1. In Table 8, the SLM-FE column, InBed has a coefficient of 0.425*** with a Std. Err. of 0.002. This is statistical noise approaching zero—implying an almost deterministic relationship. Given you used log-transformed data and a panel, this is impossible. If this is a typo, the significance is suspect; if real, it suggests severe multicollinearity or data leakage. Which is it?
2. Table 4 shows Global Moran’s I dropping from 0.298 (2016) to 0.087 (z=1.99, barely significant) and then rebounding to 0.358 in 2018, a near-complete structural break. You never mention this. What policy, data revision, or boundary change occurred that year? Ignoring this anomaly means your fixed-effects SDM likely masks a misspecified temporal dynamic.
3. You use “road infrastructure density” as an instrument for income. Road density directly determines healthcare accessibility and workforce attraction (rural doctors won’t move to areas with poor roads). It violates the exclusion restriction. Your IV results (coefficient dropping from 0.41 to 0.07) are therefore meaningless, and your claim of robust income effects is unsubstantiated.
4. In Table 1, you report Min/Max for InIncome as 9011 and 59551 (raw CNY), yet your Methods clearly state all variables (except PG) were log-transformed. This is not a minor typo, it misrepresents the actual distribution fed into the SDM. The reader cannot verify whether outliers drove your results.
5. Your indirect effect for InIncome is 1.376; meaning a 1% rise in a neighbor’s income increases local HW by >1%. In a fixed labor pool, this defies economic competition (staff flow toward richer neighbors, not away). You attribute this to resource sharing, but the SDM captures unobserved confounders (e.g., central grants), not causal spillovers. This interpretation is misleading.