# Nov 17 2017 Revisions ########################################################################## ################################################################################################# ## Passerine only analysis ####################################################################### # Loading data: pass.dat = read.csv("passerines only data.csv", header = T, stringsAsFactors = F) head(pass.dat) str(pass.dat) # Organizing data: pass.dat$Carot <- as.factor(pass.dat$Carot) pass.dat$Author <- as.factor(pass.dat$Author) pass.dat$Species <- as.factor(pass.dat$Species) pass.dat$animal <- as.factor(pass.dat$animal) pass.dat$Category <- as.factor(pass.dat$Category) pass.dat$Color.Variable <- as.factor(pass.dat$Color.Variable) pass.dat$Sex <- as.factor(pass.dat$Sex) # Creating weight for Zr effect size: pass.dat$wi <- pass.dat$Sample.Size - 3 ############################################ ### Covariance matrix among effect sizes ### ############################################ # create square matrix matching N of ES, filled with zeros vc_matrixB <- matrix(0,nrow = dim(pass.dat)[1],ncol = dim(pass.dat)[1]) pass.dat$EffectID<-as.factor(1:length(pass.dat$ID)) # Naming matrix rows and columns by Effect Size ID: rownames(vc_matrixB) <- pass.dat$EffectID colnames(vc_matrixB) <- pass.dat$EffectID # Find start and end coordinates for the subsets based on same study: shared_coord <- which(pass.dat$ExpPop %in% pass.dat$ExpPop[duplicated(pass.dat$ExpPop)]==TRUE) shared_coord # Matrix of combinations of coordinates for each study with shared effect sizes: combinations <- do.call("rbind", tapply(shared_coord, pass.dat[shared_coord,"ExpPop"], function(x) t(combn(x,2)))) combinations # Calculate covariance values between effect sizes from the same study at the positions in shared_list and place them on the matrix. "SE" is the SD of estimate, so that the covariance is r[corr coefficient] * SD * SD. We are assuming that effect sizes from the same study have a correlation of 0.5. If the correlation is 0, then we can model without this matrix. If r = 1, then we should collapse the related data points: for (i in 1:dim(combinations)[1]){ p1 <- combinations[i,1] p2 <- combinations[i,2] p1_p2_cov <- 0.5*pass.dat[p1,"SE"] * pass.dat[p2,"SE"] vc_matrixB[p1,p2] <- p1_p2_cov vc_matrixB[p2,p1] <- p1_p2_cov } # add the diagonal - use "Variance" #created variance-covariance matrix diag(vc_matrixB) <- pass.dat$Variance head(vc_matrixB) is.positive.definite(vc_matrixB) #[1] TRUE # you could save this if you like #save(vc_matrixB,file="vc_matrixB.RData") #load("vc_matrixB.RData") # Prepare for MCMCglmm: AinvGA <- solve(vc_matrixB) AinvGA <- as(AinvGA,"dgCMatrix") str(AinvGA) ## Passerine-only data trees<-read.nexus("12928.tre") to.remove = c(1, 21) pass.tree = lapply(trees, drop.tip, to.remove) save(pass.tree,file="passerine-only-trees.tre") # phylogenetic meta-analysis: invB <- inverseA(pass.tree[[1]], nodes = "TIPS") # Prior, parameter expanded: prior <- list(R=list(V=1, nu=0.002), G=list(G1=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000), G2=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000), G3=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000), G4=list(V=1, fix=1))) # Prior, same as above, but without the random effect for EffectID, in models that don't assume a 0.5 correlation between effect sizes: prior1 <- list(R=list(V=1, nu=0.002), G=list(G1=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000), G2=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000), G3=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000))) prior2 <- list(R=list(V=1, nu=0.002), G=list(G1=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000))) ######################################################## ### Model with overall effect: Assumming effect sizes are correlated ######################################################## mc.start0_pass <- MCMCglmm(Z.transf ~ 1, data = pass.dat, random = ~animal + Species + Author + EffectID, mev = 1/pass.dat$wi, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, pr = T, verbose = F) mphylMCMC0_pass <- mc.start0_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab = mphylMCMC0_pass$Liab[1,], R = mphylMCMC0_pass$VCV[1,1], G = list(G1=mphylMCMC0_pass$VCV[1,2], G2=mphylMCMC0_pass$VCV[1,3], G3=mphylMCMC0_pass$VCV[1,4], G4=mphylMCMC0_pass$VCV[1,5])) mphylMCMC0_pass <- MCMCglmm(Z.transf ~ 1, data = pass.dat, random = ~ animal + Species + Author + EffectID, mev = 1/pass.dat$wi, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, pr = T, verbose = F) if(i>300){ mc.start0_pass$VCV[i-300,]<-mphylMCMC0_pass$VCV[1,] mc.start0_pass$Sol[i-300,]<-mphylMCMC0_pass$Sol[1,] mc.start0_pass$Liab[i-300,]<-mphylMCMC0_pass$Liab[1,] } print(i) } hist(mc.start0_pass$Liab) hist(mc.start0$Liab) hist(mc.start0_pass$VCV[,1]) mean(mc.start0_pass$VCV[,1]) HPDinterval(mc.start0_pass$VCV[,1]) summary(mc.start0_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -472.5074 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.04778 1.992e-11 0.08537 1000 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01822 1.044e-07 0.05483 533.5 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.005617 1.572e-07 0.01828 248.5 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 1 1 1 0 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.002698 0.0002548 0.007996 63.75 # # Location effects: Z.transf ~ 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.1609 -0.0797 0.3686 1000 0.102 s2m.0_pass <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.0_pass <- mc.start0_pass$VCV[,"animal"] + mc.start0_pass$VCV[,"Species"] + mc.start0_pass$VCV[,"Author"] + mc.start0_pass$VCV[,"EffectID"] total.var.0_pass <- mc.start0_pass$VCV[,"animal"] + mc.start0_pass$VCV[,"Species"] + mc.start0_pass$VCV[,"Author"] + mc.start0_pass$VCV[,"EffectID"] + mc.start0_pass$VCV[,"units"] +s2m.0_pass I2.0_pass<-100*(btw.study.var.0_pass)/(total.var.0_pass) posterior.mode(I2.0_pass) # var1 # 98.3291 summary(I2.0_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 98.110564 0.284002 0.008981 0.041884 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 97.41 98.00 98.18 98.32 98.45 HPDinterval(I2.0_pass) # lower upper # var1 97.47192 98.4865 # attr(,"Probability") # [1] 0.95 #=== # MCMCglmm phylogenetic signal = lambda: #=== I2.animal.0_pass <- 100*(mc.start0_pass$VCV[,"animal"]/total.var.0_pass) posterior.mode(I2.animal.0_pass) # var1 # 0.1190537 summary(I2.animal.0_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 2.2494 3.5227 0.1114 0.1432 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 0.0038 0.3960 1.2704 2.6784 11.1902 HPDinterval(I2.animal.0_pass) # lower upper # var1 3.813853e-06 7.882944 # attr(,"Probability") # [1] 0.95 #=== # I2 Species #=== I2.species.0_pass <- 100*(mc.start0_pass$VCV[,"Species"]/total.var.0_pass) posterior.mode(I2.species.0_pass) # var1 # 0.02740433 summary(I2.species.0_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 1.17077 1.43679 0.04544 0.10831 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 0.005361 0.210823 0.688592 1.500106 5.146467 HPDinterval(I2.species.0_pass) # # lower upper # var1 2.137837e-05 4.102612 # attr(,"Probability") # [1] 0.95 #=== # I2 Study #=== I2.study.0_pass <- 100*(mc.start0_pass$VCV[,"Author"]/total.var.0_pass) posterior.mode(I2.study.0_pass) # var1 # 0.01111063 summary(I2.study.0_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 0.68734 0.78384 0.02479 0.08675 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 0.0009968 0.0977300 0.3976548 0.9745582 2.8146062 HPDinterval(I2.study.0_pass) # lower upper # var1 4.86142e-06 2.356475 # attr(,"Probability") # [1] 0.95 ######################################################################## ### Egger regression & trim-and-fill (overall effect): ######################################################################## prior.egger <- list(R=list(V=1, nu=0.002), G=list(G1=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000), G2=list(V=1, nu=1, alpha.mu = 0, alpha.V = 1000))) mc0egger_pass <- MCMCglmm(Z.transf ~ 1, data = pass.dat, random = ~ Species + Author, mev = 1/pass.dat$wi, prior = prior.egger, nitt = 1000000, burnin = 900000, thin = 100, pl = T, pr = T, verbose = F) summary(mc0egger_pass) save(mc0egger_pass, file = "mc0egger.pass.Rdata") # Iterations = 900001:999901 # Thinning interval = 100 # Sample size = 1000 # # DIC: -118.6474 # # G-structure: ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01609 3.658e-08 0.05113 823.6 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03884 0.01136 0.06987 1000 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.01801 0.005877 0.02997 1000 # # Location effects: Z.transf ~ 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.18003 0.07693 0.26999 1000 0.006 ** # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 model<-mc0egger_pass model$Random$formula<-update(model$Random$formula, ~.+leg(mev, -1, FALSE):units) Pred<-predict(model, marginal=~leg(mev, -1, FALSE):units) Prec<-1/sqrt(pass.dat$Variance) # Precision = 1/sqrt(V) es<-pass.dat$Z.transf-Pred #save(es,file="newResiduals_pass.RData") Data1<-data.frame(ES=es,Prec=Prec) # meta-analysis on residual - we should get meta-analytic mean of 0 mmm2<-rma(yi= es, sei= 1/Prec, method="REML", data=Data1) summary(mmm2) # Random-Effects Model (k = 191; tau^2 estimator: REML) # # logLik deviance AIC BIC AICc # 26.1897 -52.3794 -48.3794 -41.8854 -48.3152 # # tau^2 (estimated amount of total heterogeneity): 0.0066 (SE = 0.0025) # tau (square root of estimated tau^2 value): 0.0813 # I^2 (total heterogeneity / total variability): 28.43% # H^2 (total variability / sampling variability): 1.40 # # Test for Heterogeneity: # Q(df = 190) = 291.8215, p-val < .0001 # # Model Results: # # estimate se zval pval ci.lb ci.ub # -0.0032 0.0125 -0.2568 0.7973 -0.0276 0.0212 par(mfrow=c(1,2)) funnel(mmm2) funnel(mmm2,yaxis="seinv") # We will do publication bias tests for mc.start0 w/o the covariances: mmmr2<-regtest(mmm2,model="lm") mmmr2 # Regression Test for Funnel Plot Asymmetry # # model: weighted regression with multiplicative dispersion # predictor: standard error # # test for funnel plot asymmetry: t = 1.9804, df = 189, p = 0.0491 mmmt2<-trimfill(mmm2,estimator="L0") summary(mmmt2) # Estimated number of missing studies on the left side: 26 (SE = 8.9767) # # Random-Effects Model (k = 217; tau^2 estimator: REML) # # logLik deviance AIC BIC AICc # -7.5537 15.1074 19.1074 25.8580 19.1638 # # tau^2 (estimated amount of total heterogeneity): 0.0218 (SE = 0.0045) # tau (square root of estimated tau^2 value): 0.1475 # I^2 (total heterogeneity / total variability): 54.91% # H^2 (total variability / sampling variability): 2.22 # # Test for Heterogeneity: # Q(df = 216) = 416.5559, p-val < .0001 # # Model Results: # # estimate se zval pval ci.lb ci.ub # -0.0413 0.0152 -2.7154 0.0066 -0.0711 -0.0115 ** ###################################################################################################### ### Model with converted vs. dietary carotenoids (ASSUMING r = 0.5 among ES): ###################################################################################################### mc.start.3_pass <- MCMCglmm(Z.transf~Carot - 1, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.3_pass) mphylMCMC.3_pass <- mc.start.3_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.3_pass$Liab[1,], R =mphylMCMC.3_pass$VCV[1,1], G = list(G1=mphylMCMC.3_pass$VCV[1,2], G2=mphylMCMC.3_pass$VCV[1,3], G3=mphylMCMC.3_pass$VCV[1,4], G4=mphylMCMC.3_pass$VCV[1,5])) mphylMCMC.3_pass <- MCMCglmm(Z.transf~Carot - 1, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.3_pass$VCV[i-300,]<-mphylMCMC.3_pass$VCV[1,] mc.start.3_pass$Sol[i-300,]<-mphylMCMC.3_pass$Sol[1,] mc.start.3_pass$Liab[i-300,]<-mphylMCMC.3_pass$Liab[1,] } print(i) } summary(mc.start.3_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -5.568277 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.01882 3.244e-08 0.06819 277.1 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.0122 4.389e-09 0.04896 264 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03217 6.233e-08 0.0712 8.031 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3445 0.05129 1 1.964 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05531 0.03223 0.0782 4.608 # # Location effects: Z.transf ~ Carot - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CarotConverted 0.26321 0.06223 0.49189 1000 0.022 * # CarotDietary 0.08957 -0.12355 0.29253 1108 0.346 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 s2m.3_pass <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.3_pass <- mc.start.3_pass$VCV[,"animal"] + mc.start.3_pass$VCV[,"Species"] + mc.start.3_pass$VCV[,"Author"] + mc.start.3_pass$VCV[,"EffectID"] total.var.3_pass <- mc.start.3_pass$VCV[,"animal"] + mc.start.3_pass$VCV[,"Species"] + mc.start.3_pass$VCV[,"Author"] + mc.start.3_pass$VCV[,"EffectID"] + mc.start.3_pass$VCV[,"units"] +s2m.3_pass I2.3_pass<-100*(btw.study.var.3_pass)/(total.var.3_pass) posterior.mode(I2.3_pass) # var1 # 94.85841 summary(I2.3_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 72.433 15.652 0.495 10.262 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 51.96 60.44 65.83 94.44 95.59 HPDinterval(I2.3_pass) # lower upper # var1 53.54631 95.85842 # attr(,"Probability") # [1] 0.95 ##################################################################################### ################## Getting difference between Converted and Dietary ################ ################################################################################### contrasts(pass.dat$Carot) <- contr.treatment(levels(pass.dat$Carot), base = 1) mc.start.3_pass.cont <- MCMCglmm(Z.transf~Carot, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.3_pass.cont) mphylMCMC.3_pass.cont <- mc.start.3_pass.cont for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.3_pass.cont$Liab[1,], R =mphylMCMC.3_pass.cont$VCV[1,1], G = list(G1=mphylMCMC.3_pass.cont$VCV[1,2], G2=mphylMCMC.3_pass.cont$VCV[1,3], G3=mphylMCMC.3_pass.cont$VCV[1,4], G4=mphylMCMC.3_pass.cont$VCV[1,5])) mphylMCMC.3_pass.cont <- MCMCglmm(Z.transf~Carot, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.3_pass.cont$VCV[i-300,]<-mphylMCMC.3_pass.cont$VCV[1,] mc.start.3_pass.cont$Sol[i-300,]<-mphylMCMC.3_pass.cont$Sol[1,] mc.start.3_pass.cont$Liab[i-300,]<-mphylMCMC.3_pass.cont$Liab[1,] } print(i) } summary(mc.start.3_pass.cont) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: 52.03296 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.02374 4.665e-10 0.1112 68.22 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01337 2.293e-08 0.05222 340.1 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.04792 0.01789 0.08355 538.7 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3445 0.05162 1 2.01 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05879 0.0402 0.07761 11.41 # # Location effects: Z.transf ~ Carot # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.26209 0.04523 0.54868 1426.6 0.036 * # CarotDietary -0.15798 -0.38890 0.09700 845.2 0.162 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ########################################################################################### ### Overall Model with Category (ASSUMING r = 0.5 BETWEEN ES): ########################################################################################### mc.start.7_pass <- MCMCglmm(Z.transf~ Category - 1, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID = AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.7_pass) mphylMCMC.7_pass <- mc.start.7_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.7_pass$Liab[1,], R =mphylMCMC.7_pass$VCV[1,1], G = list(G1=mphylMCMC.7_pass$VCV[1,2], G2=mphylMCMC.7_pass$VCV[1,3], G3=mphylMCMC.7_pass$VCV[1,4], G4=mphylMCMC.7_pass$VCV[1,5])) mphylMCMC.7_pass <- MCMCglmm(Z.transf~Category - 1, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.7_pass$VCV[i-300,]<-mphylMCMC.7_pass$VCV[1,] mc.start.7_pass$Sol[i-300,]<-mphylMCMC.7_pass$Sol[1,] mc.start.7_pass$Liab[i-300,]<-mphylMCMC.7_pass$Liab[1,] } print(i) } summary(mc.start.7_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -14.11343 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.03287 3.425e-08 0.114 670 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01396 3.992e-07 0.04937 530.8 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03035 3.132e-10 0.0649 6.816 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3436 0.05077 1 2 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05315 0.02544 0.07312 4.122 # # Location effects: Z.transf ~ Category - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CategoryCondition 0.06451 -0.21849 0.35237 611.9 0.540 # CategoryImmune 0.10154 -0.21075 0.36287 1000.0 0.398 # CategoryParasite 0.24330 -0.07472 0.51021 901.8 0.096 . # CategoryParental/Reproduction 0.22311 -0.09988 0.46221 1000.0 0.122 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 s2m.7_pass <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.7_pass <- mc.start.7_pass$VCV[,"animal"] + mc.start.7_pass$VCV[,"Species"] + mc.start.7_pass$VCV[,"Author"] + mc.start.7_pass$VCV[,"EffectID"] total.var.7_pass <- mc.start.7_pass$VCV[,"animal"] + mc.start.7_pass$VCV[,"Species"] + mc.start.7_pass$VCV[,"Author"] + mc.start.7_pass$VCV[,"EffectID"] + mc.start.7_pass$VCV[,"units"] +s2m.7_pass I2.7_pass<-100*(btw.study.var.7_pass)/(total.var.7_pass) posterior.mode(I2.7_pass) # var1 # 94.36402 summary(I2.7_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 94.39660 0.95044 0.03006 0.05183 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 93.02 93.99 94.43 94.88 95.70 HPDinterval(I2.7_pass) # lower upper # var1 93.12494 95.74877 # attr(,"Probability") # [1] 0.95 ########################################################################################### ### Model with Category and type of Carotenoid (Assuming r = 0.5 between ESs): ########################################################################################### mc.start.9_pass <- MCMCglmm(Z.transf~ Category:Carot -1, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.9_pass) mphylMCMC.9_pass <- mc.start.9_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.9_pass$Liab[1,], R =mphylMCMC.9_pass$VCV[1,1], G = list(G1=mphylMCMC.9_pass$VCV[1,2], G2=mphylMCMC.9_pass$VCV[1,3], G3=mphylMCMC.9_pass$VCV[1,4], G4=mphylMCMC.9_pass$VCV[1,5])) mphylMCMC.9_pass <- MCMCglmm(Z.transf~Category:Carot - 1, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.9_pass$VCV[i-300,]<-mphylMCMC.9_pass$VCV[1,] mc.start.9_pass$Sol[i-300,]<-mphylMCMC.9_pass$Sol[1,] mc.start.9_pass$Liab[i-300,]<-mphylMCMC.9_pass$Liab[1,] } print(i) } summary(mc.start.9_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -26.44002 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.02248 9.183e-09 0.08914 346.2 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01093 4.278e-09 0.03915 279.8 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.02162 1.964e-07 0.05081 9.971 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3418 0.04901 1 1.994 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05092 0.02593 0.07172 2.176 # # Location effects: Z.transf ~ Category:Carot - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CategoryCondition:CarotConverted 0.084007 -0.158547 0.358236 1000 0.446 # CategoryImmune:CarotConverted 0.097922 -0.185954 0.367045 1000 0.436 # CategoryParasite:CarotConverted 0.435114 0.174715 0.688995 1000 0.006 ** # CategoryParental/Reproduction:CarotConverted 0.337485 0.021148 0.618882 1000 0.034 * # CategoryCondition:CarotDietary 0.106965 -0.186198 0.387852 1158 0.398 # CategoryImmune:CarotDietary 0.113537 -0.135304 0.371149 1000 0.330 # CategoryParasite:CarotDietary 0.009203 -0.241867 0.283083 1000 0.920 # CategoryParental/Reproduction:CarotDietary 0.095383 -0.182517 0.311780 1000 0.394 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 s2m.9_pass <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.9_pass <- mc.start.9_pass$VCV[,"animal"] + mc.start.9_pass$VCV[,"Species"] + mc.start.9_pass$VCV[,"Author"] + mc.start.9_pass$VCV[,"EffectID"] total.var.9_pass <- mc.start.9_pass$VCV[,"animal"] + mc.start.9_pass$VCV[,"Species"] + mc.start.9_pass$VCV[,"Author"] + mc.start.9_pass$VCV[,"EffectID"] + mc.start.9_pass$VCV[,"units"] +s2m.9_pass I2.9_pass<-100*(btw.study.var.9_pass)/(total.var.9_pass) posterior.mode(I2.9_pass) # var1 # 94.44083 summary(I2.9_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 72.4043 15.7622 0.4984 8.6347 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 49.86 59.90 67.22 93.76 95.21 HPDinterval(I2.9_pass) # lower upper # var1 52.19162 95.6872 # attr(,"Probability") # [1] 0.9 ########################################################################################### ### Model with Category and type of Carotenoid (getting contrasts. Assuming r = 0.5 between ESs): ########################################################################################### contrasts(pass.dat$Carot) <- contr.treatment(levels(pass.dat$Carot), base = 1) pass.dat$Category <- relevel(pass.dat$Category, ref = "Parental/Reproduction") mc.start.10_pass <- MCMCglmm(Z.transf~ Category*Carot , data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.10_pass) mphylMCMC.10_pass <- mc.start.10_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.10_pass$Liab[1,], R =mphylMCMC.10_pass$VCV[1,1], G = list(G1=mphylMCMC.10_pass$VCV[1,2], G2=mphylMCMC.10_pass$VCV[1,3], G3=mphylMCMC.10_pass$VCV[1,4], G4=mphylMCMC.10_pass$VCV[1,5])) mphylMCMC.10_pass <- MCMCglmm(Z.transf~Category*Carot, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.10_pass$VCV[i-300,]<-mphylMCMC.10_pass$VCV[1,] mc.start.10_pass$Sol[i-300,]<-mphylMCMC.10_pass$Sol[1,] mc.start.10_pass$Liab[i-300,]<-mphylMCMC.10_pass$Liab[1,] } print(i) } summary(mc.start.10_pass) ############## Parasite Contrast Converted vs Dietary ################################ # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: 38.56197 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.02853 7.2e-10 0.1094 222.3 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01436 2.642e-11 0.04804 437.9 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03286 0.01114 0.06415 497.5 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3427 0.04863 1 1.989 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.056 0.03762 0.07449 13.31 # # Location effects: Z.transf ~ Category * Carot # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.44854 0.21269 0.74759 1000.0 0.004 ** # CategoryImmune -0.36236 -0.55014 -0.17089 587.7 <0.001 *** # CategoryCondition -0.35844 -0.52029 -0.16418 907.0 <0.001 *** # CategoryParental/Reproduction -0.10354 -0.35971 0.14681 1000.0 0.454 # CarotDietary -0.42016 -0.69247 -0.16157 1000.0 0.006 ** # CategoryImmune:CarotDietary 0.46604 0.19554 0.72888 1000.0 <0.001 *** # CategoryCondition:CarotDietary 0.45638 0.20110 0.74210 1000.0 0.004 ** # CategoryParental/Reproduction:CarotDietary 0.19616 -0.09604 0.53328 1000.0 0.202 --- ######################################################################################### ############## Parental/Reproduction Contrast Converted vs Dietary ################################ # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: 38.21046 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.02826 6.802e-09 0.1232 240.3 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01658 1.672e-08 0.05429 247.4 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03184 0.008112 0.06106 424.8 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3428 0.04961 1 1.964 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.0554 0.0362 0.07428 32.03 # # Location effects: Z.transf ~ Category * Carot # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.329958 0.029001 0.705035 1000.0 0.074 . # CategoryParasite 0.106467 -0.132606 0.384893 1000.0 0.422 # CategoryImmune -0.260137 -0.507756 0.013516 1000.0 0.062 . # CategoryCondition -0.251666 -0.496756 0.009437 1000.0 0.060 . # CarotDietary -0.227279 -0.536816 0.153791 1000.0 0.188 # CategoryParasite:CarotDietary -0.202335 -0.560660 0.077459 1000.0 0.214 # CategoryImmune:CarotDietary 0.270533 -0.063736 0.640302 1000.0 0.136 # CategoryCondition:CarotDietary 0.254847 -0.054931 0.648269 895.8 0.154 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ######################################################################################### ########### Species Effect Sizes With Only Passerines ################################ ############ ############################### ######################################################################################## mc.start.14_pass <- MCMCglmm(Z.transf~ Species - 1, data = pass.dat, random = ~ Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior2, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) mphylMCMC.14_pass <- mc.start.14_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.14_pass$Liab[1,], R =mphylMCMC.14_pass$VCV[1,1], G = list(G1=mphylMCMC.14_pass$VCV[1,2])) mphylMCMC.14_pass <- MCMCglmm(Z.transf~ Species - 1, data = pass.dat, random = ~Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior2, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.14_pass$VCV[i-300,]<-mphylMCMC.14_pass$VCV[1,] mc.start.14_pass$Sol[i-300,]<-mphylMCMC.14_pass$Sol[1,] mc.start.14_pass$Liab[i-300,]<-mphylMCMC.14_pass$Liab[1,] } print(i) } summary(mc.start.14_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -105.0481 # # G-structure: ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03491 0.009886 0.06648 98.4 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.01769 0.006112 0.02804 309.7 # # Location effects: Z.transf ~ Species - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # SpeciesAmerican Goldfinch 0.97737 0.44929 1.45653 1000.0 0.002 ** # SpeciesBlue Tit 0.19733 -0.06987 0.41935 897.1 0.130 # SpeciesCirl Bunting -0.28475 -0.82989 0.22441 1153.1 0.286 # SpeciesCommon Redpoll -0.36304 -0.90784 0.16795 1000.0 0.182 # SpeciesCommon Rosefinch 0.38901 -0.11061 0.93370 1525.9 0.136 # SpeciesCommon Yellowthroat 0.07632 -0.14659 0.31485 1000.0 0.508 # SpeciesEuropean Greenfinch 0.12989 -0.05268 0.29426 1000.0 0.170 # SpeciesEuropean Serin 0.41891 -0.04887 0.95989 1000.0 0.100 # SpeciesGolden-Collared Manakin 0.21609 -0.23979 0.65068 1102.5 0.340 # SpeciesGolden-crowned Kinglet 0.12185 -0.24497 0.51106 1707.4 0.540 # SpeciesGreat Tit 0.13318 0.02762 0.24797 1000.0 0.022 * # SpeciesHouse Finch 0.38645 0.22058 0.55782 1000.0 <0.001 *** # SpeciesKentucky Warbler 0.04108 -0.41631 0.53945 1000.0 0.876 # SpeciesLinnet 0.12288 -0.14729 0.40706 1000.0 0.392 # SpeciesNorthern Cardinal 0.27164 0.09546 0.48522 899.4 0.010 * # SpeciesRed-winged Blackbird 0.14362 -0.32086 0.60549 1000.0 0.526 # SpeciesRed Fody 0.05935 -0.33474 0.50830 1000.0 0.810 # SpeciesSouthern Red Bishop 0.34237 -0.09536 0.80193 1000.0 0.120 # SpeciesYellowhammer -0.05809 -0.26685 0.16384 1000.0 0.610 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ############################################################################################### ############################################################################################## ################### parasite lumped with immune data ######################################## ############################################################################################# Para.Data = read.csv("parasite lump immune data.csv", header = T, stringsAsFactors = F) head(Para.Data) str(Para.Data) # Organizing data: Para.Data$Carot <- as.factor(Para.Data$Carot) Para.Data$Author <- as.factor(Para.Data$Author) Para.Data$Species <- as.factor(Para.Data$Species) Para.Data$animal <- as.factor(Para.Data$animal) Para.Data$Category <- as.factor(Para.Data$Category) Para.Data$Color.Variable <- as.factor(Para.Data$Color.Variable) Para.Data$Sex <- as.factor(Para.Data$Sex) # Creating weight for Zr effect size: Para.Data$wi <- Para.Data$Sample.Size - 3 ############################################ ### Covariance matrix among effect sizes ### ############################################ # create square matrix matching N of ES, filled with zeros vc_matrixA <- matrix(0,nrow = dim(Para.Data)[1],ncol = dim(Para.Data)[1]) Para.Data$EffectID<-as.factor(1:length(Para.Data$ID)) # Naming matrix rows and columns by Effect Size ID: rownames(vc_matrixA) <- Para.Data$EffectID colnames(vc_matrixA) <- Para.Data$EffectID # Find start and end coordinates for the subsets based on same study: shared_coord <- which(Para.Data$ExpPop %in% Para.Data$ExpPop[duplicated(Para.Data$ExpPop)]==TRUE) shared_coord # Matrix of combinations of coordinates for each study with shared effect sizes: combinations <- do.call("rbind", tapply(shared_coord, Para.Data[shared_coord,"ExpPop"], function(x) t(combn(x,2)))) combinations # Calculate covariance values between effect sizes from the same study at the positions in shared_list and place them on the matrix. "SE" is the SD of estimate, so that the covariance is r[corr coefficient] * SD * SD. We are assuming that effect sizes from the same study have a correlation of 0.5. If the correlation is 0, then we can model without this matrix. If r = 1, then we should collapse the related data points: for (i in 1:dim(combinations)[1]){ p1 <- combinations[i,1] p2 <- combinations[i,2] p1_p2_cov <- 0.5*Para.Data[p1,"SE"] * Para.Data[p2,"SE"] vc_matrixA[p1,p2] <- p1_p2_cov vc_matrixA[p2,p1] <- p1_p2_cov } # add the diagonal - use "Variance" #created variance-covariance matrix diag(vc_matrixA) <- Para.Data$Variance head(vc_matrixA) is.positive.definite(vc_matrixA) #[1] TRUE # you could save this if you like save(vc_matrixA,file="vc_matrixA.RData") load("vc_matrixA.RData") # Prepare for MCMCglmm: AinvGA <- solve(vc_matrixA) AinvGA <- as(AinvGA,"dgCMatrix") str(AinvGA) #################################################################################################### ################# Parasite lumped with Immune category, but Oxidative measures separate Cat ##################################################################################################### mc.start.9.para <- MCMCglmm(Z.transf~ Category:Carot - 1, data = Para.Data, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invA$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.9.para) mphylMCMC.9.para <- mc.start.9.para for(i in 1:length(pass.tree)){ invA <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.9.para$Liab[1,], R =mphylMCMC.9.para$VCV[1,1], G = list(G1=mphylMCMC.9.para$VCV[1,2], G2=mphylMCMC.9.para$VCV[1,3], G3=mphylMCMC.9.para$VCV[1,4], G4=mphylMCMC.9.para$VCV[1,5])) mphylMCMC.9.para <- MCMCglmm(Z.transf~Category:Carot - 1, data = Para.Data, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invA$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.9.para$VCV[i-300,]<-mphylMCMC.9.para$VCV[1,] mc.start.9.para$Sol[i-300,]<-mphylMCMC.9.para$Sol[1,] mc.start.9.para$Liab[i-300,]<-mphylMCMC.9.para$Liab[1,] } print(i) } summary(mc.start.9.para) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -18.35145 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.01512 1.351e-09 0.06073 356.6 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.008523 2.961e-09 0.033 525.7 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.0292 4.521e-08 0.06379 12.57 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3438 0.0506 1 1.998 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05344 0.02533 0.07455 3.763 # # Location effects: Z.transf ~ Category:Carot - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CategoryCondition:CarotConverted 0.07799 -0.13839 0.34085 889.5 0.464 # CategoryImmune:CarotConverted 0.31881 0.09434 0.55508 1000.0 0.012 * # CategoryOxidative:CarotConverted 0.08194 -0.53372 0.75162 1000.0 0.824 # CategoryParental/Reproduction:CarotConverted 0.35361 0.09172 0.66097 1000.0 0.018 * # CategoryCondition:CarotDietary 0.09672 -0.19545 0.33885 1000.0 0.378 # CategoryImmune:CarotDietary 0.02012 -0.19559 0.26759 592.5 0.864 # CategoryOxidative:CarotDietary 0.23630 -0.10725 0.52630 1000.0 0.130 # CategoryParental/Reproduction:CarotDietary 0.10334 -0.11561 0.36005 790.8 0.306 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ########################################################################################## ###################### Immune Split from Oxidative ############################ Ox.Data = read.csv("Ox split immune data.csv", header = T, stringsAsFactors = F) head(Ox.Data) str(Ox.Data) # Organizing data: Ox.Data$Carot <- as.factor(Ox.Data$Carot) Ox.Data$Author <- as.factor(Ox.Data$Author) Ox.Data$Species <- as.factor(Ox.Data$Species) Ox.Data$animal <- as.factor(Ox.Data$animal) Ox.Data$Category <- as.factor(Ox.Data$Category) Ox.Data$Color.Variable <- as.factor(Ox.Data$Color.Variable) Ox.Data$Sex <- as.factor(Ox.Data$Sex) # Creating weight for Zr effect size: Ox.Data$wi <- Ox.Data$Sample.Size - 3 ############################################ ### Covariance matrix among effect sizes ### ############################################ # create square matrix matching N of ES, filled with zeros vc_matrixA.2 <- matrix(0,nrow = dim(Ox.Data)[1],ncol = dim(Ox.Data)[1]) Ox.Data$EffectID<-as.factor(1:length(Ox.Data$ID)) # Naming matrix rows and columns by Effect Size ID: rownames(vc_matrixA.2) <- Ox.Data$EffectID colnames(vc_matrixA.2) <- Ox.Data$EffectID # Find start and end coordinates for the subsets based on same study: shared_coord <- which(Ox.Data$ExpPop %in% Ox.Data$ExpPop[duplicated(Ox.Data$ExpPop)]==TRUE) shared_coord # Matrix of combinations of coordinates for each study with shared effect sizes: combinations <- do.call("rbind", tapply(shared_coord, Ox.Data[shared_coord,"ExpPop"], function(x) t(combn(x,2)))) combinations # Calculate covariance values between effect sizes from the same study at the positions in shared_list and place them on the matrix. "SE" is the SD of estimate, so that the covariance is r[corr coefficient] * SD * SD. We are assuming that effect sizes from the same study have a correlation of 0.5. If the correlation is 0, then we can model without this matrix. If r = 1, then we should collapse the related data points: for (i in 1:dim(combinations)[1]){ p1 <- combinations[i,1] p2 <- combinations[i,2] p1_p2_cov <- 0.5*Ox.Data[p1,"SE"] * Ox.Data[p2,"SE"] vc_matrixA.2[p1,p2] <- p1_p2_cov vc_matrixA.2[p2,p1] <- p1_p2_cov } # add the diagonal - use "Variance" #created variance-covariance matrix diag(vc_matrixA.2) <- Ox.Data$Variance head(vc_matrixA.2) is.positive.definite(vc_matrixA.2) #[1] TRUE # you could save this if you like save(vc_matrixA.2,file="vc_matrixA.2.RData") load("vc_matrixA.2.RData") # Prepare for MCMCglmm: AinvGA <- solve(vc_matrixA.2) AinvGA <- as(AinvGA,"dgCMatrix") str(AinvGA) ############################################################################################## ############ Model with parasite, oxidative, and immune separated. Passerines Only, effect sizes correlated ############################################################################################# mc.start.9_ox <- MCMCglmm(Z.transf~ Category:Carot - 1, data = Ox.Data, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invA$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.9_ox) mphylMCMC.9_ox <- mc.start.9_ox for(i in 1:length(pass.tree)){ invA <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.9_ox$Liab[1,], R =mphylMCMC.9_ox$VCV[1,1], G = list(G1=mphylMCMC.9_ox$VCV[1,2], G2=mphylMCMC.9_ox$VCV[1,3], G3=mphylMCMC.9_ox$VCV[1,4], G4=mphylMCMC.9_ox$VCV[1,5])) mphylMCMC.9_ox <- MCMCglmm(Z.transf~Category:Carot - 1, data = Ox.Data, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invA$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.9_ox$VCV[i-300,]<-mphylMCMC.9_ox$VCV[1,] mc.start.9_ox$Sol[i-300,]<-mphylMCMC.9_ox$Sol[1,] mc.start.9_ox$Liab[i-300,]<-mphylMCMC.9_ox$Liab[1,] } print(i) } summary(mc.start.9_ox) # # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -22.95713 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.06377 8.729e-08 0.1003 1000 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01156 9.432e-10 0.0466 297.9 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.02025 5.035e-08 0.04669 30.18 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3429 0.04911 1 2.017 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05183 0.02527 0.07335 3.226 # # Location effects: Z.transf ~ Category:Carot - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CategoryCondition:CarotConverted 0.077725 -0.238385 0.347682 1000.0 0.562 # CategoryImmune:CarotConverted 0.095474 -0.205007 0.420169 1000.0 0.478 # CategoryOxidative:CarotConverted 0.107655 -0.515774 0.724251 1000.0 0.748 # CategoryParasite:CarotConverted 0.437939 0.173215 0.750596 1000.0 0.014 * # CategoryParental/Reproduction:CarotConverted 0.333549 0.011287 0.647277 1000.0 0.048 * # CategoryCondition:CarotDietary 0.101476 -0.196321 0.397650 1000.0 0.458 # CategoryImmune:CarotDietary 0.044895 -0.285937 0.359493 1000.0 0.748 # CategoryOxidative:CarotDietary 0.221911 -0.116100 0.592585 1000.0 0.182 # CategoryParasite:CarotDietary 0.002566 -0.256272 0.297976 900.6 0.994 # CategoryParental/Reproduction:CarotDietary 0.089241 -0.210574 0.357996 1000.0 0.438 ################################################################################# ############################## The effect of sex on effect size estimates ###### #################################################################################### contrasts(pass.dat$Carot) <- contr.treatment(levels(pass.dat$Carot), base = 2) pass.dat$Sex <- relevel(pass.dat$Sex, ref = "M") mc.start.20_pass <- MCMCglmm(Z.transf~ Carot * Sex, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.20_pass) mphylMCMC.20_pass <- mc.start.20_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.20_pass$Liab[1,], R =mphylMCMC.20_pass$VCV[1,1], G = list(G1=mphylMCMC.20_pass$VCV[1,2], G2=mphylMCMC.20_pass$VCV[1,3], G3=mphylMCMC.20_pass$VCV[1,4], G4=mphylMCMC.20_pass$VCV[1,5])) mphylMCMC.20_pass <- MCMCglmm(Z.transf~ Carot * Sex, data = pass.dat, random = ~animal + Species + Author + EffectID, ginverse = list(animal=invB$Ainv, EffectID=AinvGA), prior = prior, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.20_pass$VCV[i-300,]<-mphylMCMC.20_pass$VCV[1,] mc.start.20_pass$Sol[i-300,]<-mphylMCMC.20_pass$Sol[1,] mc.start.20_pass$Liab[i-300,]<-mphylMCMC.20_pass$Liab[1,] } print(i) } summary(mc.start.20_pass) ######################################################################### ######################## Converted Male as baseline #################### # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: 43.24995 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.0224 2.922e-09 0.07746 677.3 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01296 1.851e-08 0.04502 274.1 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.05834 0.0216 0.09951 750.8 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3436 0.05083 1 1.992 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05663 0.03499 0.07424 20.81 # # Location effects: Z.transf ~ Carot * Sex # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.25113 0.03249 0.52189 880.8 0.044 * # CarotDietary -0.10435 -0.33548 0.16522 1000.0 0.392 # SexBoth 0.08471 -0.51019 0.73094 960.5 0.800 # SexF 0.10103 -0.07846 0.29449 1000.0 0.276 # CarotDietary:SexBoth -0.25792 -0.87708 0.40633 937.8 0.410 # CarotDietary:SexF -0.08750 -0.33620 0.13440 1000.0 0.464 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ######################################################################################## ######################### Dietary Male as baseline ################################### # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: 47.07863 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.09698 1.388e-07 0.08465 1000 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01149 1.232e-08 0.04135 1000 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.05703 0.02265 0.09864 747.5 # # ~EffectID # # post.mean l-95% CI u-95% CI eff.samp # EffectID 0.3436 0.05015 1 1.982 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.05749 0.03748 0.07593 24.93 # # Location effects: Z.transf ~ Carot * Sex # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.152945 -0.120056 0.399400 1000.0 0.226 # CarotConverted 0.100703 -0.126370 0.395352 1000.0 0.406 # SexF 0.008555 -0.140245 0.170171 847.8 0.934 # SexBoth -0.172165 -0.391642 0.059954 815.5 0.154 # CarotConverted:SexF 0.086097 -0.176962 0.306320 842.7 0.486 # CarotConverted:SexBoth 0.260407 -0.310638 1.039737 677.3 0.402 ############################################################################################ ############################################################################################ ################### Creating Figures #################################################### ############################################################################################ require(ggplot2) gdat = read.csv("ggplotdat.pass.only.csv", header =T, stringsAsFactors=F) gdat$invar<- (1/(1/(gdat$N-3))) gdat$Category<- factor(gdat$Category, levels=c("Overall","Condition","Immune","Parasite","Reproduction")) gdat$Carot<- factor(gdat$Carot, levels=c("Dietary","Converted","Combined")) ggplot(gdat, aes(x=Carot))+ geom_linerange(aes( ymin = LCL, ymax = UCL))+ geom_point(aes(y=Estimate,color=factor(Carot), size= invar),shape =15) + scale_color_manual(values=c("Yellow2","Red2","Black")) + scale_size_continuous(range= c(4,20)) + xlab("")+ ylab(expression(Effect~Size~italic((Zr)))) + theme_bw()+ geom_hline(aes(yintercept = 0),linetype = "dashed", lwd=1)+ facet_grid(Category~.)+ coord_flip()+ theme(strip.text = element_text(size = 20), axis.text = element_text(size=16), axis.title = element_text(size=17)) + theme_bw() + theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(), axis.line = element_line(colour = "black")) ########################################################################## ############### Plot of Species by Effect Size ######################### ######################################################################### head(Data) require(tidyverse) Data$animal = as.character(Data$animal) for(i in 1:nrow(Data)){ x = strsplit(Data$animal[i], split = "_") Data$Genus[i] = x[[1]][1] } # plot species colored by genus Data %>% group_by(Carot, Genus) %>% count(Species) %>% arrange(desc(Carot), Genus, Species) %>% ungroup() %>% mutate(ord = c(1:length(Species)), sp = factor(ord, labels = Species)) %>% ggplot(aes(x = sp, y = n, fill = Genus)) + geom_bar(stat = "identity") + scale_fill_viridis(discrete = T)+ theme_bw() + ylab("Number of effect sizes") + xlab("Species") + theme(axis.text.x = element_text(size = 12, angle = 90, vjust = 0.5, hjust =1), axis.text.y = element_text(size = 12), axis.title = element_text(size = 14), panel.grid.major = element_blank(), panel.grid.minor = element_blank(), axis.line = element_line(colour = "black")) # plot data colored by carot type Data %>% group_by(Carot, Genus) %>% count(Species) %>% arrange(desc(Carot), Genus, Species) %>% ungroup() %>% mutate(ord = c(1:length(Species)), sp = factor(ord, labels = Species)) %>% ggplot(aes(x = sp, y = n, fill = Carot)) + geom_bar(stat = "identity") + scale_fill_manual(values = c("red2", "yellow2"), name = "Carotenoid Type") + theme_bw() + ylab("Number of effect sizes") + xlab("Species") + theme(axis.text.x = element_text(size = 12, angle = 90, vjust = 0.5, hjust =1), axis.text.y = element_text(size = 12), axis.title = element_text(size = 14), panel.grid.major = element_blank(), panel.grid.minor = element_blank(), axis.line = element_line(colour = "black")) ############################# ### Plot concensus tree ##### ############################ contree1 <- consensus(pass.tree, p=0.8) plot(contree1) contree2 <- consensus(pass.tree, p=1) plot(contree2) contree3 <- consensus(pass.tree, p=0.5) plot(contree3) contree4 <- consensus(pass.tree) tiff(file = "phylotree.pass.tiff", units = "in", width = 4, height = 11, res = 300) plot(contree4) dev.off() ######################################################## ### plot number of effect sizes by species colored by carot type and in Phlyo tree order ####################################################### Data %>% group_by(Phylo.Graph, Carot) %>% count(Species) %>% arrange(Phylo.Graph, Carot) %>% ungroup() %>% mutate(ord = c(1:length(Species)), sp = factor(ord, labels = Species)) %>% ggplot(aes(x = sp, y = n, fill = Carot)) + geom_bar(stat = "identity") + scale_fill_manual(values = c("red2", "yellow2"), name = "Carotenoid Type") + theme_bw() + ylab("Number of effect sizes") + xlab("Species") + theme(axis.text.x = element_text(size = 12, angle = 90, vjust = 0.5, hjust =1), axis.text.y = element_text(size = 12), axis.title = element_text(size = 14), panel.grid.major = element_blank(), panel.grid.minor = element_blank(), axis.line = element_line(colour = "black")) ########################################################################################## ####### plot effect size for each species colored by carot type and in Phlyo tree order ########################################################################################## sdat = read.csv("SpeciesPlotData.csv", header =T, stringsAsFactors=F) sdat$invar<- (1/(1/(sdat$N-3))) head(sdat) sdat$Carot<- factor(sdat$Carot, levels=c("Dietary","Converted")) sdat %>% mutate(sp = factor(Phylo.Ord, labels = Species)) %>% ggplot(aes(x=sp))+ geom_linerange(aes( ymin = lcl, ymax = ucl))+ geom_point(aes(y=Zr,color=factor(Carot), size= invar),shape =15) + scale_size_continuous(range = c(4,15)) + scale_color_manual(values = c("red2", "yellow2"), name = "Carotenoid Type") + theme_bw() + ylab("Effect Size (Zr)") + xlab("Species") + geom_hline(aes(yintercept = 0),linetype = "dashed", lwd=1)+ theme(axis.text.x = element_text(size = 12, angle = 90, vjust = 0.5, hjust =1), axis.text.y = element_text(size = 12), axis.title = element_text(size = 14), panel.grid.major = element_blank(), panel.grid.minor = element_blank(), axis.line = element_line(colour = "black")) # plot genus Data %>% group_by(Carot, Genus) %>% count(Species) %>% arrange(desc(Carot), Genus, Species) %>% ungroup() %>% mutate(ord = c(1:length(Species)), sp = factor(ord, labels = Species)) %>% ggplot(aes(x = sp, y = n, fill = Genus)) + geom_bar(stat = "identity") + scale_fill_viridis(discrete = T)+ theme_bw() + ylab("Number of effect sizes") + xlab("Species") + theme(axis.text.x = element_text(size = 12, angle = 90, vjust = 0.5, hjust =1), axis.text.y = element_text(size = 12), axis.title = element_text(size = 14), panel.grid.major = element_blank(), panel.grid.minor = element_blank(), axis.line = element_line(colour = "black")) ################################################### ####### Counting Species per Category ############ require(tidyverse) ca = unique(Data$Carot) ct = unique(Data$Category) res = data.frame() for(i in 1:length(ca)){ for(j in 1:length(ct)){ d = subset(Data, Carot == ca[i] & Category == ct[j]) sp = length(unique(d$Species)) df = data.frame(Carot = ca[i], Category = ct[j], n.sp = sp) res = rbind(res, df) } } ######################################################################################## ############# Models assuming effect sizes are Not correlated ######################### ####################################################################################### ######################################################## ### Model with overall effect ######################################################## mc.start1_pass <- MCMCglmm(Z.transf ~ 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invB$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 1000, burnin = 0, thin = 1, pl = T, pr = T, verbose = F) mphylMCMC1_pass <- mc.start1_pass for(i in 1:length(pass.tree)){ invB <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab = mphylMCMC1_pass$Liab[1,], R = mphylMCMC1_pass$VCV[1,1], G = list(G1=mphylMCMC1_pass$VCV[1,2], G2=mphylMCMC1_pass$VCV[1,3], G3=mphylMCMC1_pass$VCV[1,4])) mphylMCMC1_pass <- MCMCglmm(Z.transf ~ 1, data = pass.dat, random = ~ animal + Species + Author, ginverse = list(animal=invB$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, pr = T, verbose = F) if(i>300){ mc.start1_pass$VCV[i-300,]<-mphylMCMC1_pass$VCV[1,] mc.start1_pass$Sol[i-300,]<-mphylMCMC1_pass$Sol[1,] mc.start1_pass$Liab[i-300,]<-mphylMCMC1_pass$Liab[1,] } print(i) } summary(mc.start1_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -117.2541 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.02351 7.899e-09 0.0842 334.3 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01195 1.42e-07 0.04808 214.1 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03612 0.01051 0.06471 413.5 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.01769 0.006754 0.029 204.1 # # Location effects: Z.transf ~ 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # (Intercept) 0.17848 -0.03808 0.41417 1000 0.088 . # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 s2m.1_pass <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.1_pass <- mc.start1_pass$VCV[,"animal"] + mc.start1_pass$VCV[,"Species"] + mc.start1_pass$VCV[,"Author"] total.var.1_pass <- mc.start1_pass$VCV[,"animal"] + mc.start1_pass$VCV[,"Species"] + mc.start1_pass$VCV[,"Author"] + mc.start1_pass$VCV[,"units"] +s2m.1_pass I2.1_pass<-100*(btw.study.var.1_pass)/(total.var.1_pass) posterior.mode(I2.1_pass) # var1 # 66.86848 summary(I2.1_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 64.6738 10.1364 0.3205 0.5265 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 45.19 57.51 64.95 71.62 83.55 HPDinterval(I2.1_pass) # lower upper # var1 45.15061 83.40505 # attr(,"Probability") # [1] 0.95 #=== # MCMCglmm phylogenetic signal = lambda: #=== I2.animal.1_pass <- 100*(mc.start1_pass$VCV[,"animal"]/total.var.1_pass) posterior.mode(I2.animal.1_pass) # var1 # 0.3711767 summary(I2.animal.1_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 17.9163 17.1573 0.5426 1.3608 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 0.07335 3.78286 12.19177 28.73383 57.83835 HPDinterval(I2.animal.1_pass) # lower upper # var1 8.61676e-06 51.47404 # attr(,"Probability") # [1] 0.95 #=== # I2 Species #=== I2.species.1_pass <- 100*(mc.start1_pass$VCV[,"Species"]/total.var.1_pass) posterior.mode(I2.species.1_pass) # var1 # 0.2003761 summary(I2.species.1_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 10.4886 12.5085 0.3956 0.9289 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 0.02395 1.31876 5.75227 14.81536 45.34363 HPDinterval(I2.species.1_pass) # # lower upper # var1 0.0001385823 38.23944 # attr(,"Probability") # [1] 0.95 #=== # I2 Study #=== I2.study.1_pass <- 100*(mc.start1_pass$VCV[,"Author"]/total.var.1_pass) posterior.mode(I2.study.1_pass) # var1 # 35.36642 summary(I2.study.1_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 36.2689 14.0311 0.4437 1.0621 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 10.68 26.09 35.53 45.77 64.14 HPDinterval(I2.study.1_pass) # lower upper # var1 9.624955 62.44123 # attr(,"Probability") # [1] 0.95 ############################################################## ### Model with converted vs. dietary carotenoids: ############################################################## mc.start.2_pass <- MCMCglmm(Z.transf~Carot - 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) summary(mc.start.2_pass) mphylMCMC.2_pass <- mc.start.2_pass for(i in 1:length(pass.tree)){ invA <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.2_pass$Liab[1,], R =mphylMCMC.2_pass$VCV[1,1], G = list(G1=mphylMCMC.2_pass$VCV[1,2], G2=mphylMCMC.2_pass$VCV[1,3], G3=mphylMCMC.2_pass$VCV[1,4])) mphylMCMC.2_pass <- MCMCglmm(Z.transf~Carot - 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.2_pass$VCV[i-300,]<-mphylMCMC.2_pass$VCV[1,] mc.start.2_pass$Sol[i-300,]<-mphylMCMC.2_pass$Sol[1,] mc.start.2_pass$Liab[i-300,]<-mphylMCMC.2_pass$Liab[1,] } print(i) } summary(mc.start.2_pass) # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -109.96 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.1025 6.074e-08 0.05956 1000 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01206 1.904e-09 0.04814 166.8 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03411 0.009806 0.06372 411.8 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.01811 0.007003 0.02996 357.3 # # Location effects: Z.transf ~ Carot - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CarotConverted 0.25526 0.03010 0.50640 1095 0.046 * # CarotDietary 0.07788 -0.13410 0.31822 1000 0.362 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 #=== # MCMCglmm I^2: #=== s2m <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.2_pass <- mc.start.2_pass$VCV[,"animal"] + mc.start.2_pass$VCV[,"Species"] + mc.start.2_pass$VCV[,"Author"] total.var.2_pass <- mc.start.2_pass$VCV[,"animal"] + mc.start.2_pass$VCV[,"Species"] + mc.start.2_pass$VCV[,"Author"] + mc.start.2_pass$VCV[,"units"] + s2m I2.2_pass<-100*(btw.study.var.2_pass)/(total.var.2_pass) posterior.mode(I2.2_pass) # var1 # 63.04606 summary(I2.2_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 61.2651 11.9555 0.3781 0.7218 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 37.45 53.60 61.66 69.39 82.41 HPDinterval(I2.2_pass) # lower upper # var1 37.45729 82.68403 # attr(,"Probability") # [1] 0.95 ########################################################################################### ### Overall Model with Category: ########################################################################################### mc.start.6_pass <- MCMCglmm(Z.transf~ Category - 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 1000, burnin = 0, thin = 1, pl = T, verbose = F) mphylMCMC.6_pass <- mc.start.6_pass for(i in 1:length(pass.tree)){ invA <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.6_pass$Liab[1,], R =mphylMCMC.6_pass$VCV[1,1], G = list(G1=mphylMCMC.6_pass$VCV[1,2], G2=mphylMCMC.6_pass$VCV[1,3], G3=mphylMCMC.6_pass$VCV[1,4])) mphylMCMC.6_pass <- MCMCglmm(Z.transf~Category - 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 1000, burnin = 999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.6_pass$VCV[i-300,]<-mphylMCMC.6_pass$VCV[1,] mc.start.6_pass$Sol[i-300,]<-mphylMCMC.6_pass$Sol[1,] mc.start.6_pass$Liab[i-300,]<-mphylMCMC.6_pass$Liab[1,] } print(i) } hist(mc.start.6_pass$Liab) hist(mc.start.6_pass$VCV[,1]) mean(mc.start.6_pass$VCV[,1]) HPDinterval(mc.start.6_pass$VCV[,1]) summary(mc.start.6_pass) # # Iterations = 1:1000 # Thinning interval = 1 # Sample size = 1000 # # DIC: -129.733 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.0263 5.547e-06 0.09806 422.8 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01226 1.478e-08 0.04532 297.4 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.03488 0.01044 0.06292 478.1 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.01648 0.007 0.02906 318.2 # # Location effects: Z.transf ~ Category - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CategoryCondition 0.06893 -0.14799 0.32779 873.6 0.482 # CategoryImmune 0.14185 -0.07739 0.40637 916.5 0.170 # CategoryParasite 0.23727 -0.02004 0.46697 884.6 0.060 . # CategoryParental/Reproduction 0.22525 -0.02229 0.45101 907.6 0.066 . #=== # MCMCglmm I^2: #=== s2m <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.6_pass <- mc.start.6_pass$VCV[,"animal"] + mc.start.6_pass$VCV[,"Species"] + mc.start.6_pass$VCV[,"Author"] total.var.6_pass <- mc.start.6_pass$VCV[,"animal"] + mc.start.6_pass$VCV[,"Species"] + mc.start.6_pass$VCV[,"Author"] + mc.start.6_pass$VCV[,"units"] + s2m I2.6_pass<-100*(btw.study.var.6_pass)/(total.var.6_pass) posterior.mode(I2.6_pass) # var1 # 66.65274 summary(I2.6_pass) # Iterations = 1:1000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 1000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 66.2701 10.1267 0.3202 0.4995 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 44.86 59.47 66.58 73.02 84.61 HPDinterval(I2.6_pass) # lower upper # var1 48.14886 86.66289 # attr(,"Probability") # [1] 0.95 ########################################################################################### ### Model with Category and type of Carotenoid: ########################################################################################### mc.start.8_pass <- MCMCglmm(Z.transf~ Category:Carot - 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 3000, burnin = 0, thin = 1, pl = T, verbose = F) mphylMCMC.8_pass <- mc.start.8_pass for(i in 1:length(pass.tree)){ invA <- inverseA(pass.tree[[i]], nodes = "TIPS") start <- list(Liab=mphylMCMC.8_pass$Liab[1,], R =mphylMCMC.8_pass$VCV[1,1], G = list(G1=mphylMCMC.8_pass$VCV[1,2], G2=mphylMCMC.8_pass$VCV[1,3], G3=mphylMCMC.8_pass$VCV[1,4])) mphylMCMC.8_pass <- MCMCglmm(Z.transf~Category:Carot - 1, data = pass.dat, random = ~animal + Species + Author, ginverse = list(animal=invA$Ainv), mev = 1/pass.dat$wi, prior = prior1, nitt = 3000, burnin = 2999, thin = 1, start = start, pl = T, verbose = F) if(i>300){ mc.start.8_pass$VCV[i-300,]<-mphylMCMC.8_pass$VCV[1,] mc.start.8_pass$Sol[i-300,]<-mphylMCMC.8_pass$Sol[1,] mc.start.8_pass$Liab[i-300,]<-mphylMCMC.8_pass$Liab[1,] } print(i) } summary(mc.start.8_pass) # Iterations = 1:3000 # Thinning interval = 1 # Sample size = 3000 # # DIC: -118.6808 # # G-structure: ~animal # # post.mean l-95% CI u-95% CI eff.samp # animal 0.02708 5.882e-10 0.09942 89.6 # # ~Species # # post.mean l-95% CI u-95% CI eff.samp # Species 0.01003 2.295e-10 0.038 291.8 # # ~Author # # post.mean l-95% CI u-95% CI eff.samp # Author 0.02106 0.005048 0.04171 494 # # R-structure: ~units # # post.mean l-95% CI u-95% CI eff.samp # units 0.01757 0.006983 0.029 181.5 # # Location effects: Z.transf ~ Category:Carot - 1 # # post.mean l-95% CI u-95% CI eff.samp pMCMC # CategoryCondition:CarotConverted 0.10364 -0.16058 0.39035 3512 0.373 # CategoryImmune:CarotConverted 0.13343 -0.16150 0.41359 3000 0.291 # CategoryParasite:CarotConverted 0.42263 0.13460 0.68851 2821 0.008 ** # CategoryParental/Reproduction:CarotConverted 0.36777 0.06685 0.67678 2957 0.028 * # CategoryCondition:CarotDietary 0.10604 -0.16292 0.38526 3000 0.380 # CategoryImmune:CarotDietary 0.12906 -0.15362 0.38447 2901 0.267 # CategoryParasite:CarotDietary 0.02994 -0.25576 0.27598 3006 0.789 # CategoryParental/Reproduction:CarotDietary 0.08649 -0.14654 0.36497 3363 0.428 # --- # Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 hist(mc.start.8_pass$Liab) hist(mc.start.8_pass$VCV[,1]) mean(mc.start.8_pass$VCV[,1]) HPDinterval(mc.start.8_pass$VCV[,1]) #=== # MCMCglmm I^2: #=== s2m <- sum(pass.dat$wi*(length(pass.dat$wi)-1))/(sum(pass.dat$wi)^2-sum(pass.dat$wi^2)) btw.study.var.8_pass <- mc.start.8_pass$VCV[,"animal"] + mc.start.8_pass$VCV[,"Species"] + mc.start.8_pass$VCV[,"Author"] total.var.8_pass <- mc.start.8_pass$VCV[,"animal"] + mc.start.8_pass$VCV[,"Species"] + mc.start.8_pass$VCV[,"Author"] + mc.start.8_pass$VCV[,"units"] + s2m I2.8_pass<-100*(btw.study.var.8_pass)/(total.var.8_pass) posterior.mode(I2.8_pass) # var1 # 56.94887 summary(I2.8_pass) # Iterations = 1:3000 # Thinning interval = 1 # Number of chains = 1 # Sample size per chain = 3000 # # 1. Empirical mean and standard deviation for each variable, # plus standard error of the mean: # # Mean SD Naive SE Time-series SE # 57.906 14.568 0.266 1.630 # # 2. Quantiles for each variable: # # 2.5% 25% 50% 75% 97.5% # 28.03 48.03 58.07 68.58 84.94 HPDinterval(I2.8_pass) # lower upper # var1 30.29974 86.32168 # attr(,"Probability") # [1] 0.95