[R-sig-Geo] GLM model with spatialspillover on categorical variables

Lena Fehlhaber @turm|echen @end|ng |rom gm@||@com
Fri Jun 5 07:14:03 CEST 2020


I did a regression analysis with categorical data with a glm model
approach, which worked fine. I have longitude and latitude coordinates for
each observation and I want to add their geographic spillover effect to the
model.

My sample data is structured:

Index DV IVI IVII IVIII IVIV Long Lat
 1  0  2  1  3  -12  -17.8  12
 2  0  1  1  6  112  11  -122
 3  1  3  6  1  91  57  53

with regression eq. DV ~ IVI + IVII + IVIII + IVIV

That mentioned, I assume that the nearer regions are, the more it may
influence my dependant variable. I found several approaches for spatial
regression models, but not for categorical data. I tried to use existing
libraries and functions, such as spdep's lagsarlm, glmmfields, spatialreg,
gstat, geoRglm and many more (I used this list as a reference:
https://cran.r-project.org/web/views/Spatial.html ). For numeric values, I
am able to do spatial regression, but for categorical values, I struggle.
The data structure is the following:

library(dplyr)
data <- data %>%
  mutate(
    DV = as.factor(DV),
    IVI = as.factor(IVI),
    IVII = as.factor(IVII),
    IVIII = as.factor(IVIII),
    IVIV = as.numeric(IVIV),
    longitude = as.numeric(longitude),
    latitude = as.numeric(latitude)
  )

My dependant variable (0|1) as well as my independant variables are
categorical and it would be no use to transform them, of course. I want to
have an other glm model in the end, but with spatial spillover effects
included. The libraries I tested so far can't handle categorical data. Any
leads/ideas would be greatly appreciated.

Thanks a lot.

	[[alternative HTML version deleted]]



More information about the R-sig-Geo mailing list