Measuring the mediating role of quality education for ensuring employability skills: An analysis of higher education student perception in Bangladesh

guarantees adequate convergent validity prior to the analysis of data. The construct measure- ment’s ability to represent the construct is called convergent validity [44]. Positive intercorrela- tion is anticipated among all indicators comprising a similar construct, given that they all reflect the same construct. [43] proposed that each construct attain an average variance extrac- tion (AVE) of at least 50% to ensure adequate convergent validity. Table 2 describes the vari- ables that yield sufficient AVE values, such as Employability Skills of Students (0.507), Quality Education (0.500), Course Structure (0.609), Institutional policy (0.585), and Physical Aspects (0.500). Hence, the present investigation satisfies the criteria for convergent validity. Discriminant validity. Discriminant validity guarantees that the constructs within com- parable frameworks are distinct from one another [43]. To ensure adequate discriminant validity, AVE’s square root must exceed the latent variable’s correlation [59], which are shown in Table 3. Structural model Direct effect. Determination of structural model coefficients was accomplished via regres- sion equations. When analysing structural relationships, the variance inflation factor (VIF) is commonly employed to ascertain whether an error is attributable to collinearity [60]. There are differing opinions among researchers regarding the optimal VIF threshold value. While [43] suggest a threshold of 5.0 [51], argue that a lower value, no more than 3, is preferable. The

VIF value for each construct was equal to or less than three, indicating the absence of collinear- ity issues. The statistical significance of path coefficients was assessed through the bootstrap- ping method, with a minimal resampling of 5,000 [43]. The correlations between exogenous and endogenous components were computed using a significance level 0.05 (p<0.05). Thus, these results are shown in Table 4. The direct effect of the relationship between CS and QE has not been supported in H1 (β = 0.057, T Value = 0.647, P Value = 0.518, F2 = 0.004, VIF = 2.427). The relationship between PA and QE is supported in H3 (β = 0.389, T Value = 4.736, P Value = 0.000, F2 = 0.208, Table 3. Discriminant validity analysis using Fornell and Lacker criterion* (Source: Authors). Constructs CS ES PA Pol QE CS 0.780 ES 0.546 0.712 PA 0.626 0.599 0.707 Pol 0.523 0.529 0.531 0.765 QE 0.684 0.62 0.69 0.519 0.733 * Note: The square root of AVE is denoted by the bold values on the diagonal, whereas the other values represent the correlations between constructs. To satisfy the discriminant validity criteria, the diagonal values must be greater than the off-diagonal values. https://doi.org/10.1371/journal.pone.0310815.t003 Table 4. Result of direct effect (path coefficient and hypothesis testing) (Source: Authors). H Path Std. Beta t-value p-value Decision VIF F2 H1 CS -> QE 0.057 0.647 0.518 Not Supported 2.427 0.004 H2 Pol -> QE 0.092 1.195 0.232 Not Supported 1.538 0.016 H3 PA -> QE 0.389 4.736 0 Supported 2.075 0.208

H4 QE -> ES 0.612 9.557 0 Supported 1 0.597 https://doi.org/10.1371/journal.pone.0310815.t004 PLOS ONE Measuring the mediating role of quality education for ensuring employability skills PLOS ONE | https://doi.org/10.1371/journal.pone.0310815 October 7, 2024 10 / 18