Modeling technique

To calculate the properties of the predictive model, a structural analysis of equations was conducted using the Partial Least Squares Structural Equation Modeling (SEM-PLS) method in the R programming language, utilizing the “SEMinR” package (R Core Team, 2022; Ray et al., 2022). The Partial Least Squares technique is a statistical method with the primary goal of maximizing the explanation of latent variables variability through a set of latent predictors. PLS method is a structural equation modeling technique that employs an algorithm encompassing two phases.

In the first phase, a measurement model is created, expressed through latent variables. In the measurement model (current), latent variables are represented as reflective variables (in accordance with the differentiation between reflective and formative measurements) (Sarstedt et al., 2016). In this measurements computation phase, a confirmatory factor analysis is conducted, with observable variables (e.g., questionnaire items) as input data. The result of this analysis is the structure of the measurement model with a series of latent variables, preparing the variables for the construction of a multidimensional path analysis in the subsequent stage. The criteria for evaluating the measurement model is fit of gathered data with the theoretical measurement model formulated by the researcher, internal validity (the reliability and variability of measuring variables), and discriminant validity (meaningful differences in information measurement of analyzed latent variables).

In the second stage of SEM-PLS analysis, iterative series of multidimensional regression models are calculated in aim to explain the variability of dependent variables. This stage serves as the verification of the researcher’s theoretical model of directional relationships using an empirical model which is based on data. The empirical model is a path structure where relationships between variables are expressed through the researcher’s theory logic. The empirical model contains information about the significance, intensity, and direction of the influence of individual predictors on dependent variables. The key criterion for evaluating the path model is a predictive ability of the tested variables in the formulated model.

In the subsequent analysis, a consistent PLS-SEM algorithm was employed, correcting correlations between reflective variables to ensure results align with the factor model (Dijkstra et al., 2015). In the conducted SEM-PLS analysis, standard errors and statistical significance were estimated using a bootstrap procedure. Bootstrap is a sampling method where multiple samples of the same size are repeatedly drawn with replacement from an original sample, meaning the same observation can be selected multiple times. In the path model, 500 bootstrap sampling steps were applied, and the random seed was set to the value 123456789. Linear relationships were predicted in the path model.

Properties of measurement model

Internal validity of measurement model

The analysis of Table 1 presents reliability coefficients (Cronbach’s Alpha, rhoC, and rhoA) and an internal consistency measure represented by the average variance extracted (AVE). These results indicate that all measurements were reliable (Cronbach, 1951) and internally valid, understood as explaining observable indicators (questionnaire items) after creating latent variables (Dijkstra et al., 2015). Cronbach’s Alpha, rhoC, rhoA for each measurement pointed out that the tested latent variables exhibited low error variability and high measurement accuracy, and AVE coefficients sugessted that latent variables explained the observable indicators a little above the 0.5 established treshold. Table 2 presents the values of factor loadings of confirmatory factor analysis. These values indicate how strongly a reflective latent variable influences its observed indicators. Values above 0.50 suggest a clear impact of the reflective variable on observed indicators, but values above 0.7 are desirable (Hair et al., 2011). These assumptions were satisfied in current model. The results in Table 2 indicate that all observable variables were significantly and strongly explained by their latent variables.

In conclusion, aforementioned results confirm the good quality of the measurement model for the analyzed variables. All results express good internal measurement accuracy. Fulfilling this assumption allows trusting the path estimations in the structural model analysis results in the next step of SEM-PLS analysis. Tables 3 and 4 provide descriptive statistics and correlations between latent variables. Figure 1 illustrates these correlations graphically.

Table 1

Internal Calidity of Measurements

Measure α ρC AVE ρA
Observer 0.86 0.86 0.55 0.87
Actor 0.89 0.89 0.67 0.90
Perceived ease 0.90 0.90 0.68 0.90
Perceived usability 0.87 0.87 0.63 0.87
Intention to use 0.86 0.86 0.61 0.87

Note: ρC = Composite Reliability (desired value > 0.70); α = Cronbach’s Alpha (desired value ≥ 0.70); AVE = Average Variance Extracted (desired value > 0.50); ρA = Dijkstra’s Coherent Reliability (desired value > 0.70).

Table 2

Results of Factor Loadings Estimated During Confirmatory Factor Analysis

Measure Loading Boot loading Boot deviation t LCI HCI
Observer1 <- Observer 0.58 0.58 0.09 6.47*** 0.40 0.73
Observer2 <- Observer 0.89 0.88 0.07 12.72*** 0.73 1.01
Observer3 <- Observer 0.73 0.72 0.08 9.58*** 0.56 0.85
Observer4 <- Observer 0.73 0.73 0.07 9.9*** 0.58 0.87
Observer5 <- Observer 0.74 0.74 0.09 8.64*** 0.56 0.91
Actor1 <- Actor 0.81 0.81 0.05 16.27*** 0.71 0.90
Actor2 <- Actor 0.92 0.92 0.03 28.06*** 0.86 0.98
Actor3 <- Actor 0.76 0.76 0.05 15.99*** 0.67 0.85
Actor4 <- Actor 0.79 0.79 0.04 17.72*** 0.69 0.87
Perceived ease1 <- Perceived ease 0.82 0.82 0.06 14.47*** 0.71 0.94
Perceived ease2 <- Perceived ease 0.86 0.86 0.05 16.57*** 0.76 0.96
Perceived ease3 <- Perceived ease 0.83 0.82 0.10 8.59*** 0.62 0.99
Perceived ease4 <- Perceived ease 0.79 0.79 0.06 13.24*** 0.67 0.89
Perceived usability1 <- Perceived usability 0.82 0.83 0.03 27.26*** 0.76 0.88
Perceived usability2 <- Perceived usability 0.80 0.80 0.03 25.28*** 0.73 0.86
Perceived usability3 <- Perceived usability 0.77 0.77 0.04 19.64*** 0.69 0.84
Perceived usability4 <- Perceived usability 0.78 0.78 0.04 20.96*** 0.70 0.85
Intention to use1 <- Intention to use 0.80 0.80 0.04 17.88*** 0.70 0.87
Intention to use2 <- Intention to use 0.75 0.75 0.04 18.07*** 0.67 0.82
Intention to use3 <- Intention to use 0.76 0.76 0.04 20.97*** 0.69 0.83
Intention to use4 <- Intention to use 0.82 0.82 0.03 25.74*** 0.76 0.88

Nota: Loading = Factor Loading; Boot loading = Bootstrap-estimated Factor Loading; Boot deviation = Bootstrap Deviation from the Base Estimate; t = Student’s t-statistic; LCI and HCI = Lower and Upper 95% Confidence Intervals for Factor Loading, respectively.

*** p < 0.001

** p < 0.01

* p < 0.05

Table 3

Statystyki opisowe analizowanych zmiennych

Item id. Missings M Me Min Max SD Curtosis Skewness
Observer1 1 0 2.83 3 1 4 0.85 2.43 -0.29
Observer2 2 0 2.58 3 1 4 0.92 2.18 -0.09
Observer3 3 0 3.05 3 1 4 0.88 2.76 -0.68
Observer4 4 0 2.62 3 1 4 0.94 2.18 -0.24
Observer5 5 0 2.88 3 1 4 0.83 2.64 -0.40
Actor1 6 0 2.08 2 1 4 0.87 2.63 0.50
Actor2 7 0 2.17 2 1 4 0.87 2.59 0.43
Actor3 8 0 2.09 2 1 4 0.83 2.68 0.43
Actor4 9 0 2.24 2 1 4 0.89 2.49 0.40
Perceived ease1 10 0 3.70 4 1 5 0.90 3.69 -0.80
Perceived ease2 11 0 3.66 4 1 5 0.89 3.50 -0.62
Perceived ease3 12 0 3.49 4 1 5 0.95 3.14 -0.59
Perceived ease4 13 0 3.61 4 1 5 0.89 3.00 -0.49
Perceived usability1 14 0 3.38 3 1 5 1.03 3.04 -0.56
Perceived usability2 15 0 3.38 3 1 5 1.02 2.83 -0.41
Perceived usability3 16 0 2.97 3 1 5 1.13 2.23 -0.18
Perceived usability4 17 0 3.20 3 1 5 1.01 2.62 -0.22
Intention to use1 18 0 3.45 4 1 5 1.24 2.45 -0.66
Intention to use2 19 0 2.55 2 1 5 1.18 2.13 0.29
Intention to use3 20 0 2.58 3 1 5 1.28 1.77 0.16
Intention to use4 21 0 2.42 2 1 5 1.17 2.24 0.40

Table 4

Correlations between latent variables

Zmienna Id. 1 2 3 4
Observer 1
Actor 2 -0.39***
Perceived ease 3 -0.29*** 0.27***
Perceived usability 4 -0.47*** 0.61*** 0.49***
Intention to use 5 -0.43*** 0.66*** 0.36*** 0.72***

Nota:

*** p < 0.001

** p < 0.01

* p < 0.05

Figure 1
Visualisation of correlations between latent variables Observer, Actor, Perceived ease, Perceived usability, Intention to use

Note: The darker the color green = More positive correlation; The darker the color red = More negative correlation. The figure is based on the obtained estimates of the correlation coefficients Pearsona.

Discriminant validity of measurement model

The assumption of doiscriminant validity is an assumption concerning the excess of information in the measurement model. To assess this phenomenon in the tested model, the Heterotrait-Monotrait (HTMT) coefficient, as well as the method of factor loadings and cross-loadings, were applied. The HTMT coefficient evaluates how different two measurements are similar (measure similar information) in terms of correlations between the observable variables of the two latent constructs. Therefore, HTMT is a measure of similarity between latent variables. If the HTMT value is less than one, it can be considered that the differential validity is confirmed. In many practical situations, a threshold of 0.85 effectively distinguishes pairs of latent variables that are differentially valid from those that are not (Henseler et al., 2014). Differential validity in the method of cross-loadings is evident when each observable variable correlates weakly with all other latent constructs except the one it is theoretically related to. It is assumed that the difference in factor loadings for a given test item between latent factors should be around 0.2 of the factor loading value.

The results presented in Table 5 show that all HTMT values are below the threshold of 0.85. Similar conclusions can be drawn from Table 6, where patterns of items loadings with their factors and cross-loadings with rest of factors indicate that each observable variable correlates weakly with other factors except the one it is supposed to be associated with due to the original assumption of the measurement model. Additionally, the assumption of discriminant validity of the Fornell-Larcker criterion (Fornell & Larcker, 1981) was tested, which indicates that if the square root of the AVE (Average Variance Extracted) of each construct is higher than the highest correlation of the construct with any other construct in the model, the assumption of discriminant validity has been met. Table 7 indicates that these defined conditions have been met in all variables condition. All the above assessment methods indicate that the discriminant validity of measurements has been confirmed. Different variables in the model measured different information about the studied phenomena. This also means that the error of similar methods (common method bias), indicating that the observed results arise from the construction of the research method rather than the nature of the phenomenon studied, does not pose a significant problem in the currently tested measurements and has slight affect on structural model estimations (Podsakoff et al., 2003).

Table 5

Discriminant validity of the tested latent variables measured by HTMT coefficients

Pair of constructs HTMT Boot deviation Bootstrap SD LCI HCI
Observer -> Actor 0.44 0.44 0.06 0.32 0.54
Observer -> Perceived ease 0.32 0.33 0.07 0.19 0.46
Observer -> Perceived usability 0.53 0.53 0.07 0.41 0.66
Observer -> Intention to use 0.50 0.50 0.06 0.37 0.62
Actor -> Perceived ease 0.30 0.29 0.06 0.18 0.40
Actor -> Perceived usability 0.69 0.69 0.04 0.60 0.77
Actor -> Intention to use 0.75 0.75 0.03 0.68 0.82
Perceived ease -> Perceived usability 0.55 0.55 0.06 0.44 0.66
Perceived ease -> Intention to use 0.41 0.41 0.06 0.29 0.53
Perceived usability -> Intention to use 0.83 0.83 0.03 0.76 0.89

Note: HTMT = Heterotrait-Monotrait Differential Validity Coefficient (desired value HTMT < 0.85); Booted HTMT = HTMT Differential Validity Coefficient estimated during the bootstrap procedure; Boot deviation = Deviation of the bootstrap estimate from the base estimate; LCI and HCI = Lower and Upper 95% Confidence Intervals for the HTMT Differential Validity Coefficient.

Table 6

Discriminant validity of the tested latent variables expressed by cross-loadings method

Item Observer Actor Perceived ease Perceived usability Intention to use
Observer1 0.75 -0.21 -0.19 -0.29 -0.28
Observer2 0.84 -0.30 -0.33 -0.41 -0.34
Observer3 0.73 -0.42 -0.19 -0.39 -0.39
Observer4 0.87 -0.29 -0.23 -0.37 -0.35
Observer5 0.80 -0.31 -0.20 -0.39 -0.37
Actor1 -0.36 0.88 0.18 0.54 0.57
Actor2 -0.36 0.91 0.29 0.58 0.63
Actor3 -0.29 0.82 0.21 0.49 0.52
Actor4 -0.33 0.86 0.24 0.50 0.57
Perceived ease1 -0.27 0.21 0.87 0.40 0.31
Perceived ease2 -0.29 0.21 0.90 0.45 0.33
Perceived ease3 -0.19 0.30 0.82 0.42 0.32
Perceived ease4 -0.26 0.22 0.89 0.43 0.30
Perceived usability1 -0.45 0.51 0.46 0.87 0.64
Perceived usability2 -0.42 0.56 0.46 0.86 0.59
Perceived usability3 -0.31 0.52 0.31 0.83 0.61
Perceived usability4 -0.40 0.48 0.43 0.84 0.61
Intention to use1 -0.37 0.54 0.34 0.62 0.80
Intention to use2 -0.32 0.50 0.28 0.58 0.85
Intention to use3 -0.32 0.55 0.31 0.59 0.87
Intention to use4 -0.44 0.64 0.29 0.64 0.85

Table 7

Discriminant validity of the tested latent variables assessed by Fornell & Larcker criterium

Pomiar Observer Actor Perceived ease Perceived usability Intention to use
Observer 0.74
Actor -0.39 0.82
Perceived ease -0.29 0.27 0.83
Perceived usability -0.47 0.61 0.49 0.80
Intention to use -0.43 0.66 0.36 0.72 0.78

Nota: The table presents the square root of AVE on the diagonal.

Results of path estimatates of structural model

The analysis of estimates of path coefficients in Table 8 and at fig 2. indicated that increased intensity of observer measure was significantly related with decreased level of perceived ease of use and perceived usability and indicated that increased intensity of actor measure was significantly related with increased level of perceived ease of use and perceived usability. Analysis showed also that perceived usability was related with incresed level of intention to use. Highest explained variability was observed in intention to use \(R^2\) = 0.69, little less was observed in perceived usability \(R^2\) = 0.54, and smallest variability was explained in perceived ease of use \(R^2\) = 0.14.

Table 8

Results of path estimatates of structural model

Path β Boot β Boot deviation t LCI HCI
Observer -> Perceived ease -0.25 -0.26 0.09 -2.71** -0.44 -0.05
Observer -> Perceived usability -0.29 -0.29 0.07 -4.16*** -0.42 -0.15
Actor -> Perceived ease 0.19 0.18 0.08 2.38** 0.04 0.34
Actor -> Perceived usability 0.56 0.56 0.05 10.21*** 0.45 0.67
Perceived ease -> Intention to use -0.07 -0.07 0.06 -1.17 -0.18 0.05
Perceived usability -> Intention to use 0.87 0.87 0.05 18.34*** 0.77 0.96

Note: β = Standardized Path Coefficient; Boot β = Standardized Path Coefficient estimated during the bootstrap procedure; Boot deviation = Deviation of the bootstrap estimate from the base estimate; t = Student’s t-statistic; LCI and HCI = Lower and Upper 95% Confidence Intervals for the standardized path coefficient.

*** p < 0.001

** p < 0.01

* p < 0.05

Fig 2

Results of confirmatory factor loadings estimates in the measurement model and path coefficients in the structural model

Note: * p < 0.05, ** p < 0.01, *** p < 0.001

References

Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16(3), 297–334. https://doi.org/10.1007/bf02310555
Dijkstra, T. K., Jörg Henseler, and, & and. (2015). Consistent partial least squares path modeling. MIS Quarterly, 39(2), 297–316. https://doi.org/10.25300/misq/2015/39.2.02
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104
Hair, J. F., Sarstedt, M., Ringle, C. M., & Mena, J. A. (2011). An assessment of the use of partial least squares structural equation modeling in marketing research. Journal of the Academy of Marketing Science, 40(3), 414–433. https://doi.org/10.1007/s11747-011-0261-6
Henseler, J., Ringle, C. M., & Sarstedt, M. (2014). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115–135. https://doi.org/10.1007/s11747-014-0403-8
Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879–903. https://doi.org/10.1037/0021-9010.88.5.879
R Core Team. (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
Ray, S., Danks, N. P., & Calero Valdez, A. (2022). Seminr: Building and estimating structural equation models. https://CRAN.R-project.org/package=seminr
Sarstedt, M., Hair, J. F., Ringle, C. M., Thiele, K. O., & Gudergan, S. P. (2016). Estimation issues with PLS and CBSEM: Where the bias lies! Journal of Business Research, 69(10), 3998–4010. https://doi.org/10.1016/j.jbusres.2016.06.007