Interpreting heterogeneous coefficient spatial autoregressive panel models

- Partial derivatives for heterogeneous coefficient SAR models are derived.
- Scalar summary measures proposed in the literature are not likely to work here.
- Observation-level marginal effects are proposed.
- Non-linear relationships between estimates and observation-level marginal effects arise as in probit.
- Spatial spill-out and spill-in effects can be quantified.

We consider interpretation of estimates from the heterogeneous coefficient spatial autoregressive panel model of Aquaro et al. (2015) and derive partial derivatives (marginal effects) for this model, an issue not discussed in Aquaro et al. (2015). We show how these differ from a conventional spatial autoregressive panel model.