Caveat Emptor

• Published in: Multiscale Geographically Weighted Regression pp. 103 - 116
• Main Authors: Fotheringham, A. Stewart, Oshan, Taylor M., Li, Ziqi
• Format: Book Chapter
• Language: English
• Published: United Kingdom CRC Press 2024; Taylor & Francis Group
• Edition: 1
• ISBN: 1032564237; 9781032564227; 1032564229; 9781032564234
• DOI: 10.1201/9781003435464-6

Having described the rationale and mechanism of the MGWR framework, this chapter provides some issues that need to be considered prior to any local model calibration. These caveats are divided into pre-calibration issues-which include thinking about whether a local model actually makes sense to a given problem and what type of data are available; calibration issues-which include model format and potential problems with the calibration of local models; and post-calibration issues-which include inference and visualization. A checklist of issues to be checked prior to and during the calibration of any local model is also given. There are three sets of concerns anyone undertaking an empirical application using local models should address: pre-calibration caveats; calibration caveats; and post-calibration caveats. The first set concerns the application itself in terms of what processes are being modeled and at what locations data are available to allow such modeling. The second set concerns the form of the model being calibrated, and the third set concerns the interpretations of the results. The starting point for any successful local model calibration must be a good, robust, defensible, global model. One of the main outputs from the calibration of a local model is a set of geocoded local parameter estimates that can be mapped to display any spatial heterogeneity in their values; sufficient heterogeneity can then be used to infer process spatial nonstationarity. Many decisions that have to be taken in the calibration of a model are subjective and hence can be the subject of criticism or debate.