ABSTRACT
Spatial panel models expressed in terms of either random or fixed effects, can be estimated employing a maximum likelihood or a generalized method of moments approach. Moving to a spatial lag version of a random effects panel data model, this can be written as a combination of a spatial filtering on the dependent variable y and a random effects structure for the disturbances. From a statistical point of view, the random effects hypothesis is associated to the idea of sampling individuals from infinite population, which has led Elhorst to consider it practically irrelevant in the spatial econometric contexts. Although there are several possible alternative statistical modelling frameworks for space–time economic data, the dominant approach in econometrics considers spatial panels of individuals. In microeconometrics, panel data models are used to control for “unobserved heterogeneity” related to individual-specific, time-invariant characteristics which are difficult or impossible to observe although they can lead to biased, inefficient estimates of the parameters of interest if omitted.
