CFA (confirmatory factor analysis) research methods and SPSS 26.0 and AMOS 24. 0 software were used for the analysis of the data from questionnaires.
5.3. Confirmatory Factor Analysis
The data presented herein indicate a CFA (see
Figure 4) aimed at assessing the relationships among multiple constructs (such as public art, citizen perceived benefits, and walking intention) and their respective observed variables (such as environmental improvement, cultural shaping, service enhancement and public interaction). CFA is a key component of SEM, used to verify whether the preconceived theoretical structure holds true in the observed data.
The construct of public art exhibits strong influences on its downstream variables (environmental improvement, cultural shaping, service enhancement, public interaction), with standardized regression coefficients (β) ranging from 0.81 to 0.832 (see
Table 5). The overall average variance extracted (AVE) value is 0.674, and the composite reliability (CR) value is 0.892, indicating good convergent validity and reliability. Similarly, the construct of citizen-perceived benefits demonstrates significant impacts on the functional, experiential, and social dimensions (standardized regression coefficients ranging from 0.835 to 0.87), with an AVE value of 0.732 and a CR value of 0.891, also indicating good convergent validity and internal consistency.
Regarding the downstream variables, each dimension shows strong explanatory power for its observed variables. Particularly noteworthy are the high standardized regression coefficients (β) for environmental improvement in Q1, cultural shaping in H1, and service enhancement in M1, all approaching or exceeding 0.9, indicating strong associations with their corresponding constructs.
Overall, most constructs exhibit AVE values exceeding the threshold of 0.5, and CR values surpassing the acceptable standard of 0.7, indicating good convergent validity and internal consistency of the data, thereby validating the adaptability of the theoretical model to the data. The results of this CFA support the hypothesized relationships between the predefined constructs and their corresponding observed variables in the research model, demonstrating a solid theoretical foundation and empirical support for the model.
The table presents correlation coefficients among a series of variables (see
Table 6), along with the square root of the Average Variance Extracted (AVE) for each variable, commonly used in SEM to assess the discriminant validity of constructs. Discriminant validity refers to the distinction between different constructs and whether each construct captures information distinct from other constructs. A commonly used criterion is that the square root of the AVE for each construct should be greater than the correlations of that construct with other constructs in the model in order to demonstrate good discriminant validity [
49].
The values on the diagonal represent the square root of the AVE for each construct. This value should be greater than the correlation of that construct with any other construct in order to demonstrate discriminant validity [
50]. Correlation coefficients between constructs, denoted by asterisks (**), indicate statistically significant correlations. It can be observed from the table that the square roots of the AVE for the constructs of public art, citizen-perceived benefits, and walking intention are 0.821, 0.787, and 0.780, respectively. Moreover, all correlations associated with these constructs are lower than the square roots of the corresponding AVEs, meeting the requirements for discriminant validity. These constructs are effectively differentiated, each capturing unique information, aiding in a clear interpretation of different factors in the research model. Such clarity in differentiation is crucial for understanding the relationships between different variables and how they influence research outcomes.
The comprehensive analysis reveals that the model fits well and is applicable (see
Table 7). The model’s χ
2 value is 485.307 with 366 degrees of freedom, resulting in a chi-square to degrees of freedom ratio of 1.326. The ratio of χ
2/df, being below three, indicates a good fit of the model. Although the
p-value of the χ
2 test is significant, in the case of large samples, this value may be overly sensitive to minor deviations. The Goodness-of-Fit Index (GFI) exceeds 0.9, the Root Mean Square Error of Approximation (RMSEA) is well below 0.10, and the Root Mean Square Residual (RMR) is close to the threshold for a good fit, indicating a good fit of the model to the data. The Comparative Fit Index (CFI), Normed Fit Index (NFI), Non-Normed Fit Index (NNFI), Tucker–Lewis Index (TLI), and Incremental Fit Index (IFI) all exceed the standard of 0.9 for good model fit, while the Adjusted Goodness-of-Fit Index (AGFI) is slightly lower but still close to the good fit criterion. The Parsimonious Goodness-of-Fit Index (PGFI), Parsimonious Normed Fit Index (PNFI), and Parsimonious Comparative Fit Index (PCFI) all exceed the standard of 0.5, indicating good parsimony of the model. The Standardized Root Mean Square Residual (SRMR) is below 0.1, and the 90% confidence interval for RMSEA further confirms the stability and excellence of the results. In summary, the model is deemed appropriate for practical application.
5.5. Path Analysis
χ
2/df: The value of 1.317, significantly below the standard threshold of three, indicates a good model fit. This is an important indicator for evaluating the overall goodness of fit of the model, with values below three typically indicating a good fit between the model and the data (see
Figure 5).
GFI: With a value of 0.909, meeting the criterion of >0.9 for good fit, indicates a well-fitted model with the observed data.
RMSEA: At 0.032, well below the threshold of 0.10, suggests a small model error and excellent fit.
RMR: With a value of 0.050, close to the ideal threshold of <0.05, indicates small residuals and a high degree of fit between the model and the data.
CFI, NFI, NNFI: These indices all exceed the threshold of 0.9 for good fit, with values of 0.980, 0.922, and 0.978, respectively, indicating very good relative fit of the model.
TLI and IFI: Similar to NNFI, TLI has a value of 0.978, and IFI has a value of 0.980, both surpassing the good fit threshold of 0.9, further confirming the excellent fit of the model.
AGFI: Slightly increased to 0.893, and slightly below the ideal value of 0.9 but close to it, this indicates a good fit between the model and the data even after considering model complexity.
PGFI, PNFI, and PCFI: These indices consider the parsimony of the model, with values of 0.769, 0.836, and 0.888, respectively, all exceeding the threshold of 0.5, indicating good fit of the model while maintaining reasonable complexity.
SRMR: With a value of 0.041, below the threshold of 0.1, this indicates small residuals and good fit.
RMSEA 90% CI (RMSEA 90% Confidence Interval): Ranging from 0.023 to 0.039, this narrow confidence interval further confirms the stability and excellence of the RMSEA results.
Based on these fit indices (see
Table 9), it can be concluded that the model has an excellent fit. Almost all indices meet or exceed their respective criteria for good fit, indicating that the model adequately reflects the structural relationships in the data. Although the
p-value from the χ
2 test indicates a statistically significant difference between the model and perfect fit, this is common in large-sample studies [
51]. Considering that other fit indices all demonstrate good model fit, it can be deemed that this model is appropriate for practical application.
The summary table of regression coefficients (
Table 10) provides a detailed overview of the relationships between variables, including Standard Error (SE), CR,
p-values, and standardized regression coefficients. Analyzing these data allows us to understand how variables interact within the model, as well as the statistical significance and strength of these interactions [
52].
Standardized regression coefficients indicate the change in the standard deviation of the dependent variable Y when the independent variable X changes by one standard deviation [
53]. Higher values suggest a stronger impact. Standard Error (SE) reflects the precision of the estimates, while CR is used to test hypotheses, with higher z-values indicating statistically significant regression coefficients [
54]. The
p-value is used to assess the significance of the regression coefficients, typically with
p < 0.05 indicating statistical significance [
55].
The impact of public art on citizens perceived benefits is manifested by a standardized regression coefficient of 1, indicating a very strong positive effect, and it is highly significant statistically (p = 0). The effect of citizen perceived benefits on walking intention is represented by a standardized regression coefficient of 0.84, indicating a strong positive effect, and similarly it is highly significant statistically (p = 0).
In other aspects, the influence of public art on environmental improvement, cultural shaping, service enhancement, and public interaction is highly significant, with standardized regression coefficients of 0.802, 0.829, 0.807, and 0.826, respectively. Similarly, the impact of citizens’ perceived benefits on functionality, experience, and social aspects is also highly significant, with standardized regression coefficients of 0.86, 0.857, and 0.83, respectively.