For W 11/8

Submitted by Professor Shortell on Tue, 11/07/2017 - 20:43

# K-means Clustering

# First, we create the subset of variables for structure analysis
ANES2012ft1<-subset(ANES2012,select=ftgr_xfund:ftgr_feminists)
ANES2012ft2<-subset(ANES2012,select=ftgr_liberals:ftgr_working)
ANES2012ft3<-subset(ANES2012,select=ftgr_gay)
ANES2012ft4<-subset(ANES2012,select=ftgr_rich:ftgr_tea)
ANES2012ft<-cbind(ANES2012ft1,ANES2012ft2,ANES2012ft3,ANES2012ft4)

# Next, we compute the correlation matrix of those variables. This is the input
# for cluster analysis.
ANES2012cor<-cor(ANES2012ft,use="complete.obs",method="pearson")
kmeans(ANES2012cor,centers=2)

# Now, like with our other methods of latent structure analysis,
# we start looking for an optimal clustering solution. We begin
# with two clusters and start adding clusters until we get a good
# solution, with low within groups variation and good between
# groups variation, and, most importantly, interpretable
# clusters, based on membership. We increment the value of centers by 1
# and compare the results. Repeat until you find the optimal solution.
kmeans(ANES2012cor,centers=3)

# We want clusters that are not too big and not too small. We keep incrementing
# the number of clusters until we go one step too far, then back up to the previous
# solution.