Post date: Aug 05, 2015 4:20:59 PM
I ran the mantel tests with the categorical distance matrixes for geography. The correlations themselves are very high (0.99 for one pair, 0.89 for the other), but the p-values are both around 0.33. Thus, there are just too few independent data points for significance in a permutation-based test (i.e. the randomization procedure in the mantel test will generate the observed configuration or one that is equivalent too often to reject the null). Incidentally, if you ignore the non-independence and just use a standard test for the correlation coefficient it is significant (not that that matters). So, I think we want the numbers or a figure with a verbal argument.
Here is the R code:
fst<-c(0.032710001,0.1245251858,0.125142929,0.1194868634,0.1201347024,0.0329899261,0.0326450048,0.429231638,0.6931755719,0.4400798212,0.7147612182,0.1409532122)
geo<-c(0,1,1,1,1,0,0,1,1,1,1,0)
x<-matrix(NA,nrow=4,ncol=4)
colnames(x)<-c("bmcg31c","bmcg3q","bmtc","bmtq")
rownames(x)<-colnames(x)
x[lower.tri(x)]<-fst[1:6]
x<-as.matrix(as.dist(x))
y<-matrix(NA,nrow=4,ncol=4)
colnames(y)<-c("bmcg31c","bmcg3q","bmtc","bmtq")
rownames(y)<-colnames(y)
y[lower.tri(y)]<-geo[1:6]
y<-as.matrix(as.dist(y))
## results
mantel.rtest(as.dist(y),as.dist(x),10000)
#Monte-Carlo test
#Observation: 0.9987995
#Call: mantel.rtest(m1 = as.dist(y), m2 = as.dist(x), nrepet = 10000)
#Based on 10000 replicates
#Simulated p-value: 0.3353665
x<-matrix(NA,nrow=4,ncol=4)
colnames(x)<-c("bsc","bsq","hq","hc")
rownames(x)<-colnames(x)
x[lower.tri(x)]<-fst[7:12]
x<-as.matrix(as.dist(x))
y<-matrix(NA,nrow=4,ncol=4)
colnames(y)<-c("bsc","bsq","hq","hc")
rownames(y)<-colnames(y)
y[lower.tri(y)]<-geo[7:12]
y<-as.matrix(as.dist(y))
## results
mantel.rtest(as.dist(y),as.dist(x),10000)
#Monte-Carlo test
#Observation: 0.8931848
#Call: mantelnoneuclid(m1 = m1, m2 = m2, nrepet = nrepet)
#Based on 10000 replicates
#Simulated p-value: 0.3330667