Gibbons and Overman argue that identification problems bedevil applied spatial economic research. Spatial econometrics typically "solves" these by assuming known functional forms and using model-comparison techniques to choose among competing specifications, but the authors contend this achieves at best very weak identification and is often uninformative about the causal economic processes at work — rendering much applied spatial-econometric research "pointless" unless the aim is mere description. They advocate instead an "experimentalist paradigm" that places identification and causality at center stage.
"We argue that in many situations of interest this achieves, at best, only very weak identification. Worse, in many cases, such an approach will be uninformative about the causal economic processes at work, rendering much applied spatial econometric research 'pointless,' unless the main aim is description of the data."
A bracing corrective to the machinery catalogued on the spatial-autocorrelation page: the LM tests and SAR/SEM estimators are fine for detecting and describing dependence, but Gibbons-Overman are right that fitting them does not, by itself, identify why the dependence exists — the reflection problem guarantees that the weights matrix and functional form silently do the identifying work. The paper is squarely in the credibility-revolution tradition (its title winks at Angrist-Pischke's "Mostly Harmless"/"Mostly Pointless" framing), and its prescription — design for identification rather than assume it — is the same discipline the joint-hypothesis and forecast-evaluation literatures impose elsewhere. The fair rejoinder is that description and prediction are legitimate goals, and spatial models remain useful there; the paper's target is specifically causal over-claiming.