Academic journal article Journal of Real Estate Literature

About the Influence of Time on Spatial Dependence: A Meta-Analysis Using Real Estate Hedonic Pricing Models

Academic journal article Journal of Real Estate Literature

About the Influence of Time on Spatial Dependence: A Meta-Analysis Using Real Estate Hedonic Pricing Models

Article excerpt

(ProQuest: ... denotes formulae omitted.)

The increasing availability of geolocated data opens up new avenues to spatial analysis, especially in real estate applications. Following the identification of the spatial dependence issue by Cliff and Ord (1968), the number of scientific publications associated with the keywords spatial autocorrelation, spatial analysis, and spatial econometrics has increased since the 1990s and illustrates the enthusiasm for such methods (Exhibit 1).1 While the classical correlation statistic refers to the relations between variables, spatial autocorrelation refers to the correlation between values of different observations for a single variable and is described as the coincidence between observed variables depending on location (Anselin and Bera, 1998; Chasco and Lopez, 2008; LeSage and Pace, 2009). Spatial autocorrelation is a key concept in studies using spatial data: ''No other concept in empirical spatial research is as central to model building as is spatial autocorrelation'' (Getis, 2008, p. 299).

Spatial analysis and spatial econometrics have been developed after the pioneer work of Cliff and Ord (1968) and Paelinck et al. (1979) and are a response on how to account for spatial dependence in statistical models. The difference between the two lies in how spatial dimension is accounted for in modeling strategies, especially through the data generating process (DGP) and the way we specify the spatial connections among geographical observations. In spatial econometrics, this connection is formalized through a weights matrix. Some argue that even in an exclusively spatial perspective, the weights matrix is the cornerstone of spatial econometrics, and due to its exogenous specification, also its main weakness (Anselin, 2002; Elhorst, 2010; Bhattacharjee, Castro, and Marques, 2012; Chen, 2012). On the other hand, others claim that too much attention is drawn to defining the good spatial weight matrix specification, which became the biggest myth in spatial econometrics (LeSage and Pace, 2010).

In recent years, spatial econometric developments tend to include the temporal dimension to the spatial cross-sectional perspective (Arbia, 2011). Adding the temporal perspective is a critical transition to analyze the dynamic process and its evolution (Anselin, 2002). By including a temporal dependency in spatial analysis, the DGP becomes more complex. Spatiotemporal models are mainly developed through the panel or pseudo-panel perspectives (Elhorst, 2013). Few attempts have been made, outside the spatial panel analysis perspective, to combine spatial and temporal dependencies for spatial data collected over time, while not being repeated, such as in the case of real estate transactions. In such a case, time constraints are often neglected, assuming the simultaneous observation of the data (Dubéand Legros, 2011, 2013b). However, this assumption remains to be verified (Dubé, Baumont, and Legros, 2013) and a misinterpretation of the data structure could affect the detection and adequate correction of spatial autocorrelation, since ignoring the temporal dimension suggests an overestimation of spatial relations.

To our knowledge, no study has attempted to empirically demonstrate the impact of such omission on the spatial dependence statistics beyond a limited number of case studies or Monte Carlo simulations (e.g., Dubéand Legros, 2014a, 2015). The stakes are high because a poor detection and correction for autocorrelation may bias estimators or affect their variance and therefore invalidate the results obtained from statistical analysis. The presence of spatial autocorrelation in residuals of ordinary least squares (OLS) models leads to increased risk of type 1 error and affects the model goodness-of-fit statistic (R2) (Anselin, 2002; Haining, 2009). ''This is problematic since failing to account for serial and spatial autocorrelation when present causes the OLS estimators to lose their property of efficiency'' (Elhorst, 2001, p. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.