bw.GWPR: Bandwidth selection for basic GWPR

In GWPR.light: Geographically Weighted Panel Regression

Bandwidth selection for basic GWPR

Description

A function for automatic bandwidth selection to calibrate a GWPR model

Usage

bw.GWPR(formula, data, index, SDF, adaptive = FALSE, p = 2, bigdata = FALSE,
               upperratio = 0.25, effect = "individual",
               model = c("pooling", "within", "random"),
               random.method = "swar", approach = c("CV","AIC"), kernel = "bisquare",
               longlat = FALSE, doParallel = FALSE, cluster.number = 2,
               human.set.range = FALSE, h.upper = NULL, h.lower = NULL,
               gradientIncrement = FALSE, GI.step = NULL, GI.upper = NULL,
               GI.lower = NULL)

Arguments

formula	The regression formula: : Y ~ X1 + ... + Xk
data	data.frame for the Panel data
index	A vector of the two indexes: (c("ID", "Time"))
SDF	Spatial*DataFrame on which is based the data, with the "ID" in the index
adaptive	If TRUE, adaptive distance bandwidth is used, otherwise, fixed distance bandwidth.
p	The power of the Minkowski distance, default is 2, i.e. the Euclidean distance
bigdata	TRUE or FALSE, if the dataset exceeds 40,000, we strongly recommend set it TRUE
upperratio	Set the ratio between upper boundary of potential bandwidth range and the forthest distance of SDF, if bigdata = T. (default value: 0.25)
effect	The effects introduced in the model, one of "individual" (default) , "time", "twoways", or "nested"
model	Panel model transformation: (c("within", "random", "pooling"))
random.method	Method of estimation for the variance components in the random effects model, one of "swar" (default), "amemiya", "walhus", or "nerlove"
approach	Score used to optimize the bandwidth, c("CV", "AIC")
kernel	bisquare: wgt = (1-(vdist/bw)^2)^2 if vdist < bw, wgt=0 otherwise (default); gaussian: wgt = exp(-.5*(vdist/bw)^2); exponential: wgt = exp(-vdist/bw); tricube: wgt = (1-(vdist/bw)^3)^3 if vdist < bw, wgt=0 otherwise; boxcar: wgt=1 if dist < bw, wgt=0 otherwise
longlat	If TRUE, great circle distances will be calculated
doParallel	If TRUE, "cluster": multi-process technique with the parallel package would be used.
cluster.number	The number of the clusters that user wants to use
human.set.range	If TRUE, the range of bandwidth selection for golden selection could be set by the user
h.upper	The upper boundary of the potential bandwidth range for golden selection.
h.lower	The lower boundary of the potential bandwidth range for golden selection.
gradientIncrement	The bandwidth selection become gradient increment, if TRUE
GI.step	The step length of the increment.
GI.upper	The upper boundary of the gradient increment selection.
GI.lower	The lower boundary of the gradient increment selection.

Value

The optimal bandwidth

Author(s)

Chao Li <[e-mail]> Shunsuke Managi

References

Fotheringham, A. Stewart, Chris Brunsdon, and Martin Charlton. Geographically weighted regression: the analysis of spatially varying relationships. John Wiley & Sons, 2003.

Examples

data(TransAirPolCalif)
data(California)
formula.GWPR <- pm25 ~ co2_mean + Developed_Open_Space_perc + Developed_Low_Intensity_perc +
   Developed_Medium_Intensity_perc + Developed_High_Intensity_perc +
   Open_Water_perc + Woody_Wetlands_perc + Emergent_Herbaceous_Wetlands_perc +
   Deciduous_Forest_perc + Evergreen_Forest_perc + Mixed_Forest_perc +
   Shrub_perc + Grassland_perc + Pasture_perc + Cultivated_Crops_perc +
   pop_density + summer_tmmx + winter_tmmx + summer_rmax + winter_rmax

bw.CV.Fix <- bw.GWPR(formula = formula.GWPR, data = TransAirPolCalif,
                     index = c("GEOID", "year"),
                     SDF = California, adaptive = FALSE, p = 2, bigdata = FALSE,
                     effect = "individual", model = "within", approach = "CV",
                     kernel = "bisquare", longlat = FALSE,
                     gradientIncrement = TRUE, GI.step = 0.5, GI.upper = 5,
                     GI.lower = 1.5)

bw.CV.Fix

bw.AIC.Fix <- bw.GWPR(formula = formula.GWPR, data = TransAirPolCalif,
                      index = c("GEOID", "year"),
                      SDF = California, adaptive = FALSE, p = 2, bigdata = FALSE,
                      effect = "individual", model = "within", approach = "AIC",
                      kernel = "bisquare", longlat = FALSE, doParallel = FALSE)
bw.AIC.Fix

GWPR.light documentation built on June 21, 2022, 5:05 p.m.