Table 6.
Estimated heterogeneous impacts of the Grassland Ecological Compensation Program (GECP) on grassland quality: county level.
| (1) | (2) | (3) | |||
|---|---|---|---|---|---|
| Low | Medium | High | |||
| Panel A by rural road intensity in 2008: | |||||
| P × T | 0.017 | 0.025*** | 0.053*** | ||
| (0.157) | (0.011) | (0.000) | |||
| Number of observations | 1,122 | 1,144 | 1,149 | ||
| R2 | 0.994 | 0.988 | 0.977 | ||
| Low | Medium | High | |||
| Panel B by NDVI in 2008: | |||||
| P × T | 0.035*** | 0.055*** | 0.026*** | ||
| (0.002) | (0.000) | (0.001) | |||
| Number of observations | 1,128 | 1,150 | 1,086 | ||
| R2 | 0.991 | 0.966 | 0.948 | ||
| Meadow | Grassland | Desert | Shrub land | Herbosa | |
| Panel C by grassland type: | |||||
| P × T | 0.001 | 0.055*** | 0.030*** | 0.018** | 0.038*** |
| (0.850) | (0.000) | (0.000) | (0.046) | (0.000) | |
| Number of observations | 2,038 | 2,380 | 2,231 | 1,894 | 1,714 |
| R2 | 0.969 | 0.979 | 0.995 | 0.952 | 0.953 |
Note. The dependent variable is NDVI in log form. NDVI stands for the Normalized Difference Vegetation Index, which is a measure for grassland quality. It was constructed based on infrared and near-infrared channel remote sensing images and has been widely used as an indicator of vegetation coverage. In Panels A and B, we present the results when we first divide the control and treatment groups into three terciles (i.e., low, medium, and high), based on each indicator in a base year (i.e., 2008). In Panel C, all of the control and treatment counties are grouped into five grassland types. We then pair the subgroups in the control and treatment groups and run the model indicated by Eq. (1). All other model specifications are the same as in Column (1) of Table 1. A two-sided t test is conducted for each coefficient. The exact p-values are in parentheses.
**p < 0.05, ***p < 0.01.