This study employed a two-step process to analyse the output of the SEM path results. The two-step process employed in this study includes (1) the measurement model assessment and (2) the structural model assessment. The measurement model assessed the reliability and validity of the study variables and items. Then, the structural model was used to evaluate the hypothesised relationships (Fig. 1) in terms of their significance.
4.1 Measurement model
The measurement model was evaluated in terms of its convergent and discriminant validity. Convergent validity is the assessment employed to measure the level of correlation of multiple indicators of the same variable that are in agreement (Hamid, Sami & Sidek., 2017:2). Convergent validity was assessed using the average variance extracted (AVE) and the composite reliability (CR). To establish convergent validity, the value of AVE should be greater than or equal to 0.50, whilst the composite reliability for all latent variables should be above 0.70 (Hamid et al., 2017:2). Table 3 presents the outcome of the item's loadings Cronbach's Alpha values, and composite reliability values. All the observations exceed the threshold values of 0.5 (AVE) and 0.70 (CR). These results were used to confirm convergent validity.
Table 3
Convergent validity assessment
Latent variable
|
Measurement item
|
Cronbach alpha
|
C.R. value
|
AVE value
|
Factor loading
|
Tax morale (TM)
|
-
TM2
TM3
TM4
TM5
-
-
-
|
0.823
|
0.829
|
0.568
|
-
0.480
0.565
0.938
0.917
-
-
-
|
Taxpayer/tax office relationship (TTR)
|
-
TTR2
TTR3
TTR4
-
TTR6.1
TTR6.2
TTR6.3
TTR6.4
TTR6.5
|
0.755
|
0.759
|
0.522
|
-
0.562
0.842
0.695
-
0.517
0.576
0.639
0.786
0.722
|
Corruption (C)
|
C1
C2
C3
-
C5
|
0.719
|
0.817
|
0.558
|
0.509
0.759
0.664
-
0.596
|
Presumptive tax compliance (PTC)
|
PTC1
PTC2
PTC3
PTC4
|
0.857
|
0.872
|
0.636
|
0.686
0.899
0.919
0.649
|
Source: own research |
After assessing the convergent validity of the measurement model, the discriminant validity of the study components was also evaluated using the Fornell and Larcker (1981) criterion. The discriminant validity is measured using the AVE square roots and the variables' correlations co-efficient. It measures the degree of differences between overlapping variables (Hamid et al., 2017:2). According to Fornell and Larker (1981:45), discriminant validity is achieved when the diagonal values in bold (square root of AVE) are higher than the values in its row and column. Table 4 shows that the square root of AVE exceeds the diagonal values for each row and column, indicating discriminant validity between the variables. Therefore, it is evident that the measurement model meets the validity and reliability requirements.
Table 4
Discriminant validity assessment
|
CR
|
AVE
|
TM
|
C
|
TTR
|
PTC
|
TM
|
0.829
|
0.568
|
0.754
|
|
|
|
TTR
|
0.759
|
0.522
|
0.021
|
0.722
|
|
|
C
|
0.817
|
0.558
|
-0.24
|
0.237
|
0.747
|
|
PTC
|
0.872
|
0.636
|
0.348
|
-0.36
|
0.082
|
0.797
|
Source: Own research |
4.2 Model goodness of fit assessment
One of the key steps in applying structural equation modelling is evaluating the model's goodness-of-fit index with data (Munyanyi & Pooe, 2020:12). The model fit can be assessed by considering the chi-square (CMIN), root mean square error of approximation (RMSEA), goodness-of-fit (GFI), adjusted goodness-of-fit (AGFI), root mean square residual (RMR), standard root mean residual (SRMR), normed fit index (NFI), Tucker Lewis Index (TLI), comparative fit index (CFI), parsimony goodness-of-fit index (PGFI) and Akaike Information Criterion (Hooper, Coughlan & Mulley, 2008:54–55; Sun, 2005:246). The RMSEA cut-off range of 0 to 0.08 is considered an indication of a good fit (Feng & Chen, 2020:9; Hooper et al., 2008:54; Sun, 2005:249). Regarding CFI, GFI and TLI, the values should be greater than or equal to 0.90 to attain an acceptable fit (Feng & Chen, 2020:9; Hooper et al., 2008:55; Sun, 2005:249). Another important fit index is chi-square, which must be greater than 0.05 for the model to fit (Walker, 2010:21). Furthermore, Feng and Chen (2020:9) report that chi-square degrees of freedom (CMIN/df) values of less than 3 are acceptable to achieve a well-fitted model. In this study, CMIN, CMIN/df, RMSEA, TLI, CFI, and GFI were considered when assessing the model fit. All indices met the acceptable range for a good model fit (Table 5). Therefore, analysing the path relationships between study variables was possible using a structural model.
Table 5
Goodness-of-fit assessment
Name of category
|
Name of index
|
Level of acceptance
|
Value
|
Absolute fit
|
Chi-square
|
P-value > 0.05
|
1021.75
|
RMSEA
|
RMSEA < 0.08
|
0.061
|
GFI
|
GFI > 0.90
|
0.916
|
Incremental fit
|
CFI
|
CFI > 0.90
|
0.920
|
TLI
|
TLI > 0.90
|
0.906
|
Parsimonious
|
Chisq/df
|
Chisq/df < 3.0
|
1.890
|
Source: Own research |
4.3 Structural model analysis
The assessment of the structural model incorporated 219 cases, and the results are demonstrated in Table 6. Using the t-values, the hypothesised relationships were analysed. The path coefficient's significance was also examined to explain the degree of association between the independent and dependent variables.
Table 6: Results of hypotheses testing
Source: own research
\({H}_{1}\) proposed that there is a significant positive relationship between tax morale (TM) and presumptive tax compliance (PTC). The results in Table 6 revealed a significant relationship between TM and PTC (β = 0.438; t = 3.183; p < 0.001). Thus, this hypothesis was supported. This result supports Nichita and Batrancea (2012:741), who found that high tax morale leads to high tax compliance levels.
Regarding the impact of taxpayer/tax office relationships (TTR) on PTC, the result of the \({,H}_{2}\) which predicted a significant positive relationship between TTR and PTC indicated that TTR had no significant influence on PTC (β = -0.007; t = -0.042; p > 0.001). Therefore, \({H}_{2 }\)is not supported. These results contrast with the views of Inasius (2019:373), who concluded that the taxpayer/tax office relationship improves if the government spends the tax revenue wisely, for example, on basic facilities such as public transportation and education, and this leads to an increase in voluntary presumptive tax compliance.
In this study, \({H}_{3 }\)proposed that there is a significant negative relationship between corruption (C) and PTC. The results show that \({H}_{3 }\)is supported (β = -0.514; t = -4.283; p < 0.001). This is consistent with Bertinelli, Bourgain and Leon (2020:366), who argue that corruption negatively influences tax payments as individuals or firms pay bribes to reduce or avoid tax payments.
Moreover, this study validated the conceptual model that was evaluated using the SEM. From the three hypotheses proposed, the results provide support for two hypotheses. From the empirical results of the study, tax morale has a positive influence on the level of compliance with presumptive taxation. However, the existence of corruption has a negative impact on decisions to comply with presumptive taxation. The validated model for presumptive tax compliance used in this study is shown in Fig. 2.