Table 7 Reliability and validity analysis of the subjective visual perception model.
Category | Method/Indicator | Description | Results/Implications |
|---|---|---|---|
Reliability (Internal Consistency) | Cronbach’s Alpha | A measure of internal consistency to assess whether items in the questionnaire consistently measure the same construct | Cronbach’s alpha coefficient was 0.83, indicating strong internal consistency (Nasar, 1994; adapted Chinese version) |
Item-Total Correlation | Examined the correlation between each item and the total score to identify redundant or irrelevant items | Items with low correlations (< 0.3) were revised during the pilot test (n = 50 participants) | |
Reliability (Test–Retest) | Repeatability Test | A subset of participants (n = 20) re-rated the same images after 1 week to assess stability of responses | Intra-class correlation coefficient (ICC) was 0.87, confirming high test–retest reliability |
Validity (Content Validity) | Expert Review | A panel of 5 experts in landscape architecture and cultural heritage validated the relevance of the 8 visual perception indicators (space, color, texture, uniqueness, history, culture, aesthetics, pleasure) | All indicators were confirmed as critical for evaluating water conservancy heritage landscapes |
Validity (Construct Validity) | Factor Analysis | Principal component analysis (PCA) was conducted to verify if the 8 indicators grouped into meaningful latent factors (environmental vs. heritage features) | Explained variance: 78.3%. Two distinct factors emerged: Environmental Features (space, color, texture) and Heritage Features (uniqueness, history, culture) |
Convergent Validity | Correlations between visual perception scores and image-based metrics (e.g., water proportion, vegetation, architectural structure) were analyzed | Strong positive correlations (r = 0.62–0.78), supporting the alignment of subjective and objective measures | |
Validity (Criterion Validity) | External Benchmarking | Scores were compared with historical records and cultural significance rankings of the 32 sites | Sites with higher historical/cultural value (e.g., Site 29) scored significantly higher (p < 0.01), validating the model’s sensitivity |
Statistical Validity | Linear Mixed-Effects Model | Random intercepts were used to account for within-subject correlations (32 sites rated by 3840 participants). Fixed effects included water proportion, vegetation, and texture complexity | Model diagnostics confirmed normality and homoscedasticity of residuals, ensuring robust statistical inferences |
Data Triangulation | Triangulation with Image Analysis | Visual perception scores were cross-validated with semantic segmentation outputs (e.g., texture complexity, color saturation) derived from drone-captured images | Consistent patterns observed: High-scoring sites (e.g., Site 29) exhibited higher color diversity and structured spatial layouts |
