[R-sig-Geo] including factors in the formula for spml?

[Ariel Ortiz-Bobea]

Hello,

I'm trying to include a "state-level time trend" in a panel model with
spatially correlated errors. I'm getting the following error when I run the
spml() function with a "factor" as part of the formula:

Error in solve.default(crossprod(xco), crossprod(xco, yco)) : 
Lapack routine dgesv: system is exactly singular

It seems spml doesnt like factors in the formula (they are ok in the pml()
function). Do you have any suggestions on how to do this without having to
de-trend my data before the analysis or creating a bunch of
dummies+interactions "manually" (using the numeric class and not factors)?

Thanks in advance.

Ariel

--- Reproductible code below -----

# Load packages
	library(spdep)
	library(plm)
	library(splm)

# Create simple spatial structure
	data    	<- expand.grid(lon=1:10, lat=1:10 )
	data$fips 	<- 1:dim(data)[1]    
	rownames(data)<- 1:dim(data)[1]
	n<-dim(data)[1]
	mylist 		<- vector('list', n)
	for(i in 1:n) { 
	   mylist[[i]] = Polygons(list(Polygon(
   	   cbind( c(data[i,"lon"]-0.5, data[i,"lon"]-0.5, data[i,"lon"]+0.5,
data[i,"lon"]+0.5, data[i,"lon"]-0.5), 
   	         
c(data[i,"lat"]-0.5,data[i,"lat"]+0.5,data[i,"lat"]+0.5,data[i,"lat"]-0.5,data[i,"lat"]-0.5)))),i) 
	} 
	polys 	<- SpatialPolygons(mylist, 1:n) 
	polys 	<- SpatialPolygonsDataFrame(polys, data) 

# Create neighbor structure and plot it
	coords 	<- coordinates(polys)  
	nb  	<- poly2nb(polys)
	dlist  	<- nbdists(nb, coords) 
	idlist 	<- lapply(dlist, function(x) 1/x) # inverse distance
	nbw    	<- nb2listw(nb, glist=idlist, style="W")  
	plot(polys) 
	plot(nb, coordinates(polys), add=T, col="red")    

# Create artificial dataset for regression
	years		<- 2000:2010     # time periods
	t    		       <- length(years) # number of time periods
	dataset   	       <- data.frame( fips  = polys at data$fips, year=rep(years,
each=n) )
	dataset$x 	<- rnorm(n*t , rep(0,times=n) ) # covariate
	dataset$a 	<- rep(rnorm(n,mean=2,sd=sqrt(3)), times=t) # fixed effect
	dataset$state<- as.factor(floor(dataset$fips/10)) # create a "state" dummy
        head(dataset)

# This is the unknown and true regression parameter to be estimated
	beta = matrix(1, nrow=1, ncol=1)

# Generate error and implied y variable
	e <- rnorm(t*n, rep(0,times=n))
	dataset$y      <- as.matrix(dataset[,c("x")]) %*% beta + dataset$a + e      

# Run FE regression with plm: this works fine...
	fit_plm  <-  plm(formula = y ~ x + state*(as.numeric(year) +
I(as.numeric(year)^2)), data = dataset, model="within", index =
c("fips","year"))
	summary(fit_plm)

# Run FE regression with spatial error: crashes (apack routine dgesv: system
is exactly singular)
	fit_spml <- spml(formula = y ~ x + state*(as.numeric(year) +
I(as.numeric(year)^2)), data = dataset, model="within",
effect=c("individual"), index = c("fips","year"), listw = nbw, lag = FALSE,
spatial.error = "b")







-----
Ariel Ortiz-Bobea
PhD Candidate in Agricultural & Resource Economics
University of Maryland - College Park
--
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/including-factors-in-the-formula-for-spml-tp7579094.html
Sent from the R-sig-geo mailing list archive at Nabble.com.