*3.2. The Assessment of the Measurement Model*

The measurement model constructs were valued via assessment of reliability, then via convergent as well as discriminant validity. Second-order external factors (implementing the repeated indicators technique—the hierarchical component model proposed by Wold [51]) were formed. Some external first-order factors did not meet the evaluation requirements—therefore, the following four factors were excluded from further analysis: the factor computer self-efficacy and the factor experience with computers were excluded from the PCIL second-order factors, and the factor ERP training and the factor organizational culture were excluded from the OPC second-order factors. Results of reliability and convergent validity for constructs are presented in Table 3. Results of discriminant validity are presented in Table 4 and hypotheses testing results are in Table 5. Table 6 brings the results of second-order factors. All results are discussed as follows.

Cronbach's *α* and *CR* measures were calculated [37,38,45]. As shown in Table 3, for each of the 14 scales used in the study Cronbach's *α* and *CR* were higher than threshold, which is 0.7, thus confirming their reliability [37,38,45].


**Table 3.** Psychometric properties of the research instrument (sample size = 208).

Note: *α* = Cronbach's *α*; *R*<sup>2</sup> = explanatory power or variance; Adj. *R*<sup>2</sup> = adjusted *R*2.

Convergent validity was examined via Fornell and Larcker's criteria: all item factor loadings should be significant and higher than threshold, which is 0.70, and *AVE* for each construct, where threshold is 0.50 [43]. All item factor loadings in our study fulfilled these criteria except for one indicator that equalled 0.69, which is also above the minimal satisfactory level (0.50) suggested by Fornell and Larcker [43]. *AVE* values were above 0.50. The Cronbach's *α*, *CR* as well as *AVE* of the second-order model are presented in Table 6. Cronbach's *α* were above 0.70, *CR*s were above 0.80, and *AVE*s were equal to or above 0.50. As shown in Figure 4 and Table 6, the loadings of the first-order factors on the second-order factors went above 0.70, with the exception of two indicators, which were 0.64 and 0.68, thus still fulfilling the criteria for the minimal satisfactory level. Measurement scales show strong convergent validity.

The discriminant validity was analyzed using the criteria described in Section 2.1. Details of this estimation are presented in Table 4. All measurement loadings were above 0.70. This represents that the reflective model fits well [37,38,45] and cross-loadings were lower (data and results available by request). Additionally, all HTMT variables were less than 1.0 (italic numbers in Table 4). All three criteria of discriminant validity are fulfilled.

**Table 4.** Results of discriminant validity (intercorrelation of the latent variables and HTMT variables (italic).


Note: Square root of *AVE* in bold text; HTMT values in italic text.

Garson [38] pointed out that "standardized root means square residual (SRMS) calculates the difference among the model-implied correlation matrix as well as the observed correlation matrix" and added that the model is well-fitted if SRMS is lower than 0.08. However, some researchers use the more lenient cut-off of 0.10. The SRMS value of the research model in this paper stands at 0.09, and presents that model as allowable.

All criteria of the measurement model were met, so we were able to continue our analysis with the structural model analysis.

#### *3.3. Structural Model*

The hypotheses listed above were tested. As already mentioned, bootstrapping (5000 sub-samples) was used to examine the statistical significance of each path coefficient by performing *t* tests [48].

Table 5 and Figure 4 include the results obtained. Construct PEOU had no significant impact neither on construct PU *(t* = 0.90, *p* > 0.05) nor on construct AT (*t* = 0.98, *p* > 0.05). Further, construct PU had a weak positive impact on construct AT (*t* = 2.79, *p* < 0.05). Construct WC had a strong and significant positive impact on construct AT (*t* = 4.81, *p* < 0.01), construct PU (*t* = 6.38, *p* < 0.01), and construct ExU (*t* = 6.56, *p* < 0.01). Additionally, construct AT had a weak but important impact on construct ExU (*t* = 2.20, *p* < 0.05).

**Table 5.** Hypothesized relationships (for all TAM constructs and significant relationships of extended TAM constructs).


Note: Path significance: \*\* *p* < 0.01; \* *p* < 0.05; n.s. = not significant. *f* <sup>2</sup> thresholds: a > 0.02 (weak effect); b > 0.15 (moderate effect); c > 0.35 (strong effect).

As shown in Figure 4 and Table 6, the second-order factors significantly positively impacted construct WC, construct PU as well as construct PEOU. Second-order factor PCIL shows significantly positive but weak impact on construct WC (*t* = 2.18, *p* < 0.05). Second-order factor STC had a weak impact on construct PU (*t* = 2.79, *p* < 0.01), a moderate impact on construct WC (*t* = 6.82, *p* < 0.01) and a very strong positive impact on construct PEOU (*t* = 19.55, *p* < 0.01). Additionally, the second-order factor OPC had a weak positive impact on construct WC (*t* = 3.50, *p* < 0.01). In addition to this, other relationships between second-order factors OPC, STC, and PCIL on one side and constructs of original TAM (namely PEOU, PU and WC, on the other side were tested, but none of the relationships were significant.

**Table 6.** Path coefficients—external constructs in the second-order model.


Note: *t* values are in brackets; all values are signed at *p* < 0.01.

The variance explained for each dependent variable is indicated by the *R*<sup>2</sup> generated for each regression equation. The structural model gives a demonstration of predictive power since *R*<sup>2</sup> and adjusted *R*<sup>2</sup> for key dependent variables are very high. *R*<sup>2</sup> is 0.59 for construct WC, 0.67 for construct PU, 0.61 for construct PEOU, 0.67 for construct AT, and 0.37 for construct ExU, as presented in Table 3. All adjusted *R*<sup>2</sup> are "moderate" according to Chin [48], except AT and PU, which are "substantial". The research discovered that the research model used explains, on average, a high proportion of the variance since the average *R*<sup>2</sup> equals 0.58. The model average *f* <sup>2</sup> (as defined in Section 2.1), which equals 0.24, reflects the moderate effect size-independent factors have on dependent factors (Table 5). The highest effect was identified for second-order factor STC on construct PEOU (*f* <sup>2</sup> = 1.59), and it contributes the most to the research model average effect size.

Moderating effects for the factors of the extended TAM were explored, and the results are presented in Table 7:

	- - The indirect effect of AT (AT; WC→AT→ExU) is significant (*β* = 0.081, *t* = 1.993, *p* = 0.046, [0.008; 0.171]);
	- - The indirect effects of PU and of AT (WC→PU→AT→ExU) do not meet the significance threshold (*β* = 0.019, *t* = 1.509, *p* = 0.131, [0.001; 0.053]);

For second-order factors, total effects on construct ExU were calculated—these effects exist and are important. Each group of second-order factors significantly impacts the construct ExU:


Further analysis shows that for all three second-order factors, only the indirect effect through construct WC is significant (OPC→WC→ExU: *β* = 0.108, *t* = 3.099, *p* = 0.002; PCIL→WC→ExU: *β* = 0.046, *t* = 2.085, *p* = 0.037; STC→WC→ExU: *β* = 0.210, *t* = 4.588, *p* = 0.000), while indirect effects through construct PU and construct PEOU are not statistically significant.

**Table 7.** Direct and indirect effect of extended TAM model.


**Figure 4.** Findings of structural model analysis. Note: Path sign.: \*\* *p* < 0.01; \* *p* < 0.05; n.s. = not significant (dotted arrows).
