Using Convex Combinations of Spatial Weights ...
Type de document :
Partie d'ouvrage
Titre :
Using Convex Combinations of Spatial Weights in Spatial Autoregressive Models
Auteur(s) :
Titre de l’ouvrage :
Handbook of Regional Science
Éditeur :
Springer Berlin Heidelberg
Lieu de publication :
Berlin, Heidelberg
Date de publication :
2019-05-31
Mot(s)-clé(s) en anglais :
Markov Chain Monte Carlo estimation
Taylor series approximation
log-marginal likelihood
multiple weight matrices
Taylor series approximation
log-marginal likelihood
multiple weight matrices
Discipline(s) HAL :
Sciences de l'Homme et Société/Méthodes et statistiques
Sciences de l'Homme et Société/Economies et finances
Sciences de l'Homme et Société/Economies et finances
Résumé en anglais : [en]
Spatial regression models rely on simultaneous autoregressive processes that model spatial or cross-sectional dependence between cross-sectional observations using a weight matrix. A criticism of applied spatial regression ...
Lire la suite >Spatial regression models rely on simultaneous autoregressive processes that model spatial or cross-sectional dependence between cross-sectional observations using a weight matrix. A criticism of applied spatial regression methods is that reliance on geographic proximity of observations to form the weight matrix that specifies the structure of cross-sectional dependence might be unrealistic in some applied modeling situations. In cases where the structure of dependence or connectivity between (cross-sectional) observations arises from non-spatial relationships, spatial weights are theoretically unjustifiable. Some literature addresses the structure of dependence between observations by introducing geographic proximity as well as other types of non-spatial proximity, resulting in a model that utilizes multiple weight matrices. Each set of weights reflect a different type of dependence specified using linear combinations of observations defined by alternative characteristics. The multiple weight matrix approach results in a simultaneous autoregressive process that poses a number of challenges for parameter estimation and model interpretation. We focus on literature that relies on a single connectivity matrix constructed from a convex combination of multiple matrices, each of which reflects a different type of dependence or interaction structure. The advantage of this approach is that the resulting simultaneous autoregressive process is amenable to conventional spatial regression estimation algorithms as well as methods developed for interpretation of estimates from these models. Estimates of the scalar parameters used to form the convex combination of the weight matrices can be used to produce an inference regarding the relative importance of each type of dependence.Lire moins >
Lire la suite >Spatial regression models rely on simultaneous autoregressive processes that model spatial or cross-sectional dependence between cross-sectional observations using a weight matrix. A criticism of applied spatial regression methods is that reliance on geographic proximity of observations to form the weight matrix that specifies the structure of cross-sectional dependence might be unrealistic in some applied modeling situations. In cases where the structure of dependence or connectivity between (cross-sectional) observations arises from non-spatial relationships, spatial weights are theoretically unjustifiable. Some literature addresses the structure of dependence between observations by introducing geographic proximity as well as other types of non-spatial proximity, resulting in a model that utilizes multiple weight matrices. Each set of weights reflect a different type of dependence specified using linear combinations of observations defined by alternative characteristics. The multiple weight matrix approach results in a simultaneous autoregressive process that poses a number of challenges for parameter estimation and model interpretation. We focus on literature that relies on a single connectivity matrix constructed from a convex combination of multiple matrices, each of which reflects a different type of dependence or interaction structure. The advantage of this approach is that the resulting simultaneous autoregressive process is amenable to conventional spatial regression estimation algorithms as well as methods developed for interpretation of estimates from these models. Estimates of the scalar parameters used to form the convex combination of the weight matrices can be used to produce an inference regarding the relative importance of each type of dependence.Lire moins >
Langue :
Anglais
Audience :
Internationale
Vulgarisation :
Non
Collections :
Source :
Fichiers
- https://halshs.archives-ouvertes.fr/halshs-03509810/document
- Accès libre
- Accéder au document
- https://halshs.archives-ouvertes.fr/halshs-03509810/document
- Accès libre
- Accéder au document
- https://halshs.archives-ouvertes.fr/halshs-03509810/document
- Accès libre
- Accéder au document
- document
- Accès libre
- Accéder au document
- Debarsy_LeSage_final.pdf
- Accès libre
- Accéder au document