DATA PROCESSING NOTES

1. Panel coverage
The principal analytical panel contains 341 observations covering 31 provinces from 2013 to 2023. No missing values or duplicate province-year keys were detected in the supplied final panel.

The desktop source workbook contained one blank separator row followed by three unlabeled percentile rows beneath the 341 province-year observations. The clean submission copy retains only rows with valid province and year identifiers. The percentile values are documented separately in fsQCA_Calibration_Anchors.csv.

2. Super-SBM inputs and outputs
The efficiency model uses four inputs: culture/sports/media expenditure, employment in culture/sports/entertainment, park area and the number of A-rated scenic attractions. The two desirable outputs are forestry leisure tourism visits and forestry leisure tourism revenue.

The supplied SuperSBM compatibility workbook records the following treatment for forestry leisure tourism visits: the 2022 value for each province is the mean for 2012-2021, and the 2023 value is the mean for 2013-2022. The workbook contains the 2012 auxiliary values and a 62-row verification log covering both imputed years for all 31 provinces.

3. FLTE calculation
The model setting is Global Super-SBM under constant returns to scale, with four inputs, two desirable outputs and no undesirable output. The 341 efficiency estimates in SuperSBM_Efficiency.xls match the FLTE values in Panel_Data_Level.xlsx and XGBoost_SHAP_Input.csv by province and year.

4. SDM transformation
SDM_Log_Transformed_Data.dta contains the natural logarithms of y and x1-x5 in Panel_Data_Level.xlsx. Exponentiating the Stata variables reproduces the level-form values for all 341 observations to numerical precision.

5. fsQCA calibration
The supplied calibration record uses the direct method with the 75th percentile as full membership, the 50th percentile as the crossover and the 25th percentile as full non-membership. Membership scores equal to exactly 0.500 are to be adjusted to 0.501. The FLTE series in the supplied fsQCA input reproduces the calibration anchors used in the revised fsQCA analysis. It is retained as an analysis-specific outcome series and is stored separately from the Global Super-SBM efficiency series.

6. XGBoost-SHAP input
The XGBoost-SHAP input contains the Global Super-SBM FLTE result, geographic coordinates and the five level-form predictors. It matches the level-form panel for FLTE and x1-x5.
