CODEWITHSUNDEEP
rspatialspdep
Miranda
Miranda

Reputation: 188

How to estimate spatial SAR model with islands?

I am trying to estimate a spatial SAR model using the lagsarlm command and I get the following error:

Error in lagsarlm(f1, data = df, spatialList, tol.solve = 1e-30) : NAs in lagged dependent variable In addition: Warning message: In lag.listw(listw, y, zero.policy = zero.policy) : NAs in lagged values

Apparently, this happens because I have islands (observations that are not connected to any other observation). If I run the following code without creating an island (by deleting W[1,1:50] <- 0), the code works just fine.

library(spdep)
library(spatialreg)
library(sna)

set.seed(123)

W <- rgraph(50, m=1, tprob=0.1, mode="graph", diag=FALSE)
W[1,1:50] <- 0

spatialList <- mat2listw(W)

y <- rnorm(50)
x <- rnorm(50)

df <- cbind.data.frame(y,x)

f1 <- y ~ x

m2s = lagsarlm(f1, data=df, spatialList, tol.solve=1.0e-30)

This is just a toy example. Given that my real matrix has several islands, any ideas about how to proceed with the estimation?

Many thanks

Upvotes: 1

Views: 856

Answers (1)

Edward
Edward

Reputation: 19169

The lagsarlm function has a zero.policy argument, which is FALSE by default, which means that the function will terminate with an error if there are zeros in the data and you omit this argument. SO, change it to TRUE.

m2s = lagsarlm(f1, data=df, spatialList, tol.solve=1.0e-30, zero.policy=TRUE)
m2s

Call:
lagsarlm(formula = f1, data = df, listw = spatialList, zero.policy = TRUE, 
    tol.solve = 1e-30)
Type: lag 

Coefficients:
         rho  (Intercept)            x 
-0.007533119  0.209226752  0.058315953 

Log likelihood: -71.21548

From the help page:

zero.policy
default NULL, use global option value; if TRUE assign zero to the lagged value of zones without neighbours, if FALSE (default) assign NA - causing lagsarlm() to terminate with an error

Upvotes: 2

Related Questions