Year: 2009
Type: article
Abstract: Geographically weighted regression (Fotheringham et al., 2002) is a method of modelling \nspatial variability in regression coefficients. The procedure yields a separate model for \neach spatial location in the study area with all models generated from the same data set \nusing a differential weighting scheme. The weighting scheme, which allows for spatial \nvariation in the model parameters, involves a bandwidth parameter which is usually deter- \nmined from the data using a cross-validation procedure. Part of the main output is a set \nof location-specific parameter estimates and associated \nt \nstatistics which can be used to \ntest hypotheses about individual model parameters. If there are \nn \nspatial locations and \np \nparameters in each model, there will be up to \nnp \nhypotheses to be tested which in most \napplications defines a very high order multiple inference problem. Solutions to problems \nof this type usually involve an adjustment to the decision rule for individual tests designed \nto contain the overall risk of mistaking chance variation for a genuine effect. An undesir- \nable by-product of achieving this control is a reduction in statistical power for individual \ntests, which may result in genuine effects going undetected. These two competing aspects \nof multiple inference have become known as the multiplicity problem. In this paper we \ndevelop a simple Bonferroni style adjustment for testing multiple hypotheses about GWR \nmodel coefficients. The adjustment takes advantage of the intrinsic dependency between \nlocal GWR models to contain the overall risk mentioned above, without the large sacrifice \nin power associated with the traditional Bonferroni correction. \nWe illustrate this adjustment and a range of other corrective procedures on two data \nsets. The first models the determinants of educational attainment in the counties of Georgia \nUSA. Using area based census data we examine the links between levels of educational \nattainment and four potential predictors: the proportion of elderly, the proportion who are \nforeign born, the proportion living below the poverty line and the proportion of ethnic \nblacks. The second model is a geographically weighted hedonic house price model based \non individual mortgage records in Greater London in 1990. In both models we show how \nthe various corrections can be used to guide the interpretation of the spatial variations in \nthe parameter estimates. Finally we compare the statistical power of the proposed method \nwith Bonferroni/Sidak corrections and those based on \nfalse discovery rate \ncontrol.
Institutions: La Trobe University, National University of Ireland, Maynooth
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W904673569
