnostics indicated no concerns: the largest VIF was 1.103 and the smallest Tolerance was .907. The Durbin-Watson statistic was 1.979, indicating no evidence of autocorrelation in the residuals. Because StudyHours did not reach significance, the hypothesis is partially supported.

Table 7: Multiple Linear Regression Coefficients for the Prediction of LS_Mean

Note. N = 400. R = .575, R² = .331, adjusted R² = .324, F(4, 395) = 48.878, p < .001. f² = 0.495, indicating a large effect size (Cohen, 1988). Durbin-Watson = 1.979.

H7

A chi-square test of independence was conducted to examine whether an association existed between Gender and DeviceType (N = 400). The test revealed a statistically significant association, χ²(6, N = 400) = 25.854, p < .001. Cramér's V = .180 (df_min = 2, based on the smaller of rows − 1 and columns − 1), indicating a small effect size (Cohen, 1988). The hypothesis was supported.

Table 8: Chi-Square Test of Independence for Gender and DeviceType

Note. Cramér's V = .180 with df_min = 2, indicating a small effect size (Cohen, 1988).

⚠️ Diagnostic Note — H7: Four cells (33.3%) had expected counts below 5, with a minimum expected count of 0.11. This is a severe violation of the chi-square assumption that all expected cell frequencies should be ≥ 5. The violation is driven primarily by Gender group 3 (n = 2), which is an extremely small subgroup. Results should be interpreted with explicit caution. Consider: (a) excluding Gender group 3 and re-running the analysis on the remaining two gender groups, (b) combining sparse DeviceType categories to reduce the table dimensions, or (c) applying Fisher's Exact Test if the table can be reduced to 2 × 2.

H8

PROCESS Model 4 mediation analysis (Hayes, 2022) was conducted to examine whether TU_Mean mediated the relationship between IQ_Mean and LS_Mean (N = 400), using 5,000 bootstrap samples and 95% bias-corrected confidence intervals. Path coefficients and effect estimates are presented in Table 9. The a-path (IQ_Mean → TU_Mean) was significant (B = 0.27, SE = 0.05, t = 5.377, p < .001, 95% CI [0.17, 0.38]), indicating that higher IQ_Mean was associated with higher TU_Mean. The b-path (TU_Mean → LS_Mean, controlling for IQ_Mean) was also significant (B = 0.28, SE = 0.04, t = 6.562, p < .001, 95% CI [0.20, 0.36]). The indirect effect of IQ_Mean on LS_Mean through TU_Mean was significant (B = 0.08, BootSE = 0.02, 95% Boot CI [0.04, 0.11]), as the confidence interval excluded zero; the completely standardized indirect effect was .074. The direct effect of IQ_Mean on LS_Mean remained significant after controlling for TU_Mean (B = 0.42, SE = 0.05, t = 9.337, p < .001, 95% CI [0.33, 0.51]), and the total effect was also significant (B = 0.50, SE = 0.05, t = 10.879, p < .001, 95% CI [0.41, 0.59]). Because both the indirect effect and the direct effect were significant, the results indicate partial mediation of the IQ_Mean → LS_Mean relationship by TU_Mean. The hypothesis was supported.

Table 9: Direct, Indirect, and Total Effects for the Mediation of LS_Mean by TU_Mean (IQ_Mean as Predictor)

Note. N = 400. Bootstrap samples = 5,000; 95% bias-corrected confidence intervals. SE for the indirect effect is BootSE. The 95% Boot CI for the indirect effect excludes zero, indicating a significant indirect effect. Completely standardized indirect effect = .074. Partial mediation was concluded because both the direct and indirect effects were significant (Hayes, 2022).

H9

PROCESS Model 1 moderation analysis (Hayes, 2022) was conducted to examine whether StudyHrs moderated the relationship between IQ_Mean and LS_Mean (N = 400), with variables mean-centered prior to analysis. The interaction term (IQ_Mean × StudyHrs) was not statistically significant, B = 0.00, SE = 0.01, t = 0.020, p = .984, 95% CI [−0.03, 0.03], ΔR² = .000, F(1, 396) = 0.000, p = .984. The interaction accounted for no additional variance in LS_Mean beyond the main effects, indicating that StudyHrs did not moderate the IQ_Mean → LS_Mean relationship. The hypothesis was not supported.

References

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates.

Hayes, A. F. (2022). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach (3rd ed.). The Guilford Press.