Replicating Franzese & Hays

#==============================================================================
# 09-04-2014
# Test script estimation of spatial model
# Code taken from Jamie Monogan:
# http://monogan.myweb.uga.edu/teaching/spatial/fhReplication.R
# Replication material from:
# Franzese and Hays. 2007.
# "Spatial Econometric Models of Cross-Sectional Interdependence in
# Political Science Panel and Time-Series-Cross-Section Data."
# Political Analysis 15:140-164.
#==============================================================================

# Set preferences
options(scipen=10)

# Load data
fh<-read.csv('~/Desktop/almMLire.csv',header=TRUE) # Dataset
adj<-as.matrix(read.csv('~/Desktop/almWeights.csv',header=FALSE)) # Adjacency matrix

# Row-standardize adjacency matrix
noNeigh<-apply(adj,1,sum)
noNeigh[7]<-.01 # Greece has no neighbours, adjust as we can't have 0 in the denominator
wmat<-adj/noNeigh # Spatial weight matrix

# Calculate spatial lag of outcome variable, using Kronekcer product
N<-length(unique(fh$cc))
T<-length(unique(fh$year))
ident<-diag(1,nrow=T) # Identity matrix
bigW<-(ident%x%wmat)
fh$SpatLag<-(ident%x%wmat)%*%fh$lnlmtue # Spatial lag

## REPLICATE TABLE 4

# Column 1. Simple OLS model

m1<-lm(lnlmtue~lnlmtue_1+DENSITY+DEIND+lngdp_pc+UR+TRADE+FDI+LLVOTE+LEFTC+
         TCDEMC+GOVCON+OLDAGE+
         as.factor(cc)+as.factor(year),
              data=fh);summary(m1)

# Gives roughly the same results as in the paper

# Column 2. OLS model including a spatial lag
m2<-lm(lnlmtue~lnlmtue_1+SpatLag+DENSITY+DEIND+lngdp_pc+UR+TRADE+FDI+LLVOTE+LEFTC+
         TCDEMC+GOVCON+OLDAGE+
         as.factor(cc)+as.factor(year),
              data=fh);summary(m2)

# Again roughly the same results

# Column 3. Spatial 2SLS

# Spatially lag all the explanatory variables
fh$l.lnlmtue_1<-(ident%x%wmat)%*%fh$lnlmtue_1
fh$l.DENSITY<-(ident%x%wmat)%*%fh$DENSITY
fh$l.DEIND<-(ident%x%wmat)%*%fh$DEIND
fh$l.lngdp_pc<-(ident%x%wmat)%*%fh$lngdp_pc
fh$l.UR<-(ident%x%wmat)%*%fh$UR
fh$l.TRADE<-(ident%x%wmat)%*%fh$TRADE
fh$l.FDI<-(ident%x%wmat)%*%fh$FDI
fh$l.LLVOTE<-(ident%x%wmat)%*%fh$LLVOTE
fh$l.LEFTC<-(ident%x%wmat)%*%fh$LEFTC
fh$l.TCDEMC<-(ident%x%wmat)%*%fh$TCDEMC
fh$l.GOVCON<-(ident%x%wmat)%*%fh$GOVCON
fh$l.OLDAGE<-(ident%x%wmat)%*%fh$OLDAGE

# First stage regression
stage.1<-lm(SpatLag~l.lnlmtue_1+l.DENSITY+l.DEIND+l.lngdp_pc+l.UR+l.TRADE+l.FDI+l.LLVOTE+
              l.LEFTC+l.TCDEMC+l.GOVCON+l.OLDAGE, data=fh);summary(stage.1)

# Write predicted values to variable
fh$iv.Lag<-stage.1$fitted.values

# Second stage estimation
m3<-lm(lnlmtue~lnlmtue_1+iv.Lag+DENSITY+DEIND+lngdp_pc+UR+TRADE+FDI+LLVOTE+
         LEFTC+TCDEMC+GOVCON+OLDAGE+
         as.factor(cc)+as.factor(year),
              data=fh);summary(m3)

# Slightly different results, this is possibly due to the fact that not all Xs were
# modelled in the first stage. Close enough anyway.

# Column 4. Spatial Maximum Likelihood (GLM model)
require(MASS)
m4<-glm(lnlmtue~lnlmtue_1+SpatLag+DENSITY+DEIND+lngdp_pc+UR+TRADE+FDI+LLVOTE+LEFTC+
         TCDEMC+GOVCON+OLDAGE+
         as.factor(cc)+as.factor(year),family=gaussian,
              data=fh);summary(m4)

# Can see that there is quite some difference in the estimates
# Probably not the right ML estimator

# Try estimating the model with 'spml'
# NB - using FE not possible
require(splm)
m4a<-spml(lnlmtue~lnlmtue_1+DENSITY+DEIND+lngdp_pc+UR+TRADE+FDI+LLVOTE+LEFTC+
         TCDEMC+GOVCON+OLDAGE,data=fh,index=c("cc","year"),listw=mat2listw(wmat),
          model="pooling",spatial.error="none",lag=T);summary(m4a)

# Results are quite different