## Code author: J. Arbour

phylANOVA.intra<-function (tree, X,ycol,f1col,f2col, nsim = 1000,method="anova") {

## This function is modified from the "phytools" package function phylANOVA (Revell, 2012)
## This function carries out an ANOVA/ANCOVA, by comparing the results to a set of Brownian Motion simulated values (following Garland et al. 1993)
## This function can include more than one individual per species by sampling around the BM simulated species mean based on the standard deviation of the trait within each species
  
# tree must be a phy object  
#X must be a data.frame
#col specifies location of variables in dataframe. ycol is the location of the dependent, f1col and f2col are the two factors or the factor (1) and the covariate (2)
# method must equal "anova" or "ancova"
# nsim is the number of BM simulations
 
  
# returns a list object with...
# 1) $ANOVA: the observed ANOVA/ANCOVA results 
# 2) $sim.test.pvalue, the phylogenetic significance test associated with the two factors and interaction, or (in the case of ANCOVA) factor, covariate and interaction (null hypothesis: ANOVA/ANCOVA results could have occured under BM evolution)
  
  
  require(HH)
  mean.perf<-as.vector(by(X[,ycol],X[,1],mean))
  sd.perf<-as.vector(by(X[,ycol],X[,1],sd))
  names(mean.perf)<-levels(X[,1])
  names(sd.perf)<-levels(X[,1])
  
  ace1<-ace(mean.perf,tree)$ace[1]
  sig2 <- mean(pic(mean.perf, tree)^2)
  
  sims.means <- fastBM(tree,a=ace1, sig2 = sig2, nsim = (nsim - 1),bounds=c(0,Inf))
  sim.all.data<-matrix(nrow=nrow(X),ncol=(nsim-1))
   
  sd.perf<-sd.perf[unique(X[,1])]  
  
  if (method=="anova"){
 
    anova.obs <- summary(aov(X[,ycol] ~ X[,f1col]*X[,f2col],X))  
  
  for (j in 1:(nsim-1)){
  sim.means<-sims.means[,j]
  sim.means<-sim.means[as.character(names(sd.perf))]
  for (i in 1:length(levels(X[,1]))){
    n<-length(X[,1])/length(levels(X[,1]))
    new.perf<-rnorm(n,sim.means[i],sd.perf[i])
    starts<-n*i-(n-1)
    stops<-n*i
  sim.all.data[starts:stops,j]<-new.perf
  }
  
  }
  
  F1.null <- vector()
  F1.null[1] <- anova.obs[[1]]$F[1]
  F2.null <- vector()
  F2.null[1] <- anova.obs[[1]]$F[2]
  FIn.null <- vector()
  FIn.null[1] <- anova.obs[[1]]$F[3]

  
  
  
  for (i in 2:nsim) {
    new.X<-sim.all.data[,(i-1)]
    anova.new <- summary(aov(new.X ~ X[,f1col]*X[,f2col],X))
    F1.null[i] <- anova.new[[1]]$F[1]
    F2.null[i] <- anova.new[[1]]$F[2]
    FIn.null[i] <- anova.new[[1]]$F[3]
      }
  P.F1 <- sum(F1.null >= F1.null[1])/nsim
  P.F2 <- sum(F2.null >= F2.null[1])/nsim
  P.FIn <- sum(FIn.null >= FIn.null[1])/nsim
  Fs<-c(F1.null[1],F2.null[1],FIn.null[1])
  Ps<-c(P.F1,P.F2,P.FIn);names(Ps)<-c("Factor 1", "Factor 2","Interaction")
  
  
  } else{
    obs.data<-cbind(X[,ycol],X[,f1col],X[,f2col])
    obs.data<-as.data.frame(obs.data)
    obs.data[,2]<-as.factor(obs.data[,2])
    colnames(obs.data)<-c("y","indep","cov")
    anova.obs <- summary(ancova(y~indep*cov,obs.data))  
  
  for (j in 1:(nsim-1)){
    sim.means<-sims.means[,j]
    sim.means<-sim.means[as.character(names(sd.perf))]
    for (i in 1:length(levels(X[,1]))){
      n<-length(X[,1])/length(levels(X[,1]))
      new.perf<-rnorm(n,sim.means[i],sd.perf[i])
      starts<-n*i-(n-1)
      stops<-n*i
      sim.all.data[starts:stops,j]<-new.perf
    }
    
  }
  
    F1.null <- vector()
    F1.null[1] <- anova.obs[[1]]$F[1]
    Fcov.null <- vector()
    Fcov.null[1] <- anova.obs[[1]]$F[2]
    FIn.null <- vector()
    FIn.null[1] <- anova.obs[[1]]$F[3]
    


for (i in 2:nsim) {
  new.X<-sim.all.data[,(i-1)]
  new.data<-cbind(new.X,X[,f1col],X[,f2col])
  colnames(new.data)<-c("y","indep","cov")
  new.data<-as.data.frame(new.data)
  new.data[,2]<-as.factor(new.data[,2])
  anova.new <- summary(ancova(y~indep*cov,new.data))
  F1.null[i] <- anova.new[[1]]$F[1]
  Fcov.null[i] <- anova.new[[1]]$F[2]
  FIn.null[i] <- anova.new[[1]]$F[2]
}
P.F1 <- sum(F1.null >= F1.null[1])/nsim
P.Fcov <- sum(Fcov.null >= Fcov.null[1])/nsim
P.FIn<- sum(FIn.null >= FIn.null[1])/nsim

    Fs<-c(F1.null[1],Fcov.null[1],FIn.null[1])
Ps<-c(P.F1,P.Fcov,P.FIn);names(Ps)<-c("Factor", "Covariate","Interaction")
}
  
  
  return(list(ANOVA=anova.obs, sim.test.pvalue=Ps))
  }