source(paste0(getwd(),"/functions/resali.R")) # A la place faire un boxplot par sample size avec une boite par méthode boxplot(as.numeric(res.dat.article$typeIerror)~as.numeric(res.dat.article$N), ylim=c(0,1),xlab="N",ylab="Type-I error",col="#ff7777",pch=3) abline(h=0.05,col="red",lty=2,lwd=2) ########################## # ICC / CCC BASE ########################## plot.tam.2 <- function(x, items=1:x$nitems, type="expected", low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#f7c611', '#61b5b7', '#ca346d', '#204776' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(low, high, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) graphics::plot(theta, expScore[,i],,col=12, type="l", lwd=3, las=1, ylab="Score", xlab="Ability", main=paste("Expected Scores Curve - Item ", colnames(tamobj$resp)[i] ) , ylim=ylim2, ... ) } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab="", col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2,las=2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } # get all the function names of the given package "mypack" r <- unclass(lsf.str(envir = asNamespace("TAM"), all = T)) # create functions in the Global Env. with the same name for(name in r) eval(parse(text=paste0(name, '<-TAM:::', name))) #### CCC zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/NoDIF/N100/scenario_2A_100.csv") zaz <- zaz[zaz$replication==1,] zaza <- tam.mml(resp = zaz[,paste0("item",1:4)]) CurlyBraces <- function(x0, x1, y0, y1, pos = 1, direction = 1, depth = 1) { a=c(1,2,3,48,50) # set flexion point for spline b=c(0,.2,.28,.7,.8) # set depth for spline flexion point curve = spline(a, b, n = 50, method = "natural")$y * depth curve = c(curve,rev(curve)) if (pos == 1){ a_sequence = seq(x0,x1,length=100) b_sequence = seq(y0,y1,length=100) } if (pos == 2){ b_sequence = seq(x0,x1,length=100) a_sequence = seq(y0,y1,length=100) } # direction if(direction==1) a_sequence = a_sequence+curve if(direction==2) a_sequence = a_sequence-curve # pos if(pos==1) lines(a_sequence,b_sequence, lwd=1.5, xpd=NA) # vertical if(pos==2) lines(b_sequence,a_sequence, lwd=1.5, xpd=NA) # horizontal } # CCC de base pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/ccc_base.pdf') par(xpd=F,mar=c(6.1,5.1,7.6,2.1)) plot.tam.2(zaza,type = "items",export=F,ylab="Probability of response",main=NULL,package = "graphics",items = 3) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[7,]$xsi,y0=0,y1=1.1,lty=3) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[8,]$xsi,y0=0,y1=1.1,lty=3) segments(x0=zaza$xsi[9,]$xsi,x1=zaza$xsi[9,]$xsi,y0=0,y1=1.1,lty=3) text(x=-2.5,y=0.85,"0",col="#f7c611") text(x=-0.65,y=0.55,"1",col="#61b5b7") text(x=0.95,y=0.55,"2",col="#ca346d") text(x=2.5,y=0.85,"3",col="#204776") par(xpd=T,mar=c(5.1,4.1,4.1,2.1)) segments(x0=-3,x1=zaza$xsi[7,]$xsi,y0=1.1,y1=1.1,col="#f7c611",lwd=2) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[8,]$xsi,y0=1.1,y1=1.1,col="#61b5b7",lwd=2) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[9,]$xsi,y0=1.1,y1=1.1,col="#ca346d",lwd=2) segments(x0=zaza$xsi[9,]$xsi,x1=3,y0=1.1,y1=1.1,col="#204776",lwd=2) points(x = zaza$xsi[7,]$xsi, y=1.1,pch=9,cex=1) points(x = zaza$xsi[8,]$xsi, y=1.1,pch=9,cex=1) points(x = zaza$xsi[9,]$xsi, y=1.1,pch=9,cex=1) text( x=mean(c(-3,zaza$xsi[7,]$xsi)), y=1.2,"0",col="#f7c611" ) text( x=mean(c(zaza$xsi[7,]$xsi,zaza$xsi[8,]$xsi)), y=1.2,"1",col="#61b5b7" ) text( x=mean(c(zaza$xsi[8,]$xsi,zaza$xsi[9,]$xsi)), y=1.2,"2",col="#ca346d" ) text( x=mean(c(zaza$xsi[9,]$xsi,3)), y=1.2,"3",col="#204776" ) text( x=mean(c(-3,zaza$xsi[7,]$xsi)),cex=0.7, y=1.15,"Much less than usual",col="#f7c611" ) text( x=mean(c(zaza$xsi[7,]$xsi,zaza$xsi[8,]$xsi)),cex=0.7, y=1.15,"Less so than usual",col="#61b5b7" ) text( x=mean(c(zaza$xsi[8,]$xsi,zaza$xsi[9,]$xsi)),cex=0.7, y=1.15,"As much as usual",col="#ca346d" ) text( x=mean(c(zaza$xsi[9,]$xsi,3)),cex=0.7, y=1.15,"More so than usual",col="#204776" ) text(x=zaza$xsi[7,]$xsi,y=1.15,expression(delta["j,1"])) text(x=zaza$xsi[8,]$xsi,y=1.15,expression(delta["j,2"])) text(x=zaza$xsi[9,]$xsi,y=1.15,expression(delta["j,3"])) text(x = 0,y=1.3,"Most probable response category") CurlyBraces(x0=-2.5, x1=2.5, y0=1.225, y1=1.225, pos = 2, direction = 1, depth=0.05) arrows(x0=-2.65,x1=-3,y0=-0.35,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.35,length = 0.15,lwd = 2) text(x=-2.55,y=-0.35,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.35,"Better\nmental\nhealth",adj=1) rect(xleft = 3,xright=5,ybottom = 0,ytop=1.1,col = "white",border = "white") lines(x=c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(0,-0.15),lty=3) lines(x=c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(0,-0.15),lty=3) lines(x=c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(0,-0.15),lty=3) title(main='Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() #### ICC BASE zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/NoDIF/N300/scenario_2A_300.csv") zaza <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza2 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) zaza3 <- tam.mml(resp = zaz[,paste0("item",1:4)]) icc.tam.base <- function(x, items=1:x$nitems, type="expected",TT=F, low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#204776', '#204776', '#204776', '#204776' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(-3, 3, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) if (TT==0) { graphics::plot(theta, expScore[,i], type="l", lwd=3, las=1, ylab="Expected item response", main=NULL,xlab="", col="#204776",ylim=ylim2,xlim= ... ) } else { if (TT==1) { lines(theta,expScore[,i], type="l", lwd=3, las=1,col="#204776") } else { lines(theta,expScore[,i], type="l", lwd=2, las=1,col="#204776",lty=2) } } } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } #graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab="", col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2,las=2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) #graphics::lines( dfr1a$Theta, dfr1a$P, #col=pall[kk], type="l", ... #) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] #graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/icc_base.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.base(zaza,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=0) axis(1) axis(2,las=2) title(main='Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) arrows(x0=-2.65,x1=-3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.45,length = 0.15,lwd = 2) text(x=-2.55,y=-0.45,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.45,"Better\nmental\nhealth",adj=1) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() ########################## # CCC DIF ########################## #### CCC zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/NoDIF/N100/scenario_2A_100.csv") zaz <- zaz[zaz$replication==1,] zaza <- tam.mml(resp = zaz[,paste0("item",1:4)]) # CCC DIF # base pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/ccc_dif_1_homo.pdf') par(xpd=F,mar=c(6.1,5.1,7.6,2.1)) plot.tam.2(zaza,type = "items",export=F,ylab="Probability of response in control group (X=0)",main=NULL,package = "graphics",items = 3) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[7,]$xsi,y0=0,y1=1.1,lty=3) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[8,]$xsi,y0=0,y1=1.1,lty=3) segments(x0=zaza$xsi[9,]$xsi,x1=zaza$xsi[9,]$xsi,y0=0,y1=1.1,lty=3) text(x=-2.5,y=0.85,"0",col="#f7c611") text(x=-0.65,y=0.55,"1",col="#61b5b7") text(x=0.95,y=0.55,"2",col="#ca346d") text(x=2.5,y=0.85,"3",col="#204776") par(xpd=T,mar=c(5.1,4.1,4.1,2.1)) segments(x0=-3,x1=zaza$xsi[7,]$xsi,y0=1.1,y1=1.1,col="#f7c611",lwd=2) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[8,]$xsi,y0=1.1,y1=1.1,col="#61b5b7",lwd=2) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[9,]$xsi,y0=1.1,y1=1.1,col="#ca346d",lwd=2) segments(x0=zaza$xsi[9,]$xsi,x1=3,y0=1.1,y1=1.1,col="#204776",lwd=2) points(x = zaza$xsi[7,]$xsi, y=1.1,pch=9,cex=1) points(x = zaza$xsi[8,]$xsi, y=1.1,pch=9,cex=1) points(x = zaza$xsi[9,]$xsi, y=1.1,pch=9,cex=1) text( x=mean(c(-3,zaza$xsi[7,]$xsi)), y=1.15,"0",col="#f7c611" ) text( x=mean(c(zaza$xsi[7,]$xsi,zaza$xsi[8,]$xsi)), y=1.15,"1",col="#61b5b7" ) text( x=mean(c(zaza$xsi[8,]$xsi,zaza$xsi[9,]$xsi)), y=1.15,"2",col="#ca346d" ) text( x=mean(c(zaza$xsi[9,]$xsi,3)), y=1.15,"3",col="#204776" ) text(x=zaza$xsi[7,]$xsi,y=1.15,expression(delta["j,1"])) text(x=zaza$xsi[8,]$xsi,y=1.15,expression(delta["j,2"])) text(x=zaza$xsi[9,]$xsi,y=1.15,expression(delta["j,3"])) text(x = 0,y=1.275,"Most probable response category") CurlyBraces(x0=-2.5, x1=2.5, y0=1.2, y1=1.2, pos = 2, direction = 1, depth=0.05) arrows(x0=-2.65,x1=-3,y0=-0.35,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.35,length = 0.15,lwd = 2) text(x=-2.55,y=-0.35,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.35,"Better\nmental\nhealth",adj=1) rect(xleft = 3,xright=5,ybottom = 0,ytop=1.1,col = "white",border = "white") lines(x=c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(0,-0.275),lty=3) lines(x=c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(-0.4,-0.6),lty=3) lines(x=c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(0,-0.275),lty=3) lines(x=c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(-0.4,-0.6),lty=3) lines(x=c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(0,-0.6),lty=3) title(xlab=expression("Latent variable " * theta * "\n (mental health)"),line=1.8) title(main='A - Homogeneous DIF', font.main=2) par(xpd=F) dev.off() # DIF homogène plot.tam.dif <- function(x, items=1:x$nitems, type="expected", low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") pall <- c('#f7c611', '#61b5b7', '#ca346d', '#204776' ) low <- low-2 high <- high+2 time1 <- NULL if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(low, high, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) graphics::plot(theta, expScore[,i],,col=12, type="l", lwd=3, las=1, ylab="Score", xlab="Ability", main=paste("Expected Scores Curve - Item ", colnames(tamobj$resp)[i] ) , ylim=ylim2, ... ) } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( ifelse(dfr1a$Theta+0.5>-3,dfr1a$Theta+0.5,NA), dfr1a$P, ylim=c(-.1,1.1),xlim=c(-3,3), xlab="", col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2,las=2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( ifelse(dfr1a$Theta+0.5>-3,dfr1a$Theta+0.5,NA), dfr1a$P,xlim=c(-3,3), col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( ifelse(dfr1a$Theta+0.5>-3,dfr1a$Theta+0.5,NA), dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/ccc_dif_2_homo.pdf') par(xpd=F,mar=c(12.6,5.1,1.1,2.1)) plot.tam.dif(zaza,type = "items",export=F,ylab="Probability of response in treatment group (X=1)",main=NULL,package = "graphics",items = 3) text(x=-2,y=0.85,"0",col="#f7c611") text(x=-.05,y=0.55,"1",col="#61b5b7") text(x=1.45,y=0.55,"2",col="#ca346d") text(x=2.5,y=0.7,"3",col="#204776") par(xpd=T,mar=c(5.1,4.1,4.1,2.1)) segments(x0=-3,x1=.5+zaza$xsi[7,]$xsi,y0=-0.5,col="#f7c611",lwd=2) segments(x0=.5+zaza$xsi[7,]$xsi,x1=.5+zaza$xsi[8,]$xsi,y0=-0.5,col="#61b5b7",lwd=2) segments(x0=.5+zaza$xsi[8,]$xsi,x1=.5+zaza$xsi[9,]$xsi,y0=-0.5,col="#ca346d",lwd=2) segments(x0=.5+zaza$xsi[9,]$xsi,x1=3,y0=-0.5,col="#204776",lwd=2) points(x =.5 + zaza$xsi[7,]$xsi, y=-0.5,pch=9,cex=1) points(x =.5 + zaza$xsi[8,]$xsi, y=-0.5,pch=9,cex=1) points(x =.5 + zaza$xsi[9,]$xsi, y=-0.5,pch=9,cex=1) text( x=mean(c(-2.5,zaza$xsi[7,]$xsi)), y=-0.55,"0",col="#f7c611" ) text( x=0.5+mean(c(zaza$xsi[7,]$xsi,zaza$xsi[8,]$xsi)), y=-0.55,"1",col="#61b5b7" ) text( x=0.5+mean(c(zaza$xsi[8,]$xsi,zaza$xsi[9,]$xsi)), y=-0.55,"2",col="#ca346d" ) text( x=0.5+mean(c(zaza$xsi[9,]$xsi,2.5)), y=-0.55,"3",col="#204776" ) text(x=0.5+zaza$xsi[7,]$xsi,y=-0.55,expression(delta["j,1"])) text(x=0.5+zaza$xsi[8,]$xsi,y=-0.55,expression(delta["j,2"])) text(x=0.5+zaza$xsi[9,]$xsi,y=-0.55,expression(delta["j,3"])) arrows(x0=zaza$xsi[7,]$xsi+0.05,x1=zaza$xsi[7,]$xsi+0.5-0.05,y0=0.625,length = 0.1, lwd = 2) arrows(x0=zaza$xsi[8,]$xsi+0.05,x1=zaza$xsi[8,]$xsi+0.5-0.05,y0=0.625,length = 0.1, lwd = 2) arrows(x0=zaza$xsi[9,]$xsi+0.05,x1=zaza$xsi[9,]$xsi+0.5-0.05,y0=0.625,length = 0.1, lwd = 2) text(x=0.25+zaza$xsi[7,]$xsi,y=0.675,"DIF") text(x=0.25+zaza$xsi[8,]$xsi,y=0.675,"DIF") text(x=0.25+zaza$xsi[9,]$xsi,y=0.675,"DIF") text(x = 0.25,y=-0.675,"Most probable response category") CurlyBraces(x0=-2.25, x1=2.75, y0=-0.6, y1=-0.6, pos = 2, direction = 2, depth=0.05) arrows(x0=-2.65,x1=-3,y0=-0.35,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.35,length = 0.15,lwd = 2) text(x=-2.55,y=-0.35,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.35,"Better\nmental\nhealth",adj=1) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[7,]$xsi,y0=0.6,y1=1.6,lty=3) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[8,]$xsi,y0=0.6,y1=1.6,lty=3) segments(x0=zaza$xsi[9,]$xsi,x1=zaza$xsi[9,]$xsi,y0=0.6,y1=1.6,lty=3) rect(xleft = 3,xright=5,ybottom = -.1,ytop=1.1,col = "white",border = "white") title(xlab=expression("Latent variable " * theta * "\n (mental health)"),line=-4.3) lines(x=.5+c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=.5+c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=.5+c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=.5+c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(-0.45,-0.5),lty=3) lines(x=.5+c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(-0.45,-0.5),lty=3) lines(x=.5+c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(-0.45,-0.5),lty=3) par(xpd=F) dev.off() # CCC DIF HETEROGENE 1 plot.tam.difhet1 <- function(x, items=1:x$nitems, type="expected", low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } low <- low-2 high <- high+2 # device.Option <- getOption("device") pall <- c('#f7c611', '#61b5b7', '#ca346d', '#204776' ) time1 <- NULL if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(low, high, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) graphics::plot(theta, expScore[,i],,col=12, type="l", lwd=3, las=1, ylab="Score", xlab="Ability", main=paste("Expected Scores Curve - Item ", colnames(tamobj$resp)[i] ) , ylim=ylim2, ... ) } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( ifelse(dfr1a$Theta+0.5>-3 & dfr1a$Theta+0.5<3, dfr1a$Theta+0.5, NA), dfr1a$P, ylim=c(-.1,1.1),xlim=c(-3,3), xlab="", col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2,las=2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( ifelse(dfr1a$Theta+0.5>-3 & dfr1a$Theta+0.5<3, dfr1a$Theta+0.5, NA), dfr1a$P,xlim=c(-3,3), col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines(ifelse(dfr1a$Theta+ifelse(kk==3,1.5,ifelse(kk==4,0.4,0.5))>-3 & dfr1a$Theta+ifelse(kk==3,1.5,ifelse(kk==4,0.4,0.5))<3, dfr1a$Theta, NA)+ifelse(kk==3,1.5,ifelse(kk==4,0.4,0.5)), dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } # DIF heterogene convergent pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/ccc_dif_2_conv.pdf') par(xpd=F,mar=c(12.6,5.1,1.1,2.1)) plot.tam.difhet1(zaza,type = "items",export=F,ylab="Probability of response in treatment group (X=1)",main=NULL,package = "graphics",items = 3) text(x=-2,y=0.85,"0",col="#f7c611") text(x=-0.15,y=0.55,"1",col="#61b5b7") text(x=1.45+1.,y=0.55,"2",col="#ca346d") text(x=2.5,y=0.7,"3",col="#204776") par(xpd=T,mar=c(5.1,4.1,4.1,2.1)) segments(x0=-3,x1=.5+zaza$xsi[7,]$xsi,y0=-0.5,col="#f7c611",lwd=2) segments(x0=.5+zaza$xsi[7,]$xsi,x1=.75+zaza$xsi[8,]$xsi+0.25,y0=-0.5,col="#61b5b7",lwd=2) segments(x0=.75+zaza$xsi[8,]$xsi+0.25,x1=.5+zaza$xsi[9,]$xsi,y0=-0.5,col="#ca346d",lwd=2) segments(x0=.5+zaza$xsi[9,]$xsi,x1=3,y0=-0.5,col="#204776",lwd=2) points(x =.5 + zaza$xsi[7,]$xsi, y=-0.5,pch=9,cex=1) points(x =.75 + zaza$xsi[8,]$xsi+0.25, y=-0.5,pch=9,cex=1) points(x =.5 + zaza$xsi[9,]$xsi, y=-0.5,pch=9,cex=1) rect(xleft = 3,xright=5,ybottom = 0,ytop=1.1,col = "white",border = "white") text( x=mean(c(-2.5,zaza$xsi[7,]$xsi)), y=-0.55,"0",col="#f7c611" ) text( x=0.75+mean(c(zaza$xsi[7,]$xsi,zaza$xsi[8,]$xsi)), y=-0.55,"1",col="#61b5b7" ) text( x=0.75+mean(c(zaza$xsi[8,]$xsi,zaza$xsi[9,]$xsi)), y=-0.55,"2",col="#ca346d" ) text( x=0.5+mean(c(zaza$xsi[9,]$xsi,2.5)), y=-0.55,"3",col="#204776" ) text(x=0.5+zaza$xsi[7,]$xsi,y=-0.55,expression(delta["j,1"])) text(x=0.75+zaza$xsi[8,]$xsi+0.25,y=-0.55,expression(delta["j,2"])) text(x=0.5+zaza$xsi[9,]$xsi,y=-0.55,expression(delta["j,3"])) arrows(x0=zaza$xsi[7,]$xsi+0.05,x1=zaza$xsi[7,]$xsi+0.5-0.05,y0=0.625,length = 0.1, lwd = 2) arrows(x0=zaza$xsi[8,]$xsi+0.05,x1=zaza$xsi[8,]$xsi+1-0.05,y0=0.625,length = 0.1, lwd = 2) arrows(x0=zaza$xsi[9,]$xsi+0.05,x1=zaza$xsi[9,]$xsi+0.5-0.05,y0=0.625,length = 0.1, lwd = 2) text(x=0.25+zaza$xsi[7,]$xsi,y=0.675,"DIF") text(x=0.25+zaza$xsi[8,]$xsi+0.25,y=0.675,"DIF") text(x=0.25+zaza$xsi[9,]$xsi,y=0.675,"DIF") text(x = 0.25,y=-0.675,"Most probable response category") CurlyBraces(x0=-2.25, x1=2.75, y0=-0.6, y1=-0.6, pos = 2, direction = 2, depth=0.05) arrows(x0=-2.65,x1=-3,y0=-0.35,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.35,length = 0.15,lwd = 2) text(x=-2.55,y=-0.35,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.35,"Better\nmental\nhealth",adj=1) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[7,]$xsi,y0=0.6,y1=1.6,lty=3) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[8,]$xsi,y0=0.6,y1=1.6,lty=3) segments(x0=zaza$xsi[9,]$xsi,x1=zaza$xsi[9,]$xsi,y0=0.6,y1=1.6,lty=3) rect(xleft = 3,xright=5,ybottom = -.1,ytop=1.1,col = "white",border = "white") title(xlab=expression("Latent variable " * theta * "\n (mental health)"),line=-4.3) lines(x=.5+c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=1+c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=1+c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(-0.45,-0.5),lty=3) lines(x=.5+c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=.5+c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(-0.45,-0.5),lty=3) lines(x=.5+c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(-0.45,-0.5),lty=3) par(xpd=F) dev.off() # CCC DIF HETEROGENE 2 plot.tam.difhet2 <- function(x, items=1:x$nitems, type="expected", low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } low <- low-2 high <- high+2 # device.Option <- getOption("device") pall <- c('#f7c611', '#61b5b7', '#ca346d', '#204776' ) time1 <- NULL if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(low, high, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) graphics::plot(theta, expScore[,i],,col=12, type="l", lwd=3, las=1, ylab="Score", xlab="Ability", main=paste("Expected Scores Curve - Item ", colnames(tamobj$resp)[i] ) , ylim=ylim2, ... ) } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot(ifelse(dfr1a$Theta-1.125>-3,dfr1a$Theta-1.125,NA) , dfr1a$P, ylim=c(-.1,1.1),xlim=c(-3,3), xlab="", col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2,las=2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( ifelse(dfr1a$Theta-1.125>-3,dfr1a$Theta-1.125,NA), dfr1a$P,xlim=c(-3,3), col=pall[kk], type="l", xlim=c(-3,3), ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines(ifelse(dfr1a$Theta+ifelse(kk==3,1.5,ifelse(kk==4,0.4,0.5))>-3 & dfr1a$Theta+ifelse(kk==3,1.5,ifelse(kk==4,0.4,0.5))<3, dfr1a$Theta, NA)+ifelse(kk==3,1.5,ifelse(kk==4,0.4,0.5)), dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } # DIF heterogene divergent pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/ccc_dif_2_div.pdf') par(xpd=F,mar=c(12.6,5.1,1.1,2.1)) plot.tam.difhet2(zaza,type = "items",export=F,ylab="Probability of response in treatment group (X=1)",main=NULL,package = "graphics",items = 3) text(x=-2.5,y=0.6,"0",col="#f7c611") text(x=-0.15,y=0.55,"1",col="#61b5b7") text(x=1.45+0.475,y=0.55,"2",col="#ca346d") text(x=2.5,y=0.7,"3",col="#204776") par(xpd=T,mar=c(5.1,4.1,4.1,2.1)) segments(x0=-3,x1=-.5+zaza$xsi[7,]$xsi,y0=-0.5,col="#f7c611",lwd=2) segments(x0=-.5+zaza$xsi[7,]$xsi,x1=.75+zaza$xsi[8,]$xsi+0.25,y0=-0.5,col="#61b5b7",lwd=2) segments(x0=.75+zaza$xsi[8,]$xsi+0.25,x1=.5+zaza$xsi[9,]$xsi,y0=-0.5,col="#ca346d",lwd=2) segments(x0=.5+zaza$xsi[9,]$xsi,x1=3,y0=-0.5,col="#204776",lwd=2) points(x =.5 + zaza$xsi[7,]$xsi-1, y=-0.5,pch=9,cex=1) points(x =.75 + zaza$xsi[8,]$xsi+0.25, y=-0.5,pch=9,cex=1) points(x =.5 + zaza$xsi[9,]$xsi, y=-0.5,pch=9,cex=1) rect(xleft = 3,xright=5,ybottom = 0,ytop=1.1,col = "white",border = "white") text( x=-0.25+mean(c(-3,zaza$xsi[7,]$xsi)), y=-0.55,"0",col="#f7c611" ) text( x=-0.25+0.125+0.5+mean(c(zaza$xsi[7,]$xsi,zaza$xsi[8,]$xsi)), y=-0.55,"1",col="#61b5b7" ) text( x=0.5+0.25+mean(c(zaza$xsi[8,]$xsi,zaza$xsi[9,]$xsi)), y=-0.55,"2",col="#ca346d" ) text( x=0.5+mean(c(zaza$xsi[9,]$xsi,2.5)), y=-0.55,"3",col="#204776" ) text(x=-0.5+zaza$xsi[7,]$xsi,y=-0.55,expression(delta["j,1"])) text(x=0.75+zaza$xsi[8,]$xsi+0.25,y=-0.55,expression(delta["j,2"])) text(x=0.5+zaza$xsi[9,]$xsi,y=-0.55,expression(delta["j,3"])) arrows(x0=zaza$xsi[7,]$xsi-0.05,x1=zaza$xsi[7,]$xsi-0.5+0.05,y0=0.625,length = 0.1, lwd = 2) arrows(x0=zaza$xsi[8,]$xsi+0.05,x1=zaza$xsi[8,]$xsi+1-0.05,y0=0.625,length = 0.1, lwd = 2) arrows(x0=zaza$xsi[9,]$xsi+0.05,x1=zaza$xsi[9,]$xsi+0.5-0.05,y0=0.625,length = 0.1, lwd = 2) text(x=-0.25+zaza$xsi[7,]$xsi,y=0.675,"DIF") text(x=0.25+zaza$xsi[8,]$xsi+0.25,y=0.675,"DIF") text(x=0.25+zaza$xsi[9,]$xsi,y=0.675,"DIF") text(x = 0.25,y=-0.675,"Most probable response category") CurlyBraces(x0=-2.25, x1=2.75, y0=-0.6, y1=-0.6, pos = 2, direction = 2, depth=0.05) arrows(x0=-2.65,x1=-3,y0=-0.35,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.35,length = 0.15,lwd = 2) text(x=-2.55,y=-0.35,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.35,"Better\nmental\nhealth",adj=1) segments(x0=zaza$xsi[7,]$xsi,x1=zaza$xsi[7,]$xsi,y0=0.6,y1=1.6,lty=3) segments(x0=zaza$xsi[8,]$xsi,x1=zaza$xsi[8,]$xsi,y0=0.6,y1=1.6,lty=3) segments(x0=zaza$xsi[9,]$xsi,x1=zaza$xsi[9,]$xsi,y0=0.6,y1=1.6,lty=3) rect(xleft = 3,xright=5,ybottom = -.1,ytop=1.1,col = "white",border = "white") title(xlab=expression("Latent variable " * theta * "\n (mental health)"),line=-4.3) lines(x=-.5+c(zaza$xsi[7,]$xsi,zaza$xsi[7,]$xsi),y=c(0.65,-0.5),lty=3) lines(x=1+c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=1+c(zaza$xsi[8,]$xsi,zaza$xsi[8,]$xsi),y=c(-0.45,-0.5),lty=3) lines(x=.5+c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(0.65,-0.25),lty=3) lines(x=.5+c(zaza$xsi[9,]$xsi,zaza$xsi[9,]$xsi),y=c(-0.45,-0.5),lty=3) par(xpd=F) dev.off() ########################## # LM FACTEURS IGNORE DIF ########################## ####### SCENARIOS SANS TE res.dat.article.ignore.long <- reshape(res.dat.article.ignore,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.ignore.long) <- NULL colnames(res.dat.article.ignore.long)[6:11] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.ignore.h0 <- res.dat.article.ignore[res.dat.article.ignore$true.beta==0,] res.dat.article.ignore.h0$prop.dif <- res.dat.article.ignore.h0$nb.dif/res.dat.article.ignore.h0$J res.dat.article.ignore.h0.long <- reshape(res.dat.article.ignore.h0,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.ignore.h0.long) <- NULL colnames(res.dat.article.ignore.h0.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.ignore.h0.long$prop.dif <- as.numeric(res.dat.article.ignore.h0.long$prop.dif) res.dat.article.ignore.h0.long$N <- as.numeric(res.dat.article.ignore.h0.long$N) res.dat.article.ignore.h0.long$true.gamma <- as.numeric(res.dat.article.ignore.h0.long$true.gamma) res.dat.article.ignore.h0.long$J <- as.numeric(res.dat.article.ignore.h0.long$J) res.dat.article.ignore.h0$true.gamma <- as.numeric(res.dat.article.ignore.h0$true.gamma) res.dat.article.ignore.h0.long$bias <- abs(res.dat.article.ignore.h0.long$bias) res.dat.article.ignore.h0.long$betahat <- abs(res.dat.article.ignore.h0.long$betahat) # bias res.dat.article.ignore.h0.long$abs.bias <- abs(res.dat.article.ignore.h0.long$bias) summary(lm(abs.bias~true.gamma+prop.dif+N+J,data = res.dat.article.ignore.h0.long)) summary(lm(abs.bias~true.gamma+prop.dif,data = res.dat.article.ignore.h0.long)) summary(res.dat.article.ignore.h1.long[res.dat.article.ignore.h1.long$abs.gamma==0.5,"abs.bias"]) summary(res.dat.article.ignore.h1.long[res.dat.article.ignore.h1.long$abs.gamma==0.3 & res.dat.article.ignore.h1.long$prop.dif<0.3,"abs.bias"]) # type I summary(lm(typeIerror~J+true.gamma+prop.dif+N,data = res.dat.article.ignore.h0.long)) summary(lm(typeIerror~true.gamma+prop.dif+N,data = res.dat.article.ignore.h0.long)) res.dat.article.ignore.h0.long$abs.gamma <- abs(res.dat.article.ignore.h0.long$true.gamma) summary(lm(typeIerror~abs.gamma+prop.dif+N,data = res.dat.article.ignore.h0.long)) res.dat.article.ignore.h0[res.dat.article.ignore.h0$true.gamma==-0.5 & res.dat.article.ignore.h0$prop.dif>0.3,]$typeIerror.300 res.dat.article.ignore.h0[res.dat.article.ignore.h0$true.gamma==-0.3 & res.dat.article.ignore.h0$prop.dif<0.3,]$typeIerror.50 # coverage summary(lm(coverage~abs.gamma+prop.dif+N,data = res.dat.article.ignore.h0.long)) ####### SCENARIOS AVEC TE MASK res.dat.article.ignore.h1 <- res.dat.article.ignore[res.dat.article.ignore$true.beta!=0 & res.dat.article.ignore$true.gamma>0,] res.dat.article.ignore.h1$prop.dif <- res.dat.article.ignore.h1$nb.dif/res.dat.article.ignore.h1$J res.dat.article.ignore.h1.long <- reshape(res.dat.article.ignore.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.ignore.h1.long) <- NULL colnames(res.dat.article.ignore.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.ignore.h1.long$prop.dif <- as.numeric(res.dat.article.ignore.h1.long$prop.dif) res.dat.article.ignore.h1.long$N <- as.numeric(res.dat.article.ignore.h1.long$N) res.dat.article.ignore.h1.long$true.gamma <- as.numeric(res.dat.article.ignore.h1.long$true.gamma) res.dat.article.ignore.h1.long$J <- as.numeric(res.dat.article.ignore.h1.long$J) res.dat.article.ignore.h1$true.gamma <- as.numeric(res.dat.article.ignore.h1$true.gamma) # bias res.dat.article.ignore.h1.long$abs.bias <- abs(res.dat.article.ignore.h1.long$bias) res.dat.article.ignore.h1.long$abs.gamma <- abs(res.dat.article.ignore.h1.long$true.gamma) res.dat.article.ignore.h1.long$sign.gamma <- sign(res.dat.article.ignore.h1.long$true.gamma) summary(lm(abs.bias~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.ignore.h1.long)) # coverage res.dat.article.ignore.h1.long$masks <- res.dat.article.ignore.h1.long$true.beta/res.dat.article.ignore.h1.long$true.gamma>0 res.dat.article.ignore.h1.long$masks <- 1*res.dat.article.ignore.h1.long$masks summary(lm(coverage~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.ignore.h1.long)) # power res.dat.article.ignore.h1.long$powerdif <- as.numeric(res.dat.article.ignore.h1.long$power)-as.numeric(res.dat.article.ignore.h1.long$theoretical.power) summary(lm(powerdif~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.ignore.h1.long)) # bias +- summary(lm(bias~abs.gamma+prop.dif+true.beta+N+masks,data = res.dat.article.ignore.h1.long)) summary(lm(bias~masks,data = res.dat.article.ignore.h1.long)) summary(res.dat.article.ignore.h1.long[res.dat.article.ignore.h1.long$masks==1,]$bias) summary(res.dat.article.ignore.h1.long[res.dat.article.ignore.h1.long$masks==0,]$bias) ####### SCENARIOS AVEC TE AMPLIFY res.dat.article.ignore.h1 <- res.dat.article.ignore[res.dat.article.ignore$true.beta!=0 & res.dat.article.ignore$true.gamma<0,] res.dat.article.ignore.h1$prop.dif <- res.dat.article.ignore.h1$nb.dif/res.dat.article.ignore.h1$J res.dat.article.ignore.h1.long <- reshape(res.dat.article.ignore.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.ignore.h1.long) <- NULL colnames(res.dat.article.ignore.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.ignore.h1.long$prop.dif <- as.numeric(res.dat.article.ignore.h1.long$prop.dif) res.dat.article.ignore.h1.long$N <- as.numeric(res.dat.article.ignore.h1.long$N) res.dat.article.ignore.h1.long$true.gamma <- as.numeric(res.dat.article.ignore.h1.long$true.gamma) res.dat.article.ignore.h1.long$J <- as.numeric(res.dat.article.ignore.h1.long$J) res.dat.article.ignore.h1$true.gamma <- as.numeric(res.dat.article.ignore.h1$true.gamma) res.dat.article.ignore.h1$true.beta <- as.numeric(res.dat.article.ignore.h1$true.beta) # bias res.dat.article.ignore.h1.long$abs.bias <- abs(res.dat.article.ignore.h1.long$bias) res.dat.article.ignore.h1.long$abs.gamma <- abs(res.dat.article.ignore.h1.long$true.gamma) res.dat.article.ignore.h1.long$sign.gamma <- sign(res.dat.article.ignore.h1.long$true.gamma) summary(lm(abs.bias~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.ignore.h1.long)) # coverage res.dat.article.ignore.h1.long$masks <- res.dat.article.ignore.h1.long$true.beta/res.dat.article.ignore.h1.long$true.gamma>0 res.dat.article.ignore.h1.long$masks <- 1*res.dat.article.ignore.h1.long$masks summary(lm(coverage~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.ignore.h1.long)) # power res.dat.article.ignore.h1.long$powerdif <- as.numeric(res.dat.article.ignore.h1.long$power)-as.numeric(res.dat.article.ignore.h1.long$theoretical.power) summary(lm(powerdif~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.ignore.h1.long)) # bias +- summary(lm(bias~abs.gamma+prop.dif+true.beta+N+masks,data = res.dat.article.ignore.h1.long)) summary(lm(bias~masks,data = res.dat.article.ignore.h1.long)) summary(res.dat.article.ignore.h1.long[res.dat.article.ignore.h1.long$masks==1,]$bias) summary(res.dat.article.ignore.h1.long[res.dat.article.ignore.h1.long$masks==0,]$bias) ########################## # DESCRIPTION PCM-DIF ########################## ####### SCENARIOS SANS TE res.dat.article.dif.h0 <- res.dat.article.dif[res.dat.article.dif$true.beta==0,] res.dat.article.dif.h0$prop.dif <- res.dat.article.dif.h0$nb.dif/res.dat.article.dif.h0$J res.dat.article.dif.h0.long <- reshape(res.dat.article.dif.h0,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.dif.h0.long) <- NULL colnames(res.dat.article.dif.h0.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.dif.h0.long$prop.dif <- as.numeric(res.dat.article.dif.h0.long$prop.dif) res.dat.article.dif.h0.long$N <- as.numeric(res.dat.article.dif.h0.long$N) res.dat.article.dif.h0.long$true.gamma <- as.numeric(res.dat.article.dif.h0.long$true.gamma) res.dat.article.dif.h0.long$J <- as.numeric(res.dat.article.dif.h0.long$J) res.dat.article.dif.h0$true.gamma <- as.numeric(res.dat.article.dif.h0$true.gamma) # typeI summary(as.numeric(res.dat.article.dif.h0.long$typeIerror)) # bias summary(abs(as.numeric(res.dat.article.dif.h0.long$bias))) # coverage summary(as.numeric(res.dat.article.dif.h0.long$coverage)) ####### SCENARIOS AVEC TE res.dat.article.dif.h1 <- res.dat.article.dif[res.dat.article.dif$true.beta!=0,] res.dat.article.dif.h1$prop.dif <- res.dat.article.dif.h1$nb.dif/res.dat.article.dif.h1$J res.dat.article.dif.h1.long <- reshape(res.dat.article.dif.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.dif.h1.long) <- NULL colnames(res.dat.article.dif.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.dif.h1.long$prop.dif <- as.numeric(res.dat.article.dif.h1.long$prop.dif) res.dat.article.dif.h1.long$N <- as.numeric(res.dat.article.dif.h1.long$N) res.dat.article.dif.h1.long$true.gamma <- as.numeric(res.dat.article.dif.h1.long$true.gamma) res.dat.article.dif.h1.long$J <- as.numeric(res.dat.article.dif.h1.long$J) res.dat.article.dif.h1$true.gamma <- as.numeric(res.dat.article.dif.h1$true.gamma) res.dat.article.dif.h1.long$powerdif <- as.numeric(res.dat.article.dif.h1.long$power)-as.numeric(res.dat.article.dif.h1.long$theoretical.power) # powerdif summary(as.numeric(res.dat.article.dif.h1.long$powerdif)) summary(lm(powerdif~1,data=res.dat.article.dif.h1.long)) # bias summary(abs(as.numeric(res.dat.article.dif.h1.long$bias))) # coverage summary(as.numeric(res.dat.article.dif.h1.long$coverage)) ########################## # LM FACTEURS ROSALI ########################## ####### SCENARIOS SANS TE res.dat.article.rosali.dif.h0 <- res.dat.article.rosali.dif[res.dat.article.rosali.dif$true.beta==0,] res.dat.article.rosali.dif.h0$prop.dif <- res.dat.article.rosali.dif.h0$nb.dif/res.dat.article.rosali.dif.h0$J res.dat.article.rosali.dif.h0.long <- reshape(res.dat.article.rosali.dif.h0,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.rosali.dif.h0.long) <- NULL colnames(res.dat.article.rosali.dif.h0.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.rosali.dif.h0.long$prop.dif <- as.numeric(res.dat.article.rosali.dif.h0.long$prop.dif) res.dat.article.rosali.dif.h0.long$N <- as.numeric(res.dat.article.rosali.dif.h0.long$N) res.dat.article.rosali.dif.h0.long$true.gamma <- as.numeric(res.dat.article.rosali.dif.h0.long$true.gamma) res.dat.article.rosali.dif.h0.long$J <- as.numeric(res.dat.article.rosali.dif.h0.long$J) res.dat.article.rosali.dif.h0.long$abs.gamma <- abs(res.dat.article.rosali.dif.h0.long$true.gamma) res.dat.article.rosali.dif.h0$true.gamma <- as.numeric(res.dat.article.rosali.dif.h0$true.gamma) # bias res.dat.article.rosali.dif.h0.long$abs.bias <- abs(res.dat.article.rosali.dif.h0.long$bias) summary(lm(abs.bias~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h0.long)) # type I summary(lm(typeIerror~abs.gamma+prop.dif+N+J,data = res.dat.article.rosali.dif.h0.long)) summary(lm(typeIerror~abs.gamma+prop.dif+N,data = res.dat.article.rosali.dif.h0.long)) summary(as.numeric(res.dat.article.rosali.dif.h0.long$typeIerror)) # coverage summary(lm(coverage~abs.gamma+prop.dif+N,data = res.dat.article.ignore.h0.long)) ####### SCENARIOS AVEC TE MASK res.dat.article.rosali.dif.h1 <- res.dat.article.rosali.dif[res.dat.article.rosali.dif$true.beta!=0 & res.dat.article.rosali.dif$true.gamma>0,] res.dat.article.rosali.dif.h1$prop.dif <- res.dat.article.rosali.dif.h1$nb.dif/res.dat.article.rosali.dif.h1$J res.dat.article.rosali.dif.h1.long <- reshape(res.dat.article.rosali.dif.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.rosali.dif.h1.long) <- NULL colnames(res.dat.article.rosali.dif.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.rosali.dif.h1.long$prop.dif <- as.numeric(res.dat.article.rosali.dif.h1.long$prop.dif) res.dat.article.rosali.dif.h1.long$N <- as.numeric(res.dat.article.rosali.dif.h1.long$N) res.dat.article.rosali.dif.h1.long$true.gamma <- as.numeric(res.dat.article.rosali.dif.h1.long$true.gamma) res.dat.article.rosali.dif.h1.long$J <- as.numeric(res.dat.article.rosali.dif.h1.long$J) res.dat.article.rosali.dif.h1$true.gamma <- as.numeric(res.dat.article.rosali.dif.h1$true.gamma) # bias res.dat.article.rosali.dif.h1.long$abs.bias <- abs(res.dat.article.rosali.dif.h1.long$bias) res.dat.article.rosali.dif.h1.long$abs.gamma <- abs(res.dat.article.rosali.dif.h1.long$true.gamma) res.dat.article.rosali.dif.h1.long$sign.gamma <- sign(res.dat.article.rosali.dif.h1.long$true.gamma) summary(lm(abs.bias~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h1.long)) # coverage summary(lm(coverage~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h1.long)) summary(res.dat.article.rosali.dif.h1.long$coverage) # power res.dat.article.rosali.dif.h1.long$powerdif <- as.numeric(res.dat.article.rosali.dif.h1.long$power)-as.numeric(res.dat.article.rosali.dif.h1.long$theoretical.power) summary(lm(powerdif~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h1.long)) # bias +- summary(lm(bias~abs.gamma+prop.dif+true.beta+N+masks,data = res.dat.article.rosali.dif.h1.long)) summary(lm(bias~masks,data = res.dat.article.rosali.dif.h1.long)) summary(res.dat.article.rosali.dif.h1.long[res.dat.article.rosali.dif.h1.long$masks==1,]$bias) summary(res.dat.article.rosali.dif.h1.long[res.dat.article.rosali.dif.h1.long$masks==0,]$bias) ####### SCENARIOS AVEC TE AMPLIFY res.dat.article.rosali.dif.h1 <- res.dat.article.rosali.dif[res.dat.article.rosali.dif$true.beta!=0 & res.dat.article.rosali.dif$true.gamma<0,] res.dat.article.rosali.dif.h1$prop.dif <- res.dat.article.rosali.dif.h1$nb.dif/res.dat.article.rosali.dif.h1$J res.dat.article.rosali.dif.h1.long <- reshape(res.dat.article.rosali.dif.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.rosali.dif.h1.long) <- NULL colnames(res.dat.article.rosali.dif.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.rosali.dif.h1.long$prop.dif <- as.numeric(res.dat.article.rosali.dif.h1.long$prop.dif) res.dat.article.rosali.dif.h1.long$N <- as.numeric(res.dat.article.rosali.dif.h1.long$N) res.dat.article.rosali.dif.h1.long$true.gamma <- as.numeric(res.dat.article.rosali.dif.h1.long$true.gamma) res.dat.article.rosali.dif.h1.long$J <- as.numeric(res.dat.article.rosali.dif.h1.long$J) res.dat.article.rosali.dif.h1$true.gamma <- as.numeric(res.dat.article.rosali.dif.h1$true.gamma) # bias res.dat.article.rosali.dif.h1.long$abs.bias <- abs(res.dat.article.rosali.dif.h1.long$bias) res.dat.article.rosali.dif.h1.long$abs.gamma <- abs(res.dat.article.rosali.dif.h1.long$true.gamma) res.dat.article.rosali.dif.h1.long$sign.gamma <- sign(res.dat.article.rosali.dif.h1.long$true.gamma) summary(lm(abs.bias~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h1.long)) # coverage summary(lm(coverage~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h1.long)) summary(res.dat.article.rosali.dif.h1.long$coverage) # power res.dat.article.rosali.dif.h1.long$powerdif <- as.numeric(res.dat.article.rosali.dif.h1.long$power)-as.numeric(res.dat.article.rosali.dif.h1.long$theoretical.power) summary(lm(powerdif~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.rosali.dif.h1.long)) ########################## # Plots ROSALI vs ignore ########################## par(mfrow=c(2,2)) plot(res.dat.article.ignore.h0.long$typeIerror,res.dat.article.rosali.dif.h0.long$typeIerror, pch=3,col="#CD5E35",xlim=c(0,1),ylim=c(0,1),cex=1.5, xlab = "Type-I error when ignoring DIF",ylab="Type-I error after ROSALI DIF detection", main="Type-I error",axes = F) segments(x0=0,y0=0,x1=1,y1=1,lty=2) axis(1) axis(2) plot(c(res.dat.article.ignore.h0.long$abs.bias,res.dat.article.ignore.h1.long$abs.bias), c(res.dat.article.rosali.dif.h0.long$abs.bias,res.dat.article.rosali.dif.h1.long$abs.bias), pch=3,col="#CD5E35",xlim=c(0,0.4),ylim=c(0,0.4),cex=1.5, xlab = "Absolute bias when ignoring DIF",ylab="Absolute bias after ROSALI DIF detection", main="Absolute bias",axes = F) segments(x0=0,y0=0,x1=0.4,y1=0.4,lty=2) axis(1) axis(2) plot(c(res.dat.article.ignore.h0.long$coverage,res.dat.article.ignore.h1.long$coverage), c(res.dat.article.rosali.dif.h0.long$coverage,res.dat.article.rosali.dif.h1.long$coverage), pch=3,col="#CD5E35",xlim=c(0,1),ylim=c(0,1),cex=1.5, xlab = "Coverage when ignoring DIF",ylab="Coverage after ROSALI DIF detection", main="Coverage",axes = F) segments(x0=0,y0=0,x1=1,y1=1,lty=2) axis(1) axis(2) plot(c(res.dat.article.ignore.h0.long$powerdif,res.dat.article.ignore.h1.long$powerdif), c(res.dat.article.rosali.dif.h0.long$powerdif,res.dat.article.rosali.dif.h1.long$powerdif), pch=3,col="#CD5E35",xlim=c(-1,1),ylim=c(-1,1),cex=1.5, xlab = "Power difference when ignoring DIF",ylab="Power difference after ROSALI DIF detection", main="Difference between expected and observed power",axes = F) segments(x0=-1,y0=-1,x1=1,y1=1,lty=2) axis(1) axis(2) par(mfrow=c(1,1)) which.typeI <- which(as.numeric(res.dat.article.rosali.dif.h0.long$typeIerror)-as.numeric(res.dat.article.ignore.h0.long$typeIerror)<=-0.1) res.dat.article.ignore.h0.long[which.typeI,] df_temp_rosali <- rbind(res.dat.article.rosali.dif.h0.long[,c(1:6,13)],res.dat.article.rosali.dif.h1.long[,c(1:6,13)]) df_temp_ignore <- rbind(res.dat.article.ignore.h0.long[,c(1:6,13)],res.dat.article.ignore.h1.long[,c(1:6,13)]) which.bias <- which(as.numeric(df_temp_rosali$abs.bias)-as.numeric(df_temp_ignore$abs.bias)<=-0.05) df_temp_rosali[which.bias,] df_temp_rosali <- rbind(res.dat.article.rosali.dif.h0.long[,c(1:6,12)],res.dat.article.rosali.dif.h1.long[,c(1:6,12)]) df_temp_ignore <- rbind(res.dat.article.ignore.h0.long[,c(1:6,12)],res.dat.article.ignore.h1.long[,c(1:6,12)]) which.coverage <- which(as.numeric(df_temp_rosali$coverage)-as.numeric(df_temp_ignore$coverage)>=0.1) df_temp_rosali[which.coverage,] df_temp_rosali <- rbind(res.dat.article.rosali.dif.h1.long) df_temp_ignore <- rbind(res.dat.article.ignore.h1.long) which.power <- which(abs(as.numeric(df_temp_rosali$powerdif))-abs(as.numeric(df_temp_ignore$powerdif))<=-0.1) df_temp_rosali[which.power,] ########################## # LM FACTEURS RESIDIF ########################## ####### SCENARIOS SANS TE res.dat.article.residif.dif.h0 <- res.dat.article.residif.dif[res.dat.article.residif.dif$true.beta==0,] res.dat.article.residif.dif.h0$prop.dif <- res.dat.article.residif.dif.h0$nb.dif/res.dat.article.residif.dif.h0$J res.dat.article.residif.dif.h0.long <- reshape(res.dat.article.residif.dif.h0,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.residif.dif.h0.long) <- NULL colnames(res.dat.article.residif.dif.h0.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.residif.dif.h0.long$prop.dif <- as.numeric(res.dat.article.residif.dif.h0.long$prop.dif) res.dat.article.residif.dif.h0.long$N <- as.numeric(res.dat.article.residif.dif.h0.long$N) res.dat.article.residif.dif.h0.long$true.gamma <- as.numeric(res.dat.article.residif.dif.h0.long$true.gamma) res.dat.article.residif.dif.h0.long$J <- as.numeric(res.dat.article.residif.dif.h0.long$J) res.dat.article.residif.dif.h0.long$abs.gamma <- as.numeric(res.dat.article.residif.dif.h0.long$true.gamma) res.dat.article.residif.dif.h0$true.gamma <- as.numeric(res.dat.article.residif.dif.h0$true.gamma) # bias res.dat.article.residif.dif.h0.long$abs.gamma <- abs(res.dat.article.residif.dif.h0.long$true.gamma) res.dat.article.residif.dif.h0.long$abs.bias <- abs(res.dat.article.residif.dif.h0.long$bias) summary(lm(abs.bias~abs.gamma+prop.dif+N+J,data = res.dat.article.residif.dif.h0.long)) summary(lm(abs.bias~abs.gamma+prop.dif+N,data = res.dat.article.residif.dif.h0.long)) summary(res.dat.article.residif.dif.h0.long$abs.bias) # type I summary(lm(typeIerror~abs.gamma+prop.dif+N+J,data = res.dat.article.residif.dif.h0.long)) summary(lm(typeIerror~abs.gamma+prop.dif+N,data = res.dat.article.residif.dif.h0.long)) summary(as.numeric(res.dat.article.residif.dif.h0.long$typeIerror)) # coverage summary(lm(coverage~abs.gamma+prop.dif+N,data = res.dat.article.ignore.h0.long)) ####### SCENARIOS AVEC TE MASK res.dat.article.residif.dif.h1 <- res.dat.article.residif.dif[res.dat.article.residif.dif$true.beta!=0 & res.dat.article.residif.dif$true.gamma>0,] res.dat.article.residif.dif.h1$prop.dif <- res.dat.article.residif.dif.h1$nb.dif/res.dat.article.residif.dif.h1$J res.dat.article.residif.dif.h1.long <- reshape(res.dat.article.residif.dif.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.residif.dif.h1.long) <- NULL colnames(res.dat.article.residif.dif.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.residif.dif.h1.long$prop.dif <- as.numeric(res.dat.article.residif.dif.h1.long$prop.dif) res.dat.article.residif.dif.h1.long$N <- as.numeric(res.dat.article.residif.dif.h1.long$N) res.dat.article.residif.dif.h1.long$true.gamma <- as.numeric(res.dat.article.residif.dif.h1.long$true.gamma) res.dat.article.residif.dif.h1.long$J <- as.numeric(res.dat.article.residif.dif.h1.long$J) res.dat.article.residif.dif.h1$true.gamma <- as.numeric(res.dat.article.residif.dif.h1$true.gamma) # bias res.dat.article.residif.dif.h1.long$abs.bias <- abs(res.dat.article.residif.dif.h1.long$bias) res.dat.article.residif.dif.h1.long$abs.gamma <- abs(res.dat.article.residif.dif.h1.long$true.gamma) summary(lm(abs.bias~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.residif.dif.h1.long)) # coverage summary(lm(coverage~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.residif.dif.h1.long)) # power res.dat.article.residif.dif.h1.long$powerdif <- as.numeric(res.dat.article.residif.dif.h1.long$power)-as.numeric(res.dat.article.residif.dif.h1.long$theoretical.power) summary(lm(powerdif~true.gamma+prop.dif+true.beta+N+J,data = res.dat.article.residif.dif.h1.long)) ####### SCENARIOS AVEC TE AMPLIFY res.dat.article.residif.dif.h1 <- res.dat.article.residif.dif[res.dat.article.residif.dif$true.beta!=0 & res.dat.article.residif.dif$true.gamma<0,] res.dat.article.residif.dif.h1$prop.dif <- res.dat.article.residif.dif.h1$nb.dif/res.dat.article.residif.dif.h1$J res.dat.article.residif.dif.h1.long <- reshape(res.dat.article.residif.dif.h1,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.residif.dif.h1.long) <- NULL colnames(res.dat.article.residif.dif.h1.long)[7:12] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.residif.dif.h1.long$prop.dif <- as.numeric(res.dat.article.residif.dif.h1.long$prop.dif) res.dat.article.residif.dif.h1.long$N <- as.numeric(res.dat.article.residif.dif.h1.long$N) res.dat.article.residif.dif.h1.long$true.gamma <- as.numeric(res.dat.article.residif.dif.h1.long$true.gamma) res.dat.article.residif.dif.h1.long$J <- as.numeric(res.dat.article.residif.dif.h1.long$J) res.dat.article.residif.dif.h1$true.gamma <- as.numeric(res.dat.article.residif.dif.h1$true.gamma) # bias res.dat.article.residif.dif.h1.long$abs.bias <- abs(res.dat.article.residif.dif.h1.long$bias) res.dat.article.residif.dif.h1.long$abs.gamma <- abs(res.dat.article.residif.dif.h1.long$true.gamma) summary(lm(bias~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.residif.dif.h1.long)) # coverage summary(lm(coverage~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.residif.dif.h1.long)) # power res.dat.article.residif.dif.h1.long$powerdif <- as.numeric(res.dat.article.residif.dif.h1.long$power)-as.numeric(res.dat.article.residif.dif.h1.long$theoretical.power) summary(lm(powerdif~abs.gamma+prop.dif+true.beta+N+J,data = res.dat.article.residif.dif.h1.long)) ########################## # Plots RESIDIF vs ignore ########################## par(mfrow=c(2,2)) plot(res.dat.article.ignore.h0.long$typeIerror,res.dat.article.residif.dif.h0.long$typeIerror, pch=3,col="#CD5E35",xlim=c(0,1),ylim=c(0,1),cex=1.5, xlab = "Type-I error when ignoring DIF",ylab="Type-I error after RESIDIF DIF detection", main="Type-I error",axes = F) segments(x0=0,y0=0,x1=1,y1=1,lty=2) axis(1) axis(2) plot(c(res.dat.article.ignore.h0.long$abs.bias,res.dat.article.ignore.h1.long$abs.bias), c(res.dat.article.residif.dif.h0.long$abs.bias,res.dat.article.residif.dif.h1.long$abs.bias), pch=3,col="#CD5E35",xlim=c(0,0.4),ylim=c(0,0.4),cex=1.5, xlab = "Absolute bias when ignoring DIF",ylab="Absolute bias after RESIDIF DIF detection", main="Absolute bias",axes = F) segments(x0=0,y0=0,x1=0.4,y1=0.4,lty=2) axis(1) axis(2) plot(c(res.dat.article.ignore.h0.long$coverage,res.dat.article.ignore.h1.long$coverage), c(res.dat.article.residif.dif.h0.long$coverage,res.dat.article.residif.dif.h1.long$coverage), pch=3,col="#CD5E35",xlim=c(0,1),ylim=c(0,1),cex=1.5, xlab = "Coverage when ignoring DIF",ylab="Coverage after RESIDIF DIF detection", main="Coverage",axes = F) segments(x0=0,y0=0,x1=1,y1=1,lty=2) axis(1) axis(2) plot(c(res.dat.article.ignore.h0.long$powerdif,res.dat.article.ignore.h1.long$powerdif), c(res.dat.article.residif.dif.h0.long$powerdif,res.dat.article.residif.dif.h1.long$powerdif), pch=3,col="#CD5E35",xlim=c(-1,1),ylim=c(-1,1),cex=1.5, xlab = "Power difference when ignoring DIF",ylab="Power difference after RESIDIF DIF detection", main="Difference between expected and observed power",axes = F) segments(x0=-1,y0=-1,x1=1,y1=1,lty=2) axis(1) axis(2) par(mfrow=c(1,1)) which.typeI <- which(as.numeric(res.dat.article.residif.dif.h0.long$typeIerror)-as.numeric(res.dat.article.ignore.h0.long$typeIerror)<=-0.1) res.dat.article.ignore.h0.long[which.typeI,] res.dat.article.residif.dif.h0.long$masks <- NA res.dat.article.ignore.h0.long$masks <- NA df_temp_residif <- rbind(res.dat.article.residif.dif.h0.long[,c(1:6,14:15)],res.dat.article.residif.dif.h1.long[,c(1:6,13,16)]) df_temp_ignore <- rbind(res.dat.article.ignore.h0.long[,c(1:6,13,15)],res.dat.article.ignore.h1.long[,c(1:6,13,16)]) which.bias <- which(as.numeric(df_temp_residif$abs.bias)-as.numeric(df_temp_ignore$abs.bias)<=-0.05) df_temp_residif[which.bias,] table(df_temp_residif[which.bias,]$masks) which.bias <- which(as.numeric(df_temp_residif$abs.bias)-as.numeric(df_temp_ignore$abs.bias)>0.05) df_temp_residif[which.bias,] df_temp_residif <- rbind(res.dat.article.residif.dif.h0.long[,c(1:6,12,15)],res.dat.article.residif.dif.h1.long[,c(1:6,12,16)]) df_temp_ignore <- rbind(res.dat.article.ignore.h0.long[,c(1:6,12,15)],res.dat.article.ignore.h1.long[,c(1:6,12,16)]) which.bias <- which(as.numeric(df_temp_residif$coverage)-as.numeric(df_temp_ignore$coverage)<=-0.1) df_temp_residif[which.coverage,] table(df_temp_residif[which.coverage,]$masks) df_temp_residif <- rbind(res.dat.article.residif.dif.h1.long) df_temp_ignore <- rbind(res.dat.article.ignore.h1.long) which.power <- which(abs(as.numeric(df_temp_residif$powerdif))-abs(as.numeric(df_temp_ignore$powerdif))<=-0.1) df_temp_residif[which.power,] table(df_temp_residif[which.power,]$masks) ########################## # BOXPLOT PERF VS NODIF ########################## par(mfrow=c(2,2)) par(bg = "white") res.dat.article.nodif <- res.dat.article[res.dat$nb.dif==0,] ### Type I error bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",120), rep("ROSALI",120), rep("RESIDIF",120), rep("PCM-DIF",120) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab="Analysis strategy",pch=3,main="Type-I error rate (TE = 0)", ylab="RCT type-I error",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ### BIAS bp.dat.bias.ignore <- as.numeric(res.dat.article[res.dat.article$nb.dif!=0,"bias"]) bp.dat.bias.nodif <- as.numeric(res.dat.article.nodif[,"bias"]) bp.dat.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$nb.dif!=0,"bias"]) bp.dat.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$nb.dif!=0,"bias"]) bp.dat.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$nb.dif!=0,"bias"]) bp.dat.bias <- data.frame(bias=c(bp.dat.bias.nodif,bp.dat.bias.ignore,bp.dat.bias.rosali,bp.dat.bias.residif, bp.dat.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",120), rep("ROSALI",120), rep("RESIDIF",120), rep("PCM-DIF",120) ) ) bp.dat.bias$method <- factor(bp.dat.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) bp.dat.bias$bias <- abs(bp.dat.bias$bias) boxplot(bp.dat.bias$bias~bp.dat.bias$method,xlab="Analysis strategy",pch=3,main="Absolute bias", ylab="Absolute bias",ylim=c(0,.5),yaxt="n", col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850"), cex.lab=1.45,cex.main=1.5,cex.axis=1.45) axis(2,seq(-.5,.5,0.25),cex.axis=1.45) abline(h=0,lty=2,col='#777777',lwd=2) ### COVERAGE bp.dat.coverage.ignore <- as.numeric(res.dat.article[res.dat.article$nb.dif!=0,"coverage"]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",120), rep("ROSALI",120), rep("RESIDIF",120), rep("PCM-DIF",120) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab="Analysis strategy",pch=3,main="Coverage", ylab="Coverage",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(-1,1,0.25),cex=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ### POWER bp.dat.power.ignore.mask <- as.numeric(res.dat.article[res.dat.article$true.gamma>0 & res.dat.article$true.beta>0 & res.dat.article$nb.dif!=0,"power"])-as.numeric(res.dat.article[res.dat.article$true.gamma>0 & res.dat.article$true.beta>0 & res.dat.article$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article[res.dat.article$true.gamma<0 & res.dat.article$true.beta>0 & res.dat.article$nb.dif!=0,"power"])-as.numeric(res.dat.article[res.dat.article$true.gamma<0 & res.dat.article$true.beta>0 & res.dat.article$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif[res.dat.article.nodif$true.beta>0,"power"])-as.numeric(res.dat.article.nodif[res.dat.article.nodif$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",48),rep("AMPLIFY1",48), rep("MASK2",48),rep("AMPLIFY2",48), rep("MASK3",48),rep("AMPLIFY3",48), rep("MASK4",48),rep("AMPLIFY4",48) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="Power (TE ≠ 0)", ylab="RCT power - expected power",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) title(cex.lab=1.45,xlab="Analysis strategy") # legend("topright",fill = c("#798A5D","#D4E8B5"),c('DIF masks treatment effect','DIF amplifies treatment effect')) par(mfrow=c(1,1)) par(bg = "white") ########################## # DESCRIPTION NO DIF ########################## ####### SCENARIOS SANS TE res.dat.article.nodif.long.h0 <- res.dat.article.nodif.long[res.dat.article.nodif.long$true.beta==0,] res.dat.article.nodif.long.h0$prop.dif <- res.dat.article.nodif.long.h0$nb.dif/res.dat.article.nodif.long.h0$J res.dat.article.nodif.long.h0$prop.dif <- as.numeric(res.dat.article.nodif.long.h0$prop.dif) res.dat.article.nodif.long.h0$N <- as.numeric(res.dat.article.nodif.long.h0$N) res.dat.article.nodif.long.h0$true.gamma <- as.numeric(res.dat.article.nodif.long.h0$true.gamma) res.dat.article.nodif.long.h0$J <- as.numeric(res.dat.article.nodif.long.h0$J) res.dat.article.nodif.long.h0$true.gamma <- as.numeric(res.dat.article.nodif.long.h0$true.gamma) # typeI summary(as.numeric(res.dat.article.dif.h0.long$typeIerror)) ####### SCENARIOS AVEC TE res.dat.article.nodif.long.h1 <- res.dat.article.nodif.long[res.dat.article.nodif.long$true.beta!=0,] res.dat.article.nodif.long.h1$prop.dif <- res.dat.article.nodif.long.h1$nb.dif/res.dat.article.nodif.long.h1$J res.dat.article.nodif.long.h1$prop.dif <- as.numeric(res.dat.article.nodif.long.h1$prop.dif) res.dat.article.nodif.long.h1$N <- as.numeric(res.dat.article.nodif.long.h1$N) res.dat.article.nodif.long.h1$true.gamma <- as.numeric(res.dat.article.nodif.long.h1$true.gamma) res.dat.article.nodif.long.h1$J <- as.numeric(res.dat.article.nodif.long.h1$J) res.dat.article.nodif.long.h1$true.gamma <- as.numeric(res.dat.article.nodif.long.h1$true.gamma) res.dat.article.nodif.long.h1$powerdif <- as.numeric(res.dat.article.nodif.long.h1$power)-as.numeric(res.dat.article.nodif.long.h1$theoretical.power) # powerdif summary(as.numeric(res.dat.article.nodif.long.h1$powerdif)) ####### Overall #coverage res.dat.article.nodif.long$bias <- as.numeric(res.dat.article.nodif.long$bias) res.dat.article.nodif.long$abs.bias <- abs(res.dat.article.nodif.long$bias) summary(res.dat.article.nodif.long$abs.bias) #coverage summary(res.dat.article.nodif.long$coverage) ####### ROSALI # SCENARIOS SANS TE # typeI summary(as.numeric(res.dat.article.rosali.2.nodif$typeIerror)) res.dat.article.rosali.2.nodif$typeIerror <- as.numeric(res.dat.article.rosali.2.nodif$typeIerror) res.dat.article.rosali.2.nodif$N <- as.numeric(res.dat.article.rosali.2.nodif$N) summary(lm(typeIerror~N+J,data = res.dat.article.rosali.2.nodif)) # bias res.dat.article.rosali.2.nodif$abs.bias <- abs(as.numeric(res.dat.article.rosali.2.nodif$bias)) summary(as.numeric(res.dat.article.rosali.2.nodif[res.dat.article.rosali.2.nodif$true.beta==0,]$abs.bias)) # SCENARIOS AVEC TE # bias summary(as.numeric(res.dat.article.rosali.2.nodif[res.dat.article.rosali.2.nodif$true.beta!=0,]$abs.bias)) # coverage res.dat.article.rosali.2.nodif$coverage <- as.numeric(res.dat.article.rosali.2.nodif$coverage) summary(as.numeric(res.dat.article.rosali.2.nodif[res.dat.article.rosali.2.nodif$true.beta!=0,]$coverage)) # power res.dat.article.rosali.2.nodif$powerdif <- as.numeric(res.dat.article.rosali.2.nodif$power)-as.numeric(res.dat.article.rosali.2.nodif$theoretical.power) summary(as.numeric(res.dat.article.rosali.2.nodif[res.dat.article.rosali.2.nodif$true.beta!=0,]$powerdif)) ####### RESIDIF # SCENARIOS SANS TE # typeI summary(as.numeric(res.dat.article.residif.2.nodif$typeIerror)) res.dat.article.residif.2.nodif$typeIerror <- as.numeric(res.dat.article.residif.2.nodif$typeIerror) res.dat.article.residif.2.nodif$N <- as.numeric(res.dat.article.residif.2.nodif$N) summary(lm(typeIerror~N+J,data = res.dat.article.residif.2.nodif)) # bias res.dat.article.residif.2.nodif$abs.bias <- abs(as.numeric(res.dat.article.residif.2.nodif$bias)) summary(as.numeric(res.dat.article.residif.2.nodif[res.dat.article.residif.2.nodif$true.beta==0,]$abs.bias)) # SCENARIOS AVEC TE # bias summary(as.numeric(res.dat.article.residif.2.nodif[res.dat.article.residif.2.nodif$true.beta!=0,]$abs.bias)) # coverage res.dat.article.residif.2.nodif$coverage <- as.numeric(res.dat.article.residif.2.nodif$coverage) summary(as.numeric(res.dat.article.residif.2.nodif[res.dat.article.residif.2.nodif$true.beta!=0,]$coverage)) # power res.dat.article.residif.2.nodif$powerdif <- as.numeric(res.dat.article.residif.2.nodif$power)-as.numeric(res.dat.article.residif.2.nodif$theoretical.power) summary(as.numeric(res.dat.article.residif.2.nodif[res.dat.article.residif.2.nodif$true.beta!=0,]$powerdif)) # Scenarios avec + perf res.dat.article.ignore.long <- reshape(res.dat.article.ignore,idvar=c("J",'true.beta',"true.gamma","nb.dif","prop.dif"),v.names=c('betahat','bias','typeIerror','power',"coverage")) rownames(res.dat.article.ignore.long) <- NULL colnames(res.dat.article.ignore.long)[6:11] <- c("betahat","bias","typeIerror",'power','theoretical.power','coverage') res.dat.article.ignore.long$abs.bias <- abs(res.dat.article.ignore.long$bias) res.dat.article.residif.2$abs.bias <- abs(res.dat.article.residif.2$bias) res.dat.article.residif.2.dif <- res.dat.article.residif.2[res.dat.article.residif.2$nb.dif>0,] res.dat.article.rosali.2$abs.bias <- abs(res.dat.article.rosali.2$bias) res.dat.article.rosali.2.dif <- res.dat.article.rosali.2[res.dat.article.rosali.2$nb.dif>0,] res.dat.article.ignore.long.dif <- res.dat.article.ignore.long[res.dat.article.ignore.long$nb.dif>0,] summary(res.dat.article.residif.2.dif[res.dat.article.residif.2.dif$abs.bias-res.dat.article.ignore.long.dif$abs.bias<=-0.1,]$abs.bias) summary(res.dat.article.ignore.long.dif[res.dat.article.residif.2.dif$abs.bias-res.dat.article.ignore.long.dif$abs.bias<=-0.1,]$abs.bias) summary(res.dat.article.rosali.2.dif[res.dat.article.rosali.2.dif$abs.bias-res.dat.article.ignore.long.dif$abs.bias<=-0.1,]$abs.bias) summary(res.dat.article.ignore.long.dif[res.dat.article.rosali.2.dif$abs.bias-res.dat.article.ignore.long.dif$abs.bias<=-0.1,]$abs.bias) ########################## # BOXPLOT PERF VS NODIF FACET TYPE1 ########################## par(mfrow=c(3,4)) par(bg = "white") res.dat.article.2$abs.gamma <- abs(as.numeric(res.dat.article.2$true.gamma)) res.dat.article.rosali.2$abs.gamma <- abs(as.numeric(res.dat.article.rosali.2$true.gamma)) res.dat.article.residif.2$abs.gamma <- abs(as.numeric(res.dat.article.residif.2$true.gamma)) res.dat.article.dif.2$abs.gamma <- abs(as.numeric(res.dat.article.dif.2$true.gamma)) res.dat.article.2$prop.dif <- res.dat.article.2$nb.dif/res.dat.article.2$J res.dat.article.rosali.2$prop.dif <- res.dat.article.rosali.2$nb.dif/res.dat.article.rosali.2$J res.dat.article.residif.2$prop.dif <- res.dat.article.residif.2$nb.dif/res.dat.article.residif.2$J res.dat.article.dif.2$prop.dif <- res.dat.article.dif.2$nb.dif/res.dat.article.dif.2$J ############# Type I error # N50 ## DIF 03 25% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="Weak DIF - 25% of items", ylab="Type-I error rate",ylim=c(0,1),yaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 03 50% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="Weak DIF - 50% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 05 25% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="Medium DIF - 25% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 05 50% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="Medium DIF - 50% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) # N100 ## DIF 03 25% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab="Type-I error rate",ylim=c(0,1),yaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 03 50% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 05 25% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 05 50% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) # N300 ## DIF 03 25% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab="Type-I error rate",ylim=c(0,1),yaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 03 50% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 05 25% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ## DIF 05 50% bp.dat.typeIerror.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("typeIerror")]) bp.dat.typeIerror.nodif <- as.numeric(res.dat.article.nodif.2[,"typeIerror"]) bp.dat.typeIerror.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"typeIerror"]) bp.dat.typeIerror <- data.frame(typeIerror=c(bp.dat.typeIerror.nodif,bp.dat.typeIerror.ignore,bp.dat.typeIerror.rosali,bp.dat.typeIerror.residif, bp.dat.typeIerror.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.typeIerror$method <- factor(bp.dat.typeIerror$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.typeIerror$typeIerror~bp.dat.typeIerror$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.05,lty=2,col='#777777',lwd=2) ########################## # BOXPLOT PERF VS NODIF FACET BIAS ########################## par(mfrow=c(3,4)) par(bg = "white") res.dat.article.2$abs.gamma <- abs(as.numeric(res.dat.article.2$true.gamma)) res.dat.article.rosali.2$abs.gamma <- abs(as.numeric(res.dat.article.rosali.2$true.gamma)) res.dat.article.residif.2$abs.gamma <- abs(as.numeric(res.dat.article.residif.2$true.gamma)) res.dat.article.dif.2$abs.gamma <- abs(as.numeric(res.dat.article.dif.2$true.gamma)) res.dat.article.2$prop.dif <- res.dat.article.2$nb.dif/res.dat.article.2$J res.dat.article.rosali.2$prop.dif <- res.dat.article.rosali.2$nb.dif/res.dat.article.rosali.2$J res.dat.article.residif.2$prop.dif <- res.dat.article.residif.2$nb.dif/res.dat.article.residif.2$J res.dat.article.dif.2$prop.dif <- res.dat.article.dif.2$nb.dif/res.dat.article.dif.2$J res.dat.article.2$abs.bias <- abs(as.numeric(res.dat.article.2$bias)) res.dat.article.rosali.2$abs.bias <- abs(as.numeric(res.dat.article.rosali.2$bias)) res.dat.article.residif.2$abs.bias <- abs(as.numeric(res.dat.article.residif.2$bias)) res.dat.article.dif.2$abs.bias <- abs(as.numeric(res.dat.article.dif.2$bias)) res.dat.article.nodif.2$abs.bias <- abs(as.numeric(res.dat.article.nodif.2$bias)) ############# Bias # N50 ## DIF 03 25% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="Weak DIF - 25% of items", ylab="Absolute bias",ylim=c(0,0.5),yaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 03 50% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="Weak DIF - 50% of items", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 05 25% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="Medium DIF - 25% of items", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 05 50% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="Medium DIF - 50% of items", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) # N100 ## DIF 03 25% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab="Absolute bias",ylim=c(0,0.5),yaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 03 50% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 05 25% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 05 50% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) # N300 ## DIF 03 25% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab="Absolute bias",ylim=c(0,0.5),yaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 03 50% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 05 25% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ## DIF 05 50% bp.dat.abs.bias.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("abs.bias")]) bp.dat.abs.bias.nodif <- as.numeric(res.dat.article.nodif.2[,"abs.bias"]) bp.dat.abs.bias.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"abs.bias"]) bp.dat.abs.bias <- data.frame(abs.bias=c(bp.dat.abs.bias.nodif,bp.dat.abs.bias.ignore,bp.dat.abs.bias.rosali,bp.dat.abs.bias.residif, bp.dat.abs.bias.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.abs.bias$method <- factor(bp.dat.abs.bias$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.abs.bias$abs.bias~bp.dat.abs.bias$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,0.5),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) ########################## # BOXPLOT PERF VS NODIF FACET COVERAGE ########################## par(mfrow=c(3,4)) par(bg = "white") res.dat.article.2$abs.gamma <- abs(as.numeric(res.dat.article.2$true.gamma)) res.dat.article.rosali.2$abs.gamma <- abs(as.numeric(res.dat.article.rosali.2$true.gamma)) res.dat.article.residif.2$abs.gamma <- abs(as.numeric(res.dat.article.residif.2$true.gamma)) res.dat.article.dif.2$abs.gamma <- abs(as.numeric(res.dat.article.dif.2$true.gamma)) res.dat.article.2$prop.dif <- res.dat.article.2$nb.dif/res.dat.article.2$J res.dat.article.rosali.2$prop.dif <- res.dat.article.rosali.2$nb.dif/res.dat.article.rosali.2$J res.dat.article.residif.2$prop.dif <- res.dat.article.residif.2$nb.dif/res.dat.article.residif.2$J res.dat.article.dif.2$prop.dif <- res.dat.article.dif.2$nb.dif/res.dat.article.dif.2$J ############# Bias # N50 ## DIF 03 25% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="Weak DIF - 25% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) title(ylab="Coverage",cex.lab=1.6) ## DIF 03 50% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="Weak DIF - 50% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ## DIF 05 25% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="Medium DIF - 25% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ## DIF 05 50% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="Medium DIF - 50% of items", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) # N100 ## DIF 03 25% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab="",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) title(ylab="Coverage",cex.lab=1.6) ## DIF 03 50% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ## DIF 05 25% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ## DIF 05 50% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) # N300 ## DIF 03 25% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) title(ylab="Coverage",cex.lab=1.6) ## DIF 03 50% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ## DIF 05 25% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ## DIF 05 50% bp.dat.coverage.ignore <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif>0.3 & res.dat.article.2$nb.dif!=0,c("coverage")]) bp.dat.coverage.nodif <- as.numeric(res.dat.article.nodif.2[,"coverage"]) bp.dat.coverage.rosali <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif>0.3 & res.dat.article.rosali.2$nb.dif!=0,"coverage"]) bp.dat.coverage.residif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif>0.3 & res.dat.article.residif.2$nb.dif!=0,"coverage"]) bp.dat.coverage.dif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif>0.3 & res.dat.article.dif.2$nb.dif!=0,"coverage"]) bp.dat.coverage <- data.frame(coverage=c(bp.dat.coverage.nodif,bp.dat.coverage.ignore,bp.dat.coverage.rosali,bp.dat.coverage.residif, bp.dat.coverage.dif),method=c(rep("NO DIF",18), rep("IGNORE-DIF",10), rep("ROSALI",10), rep("RESIDIF",10), rep("PCM-DIF",10) ) ) bp.dat.coverage$method <- factor(bp.dat.coverage$method,levels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF")) boxplot(bp.dat.coverage$coverage~bp.dat.coverage$method,xlab=" ",pch=3,main="", ylab=" ",ylim=c(0,1),yaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.45, col=c("#e69875","#a7c080","#a7c080","#a7c080","#a7c080") ,border=c("#CD5E35","#697850","#697850","#697850","#697850")) axis(2,seq(0,1,0.1),cex.axis=1.45) abline(h=0.95,lty=2,col='#777777',lwd=2) ########################## # BOXPLOT PERF VS NODIF FACET POWER ########################## par(mfrow=c(3,4)) par(bg = "white") res.dat.article.2$abs.gamma <- abs(as.numeric(res.dat.article.2$true.gamma)) res.dat.article.rosali.2$abs.gamma <- abs(as.numeric(res.dat.article.rosali.2$true.gamma)) res.dat.article.residif.2$abs.gamma <- abs(as.numeric(res.dat.article.residif.2$true.gamma)) res.dat.article.dif.2$abs.gamma <- abs(as.numeric(res.dat.article.dif.2$true.gamma)) res.dat.article.2$prop.dif <- res.dat.article.2$nb.dif/res.dat.article.2$J res.dat.article.rosali.2$prop.dif <- res.dat.article.rosali.2$nb.dif/res.dat.article.rosali.2$J res.dat.article.residif.2$prop.dif <- res.dat.article.residif.2$nb.dif/res.dat.article.residif.2$J res.dat.article.dif.2$prop.dif <- res.dat.article.dif.2$nb.dif/res.dat.article.dif.2$J # N50 ## DIF 03 25% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="Weak DIF - 25% of items", ylab="RCT power - expected power",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 03 50% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="Weak DIF - 50% of items", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 05 25% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="Medium DIF - 25% of items", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 05 50% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==50 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==50 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==50 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==50 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="Medium DIF - 50% of items", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) # N100 ## DIF 03 25% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="RCT power - expected power",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 03 50% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 05 25% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 05 50% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==100 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==100 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==100 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==100 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) # N300 ## DIF 03 25% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="RCT power - expected power",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.6,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 03 50% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.3 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.3 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.3 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.3 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 05 25% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) ## DIF 05 50% bp.dat.power.ignore.mask <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma>0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.ignore.magnif <- as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.2[res.dat.article.2$N==300 & res.dat.article.2$abs.gamma==0.5 & res.dat.article.2$prop.dif<0.3 & res.dat.article.2$true.gamma<0 & res.dat.article.2$true.beta>0 & res.dat.article.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.nodif <- as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"power"])-as.numeric(res.dat.article.nodif.2[res.dat.article.nodif.2$true.beta>0,"theoretical.power"]) bp.dat.power.rosali.mask <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma>0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.rosali.magnif <- as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.rosali.2[res.dat.article.rosali.2$N==300 & res.dat.article.rosali.2$abs.gamma==0.5 & res.dat.article.rosali.2$prop.dif<0.3 & res.dat.article.rosali.2$true.beta>0 & res.dat.article.rosali.2$true.gamma<0 & res.dat.article.rosali.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.mask <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma>0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.residif.magnif <- as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.residif.2[res.dat.article.residif.2$N==300 & res.dat.article.residif.2$abs.gamma==0.5 & res.dat.article.residif.2$prop.dif<0.3 & res.dat.article.residif.2$true.beta>0 & res.dat.article.residif.2$true.gamma<0 & res.dat.article.residif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.mask <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma>0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power.dif.magnif <- as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"power"])-as.numeric(res.dat.article.dif.2[res.dat.article.dif.2$N==300 & res.dat.article.dif.2$abs.gamma==0.5 & res.dat.article.dif.2$prop.dif<0.3 & res.dat.article.dif.2$true.beta>0 & res.dat.article.dif.2$true.gamma<0 & res.dat.article.dif.2$nb.dif!=0,"theoretical.power"]) bp.dat.power <- data.frame(power=c(bp.dat.power.nodif,bp.dat.power.ignore.mask,bp.dat.power.ignore.magnif,bp.dat.power.rosali.mask,bp.dat.power.rosali.magnif,bp.dat.power.residif.mask,bp.dat.power.residif.magnif, bp.dat.power.dif.mask,bp.dat.power.dif.magnif), method=c(rep("NO DIF",12), rep("MASK1",4),rep("AMPLIFY1",4), rep("MASK2",4),rep("AMPLIFY2",4), rep("MASK3",4),rep("AMPLIFY3",4), rep("MASK4",4),rep("AMPLIFY4",4) )) bp.dat.power$method <- factor(bp.dat.power$method,levels=c("NO DIF","MASK1","AMPLIFY1","MASK2","AMPLIFY2","MASK3","AMPLIFY3","MASK4","AMPLIFY4")) boxplot(bp.dat.power$power~bp.dat.power$method,xlab="",pch=3,main="", ylab="",ylim=c(-1,1),yaxt="n",xaxt="n", cex.lab=1.45,cex.main=1.5,cex.axis=1.15, col=c("#e69875","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5","#798A5D","#D4E8B5") ,border=c("#CD5E35","#697850","#697850","#697850","#697850","#697850","#697850","#697850","#697850"), width=c(0.8,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4), at=c(1,2,2.5,3.25,3.75,4.5,5,5.75,6.25)) axis(2,seq(-1,1,0.25),cex.axis=1.45) axis(1,c(1,2.25,3.5,4.75,6),labels=c("NO DIF","IGNORE-DIF","ROSALI","RESIDIF","PCM-DIF"),cex.axis=1.45) abline(h=0,lty=2,col='#595959',lwd=2) par(mfrow=c(1,1)) ########################## # ICC DIF ########################## # DIF HOMOGENE zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N300/scenario_XA_300.csv") zaza0 <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza1 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) icc.tam.difh <- function(x, items=1:x$nitems, type="expected",TT=F, low=-4, high=4, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#f7c611', '#61b5b7', '#ca346d', '#204776' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(-3, 3, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) if (!TT) { graphics::plot(theta, expScore[,i], type="l", lwd=3, las=1, ylab="Expected item response", main=NULL, col='#204776',ylim=ylim2,xlim=c(-4,4),xlab="", ... ) } else { lines(theta,expScore[,i], type="l", lwd=3, las=1,col='#ca346d') } } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } #graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab=expression(theta), col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/icc_difh.pdf') par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.difh(zaza0,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=F) icc.tam.difh(zaza1,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=T) axis(1) axis(2,las=2) title(main='A \n \n ', font.main=2) arrows(x0=-3.65,x1=-4,y0=-0.55,length = 0.15,lwd = 2) arrows(x0=3.65,x1=4,y0=-0.55,length = 0.15,lwd = 2) text(x=-3.55,y=-0.55,"Worse\nmental\nhealth",adj=0) text(x=3.5,y=-0.55,"Better\nmental\nhealth",adj=1) arrows(x1=-.65,x0=.15,y0=1.5,length = 0.15,lwd = 2) arrows(x1=-1.7,x0=-0.9,y0=.75,length = 0.15,lwd = 2) arrows(x1=0.4,x0=1.2,y0=2.25,length = 0.15,lwd = 2) legend(y=0.5,x=1.5,lty=c(1,1),lwd=c(2,2),col=c('#204776', '#ca346d'),c("Control group","Treatment group")) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() # DIF CONVERGENT zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N300/scenario_YA_300.csv") zaza0 <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza1 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/icc_difnh1.pdf') par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.difh(zaza0,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=F) icc.tam.difh(zaza1,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=T) axis(1) axis(2,las=2) title(main='B \n \n Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) arrows(x0=-3.65,x1=-4,y0=-0.55,length = 0.15,lwd = 2) arrows(x0=3.65,x1=4,y0=-0.55,length = 0.15,lwd = 2) text(x=-3.55,y=-0.55,"Worse\nmental\nhealth",adj=0) text(x=3.5,y=-0.55,"Better\nmental\nhealth",adj=1) arrows(x1=-0.1,x0=1.2,y0=2.25,length = 0.15,lwd = 2) arrows(x1=-.9,x0=.15,y0=1.5,length = 0.15,lwd = 2) arrows(x1=-1.65,x0=-0.9,y0=.75,length = 0.15,lwd = 2) legend(y=0.5,x=1.5,lty=c(1,1),lwd=c(2,2),col=c('#204776', '#ca346d'),c("Control group","Treatment group")) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() # DIF DIVERGENT zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N300/scenario_ZA_300.csv") zaza0 <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza1 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) pdf(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PDF/icc_difnh2.pdf') par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.difh(zaza0,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=F) icc.tam.difh(zaza1,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=T) axis(1) axis(2,las=2) title(main='C \n \n ', font.main=2) arrows(x0=-3.65,x1=-4,y0=-0.55,length = 0.15,lwd = 2) arrows(x0=3.65,x1=4,y0=-0.55,length = 0.15,lwd = 2) text(x=-3.55,y=-0.55,"Worse\nmental\nhealth",adj=0) text(x=3.5,y=-0.55,"Better\nmental\nhealth",adj=1) legend(y=0.5,x=1.5,lty=c(1,1),lwd=c(2,2),col=c('#204776', '#ca346d'),c("Control group","Treatment group")) arrows(x1=1.9,x0=1.4,y0=2.25,length = 0.15,lwd = 2) arrows(x1=-1.4,x0=-.9,y0=0.75,length = 0.15,lwd = 2) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() #### ICC DIF BASE zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N300/scenario_XB_300.csv") zaza <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza2 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) zaza3 <- tam.mml(resp = zaz[,paste0("item",1:4)]) icc.tam <- function(x, items=1:x$nitems, type="expected",TT=F, low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#200c23', '#62403d', '#a87b5e', '#e9bf98' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(-3, 3, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) if (TT==0) { graphics::plot(theta, expScore[,i], type="l", lwd=3, las=1, ylab="Expected item response", main=NULL,xlab="", col="#204776",ylim=ylim2,xlim= ... ) } else { if (TT==1) { lines(theta,expScore[,i], type="l", lwd=3, las=1,col="#ca346d") } else { lines(theta,expScore[,i], type="l", lwd=2, las=1,col="#999999",lty=2) } } } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } #graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab=expression(theta), col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/icc_beta.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam(zaza,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=0) icc.tam(zaza2,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=1) icc.tam(zaza3,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=2) axis(1) axis(2,las=2) title(main='A - Item Characteristic Curve \n Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) arrows(x0=-2.65,x1=-3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.45,length = 0.15,lwd = 2) text(x=-2.55,y=-0.45,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.45,"Better\nmental\nhealth",adj=1) arrows(x0=0.175,x1=-0.175,y0=1.5,length = 0.15,lwd = 2,col="black") text(x=0.05,y=1.575,"DIF",col="black") legend(y=3,x=-3,lty=c(1,1,2),lwd=c(2,2,2),col=c('#204776', '#ca346d',"#999999"),c("Control group","Treatment group","True ICC")) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() icc.tam.2 <- function(x, items=1:x$nitems, type="expected",TT=F, low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#204776', '#ca346d' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(-3, 3, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) if (TT==0) { graphics::plot(theta, expScore[,i], type="l", lwd=3, las=1, ylab="Expected item response", main=NULL,xlab="", col="#ca346d",ylim=ylim2,xlim= ... ) } else { if (TT==1) { lines(theta,expScore[,i], type="l", lwd=3, las=1,col="#204776") } else { lines(theta,expScore[,i], type="l", lwd=2, las=1,col="#999999",lty=2) } } } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } #graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab=expression(theta), col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/icc_beta2.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.2(zaza,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=0) icc.tam.2(zaza2,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=1) icc.tam.2(zaza3,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=2) axis(1) axis(2,las=2) title(main='A - Item Characteristic Curve \n Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) arrows(x0=-2.65,x1=-3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.45,length = 0.15,lwd = 2) text(x=-2.55,y=-0.45,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.45,"Better\nmental\nhealth",adj=1) arrows(x1=0.175,x0=-0.175,y0=1.5,length = 0.15,lwd = 2,col="black") text(x=0.05,y=1.575,"DIF",col="black") legend(y=3,x=-3,lty=c(1,1,2),lwd=c(2,2,2),col=c('#204776', '#ca346d',"#999999"),c("Control group","Treatment group","True ICC")) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() #### Densité theta beta0 png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/densite_theta_null.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) plot(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 0,sd=1),type="l",lwd=2,col="#204776",axes=F, xlab="",ylab="Probability density") axis(1) axis(2,las=2) title(main='B - true vs. observed latent variable density \n Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2,line=2.8) lines(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 0,sd=1),type="l",lwd=0.5,col="#ca346d",lty=1) text(x=0,y=0.4225,expression(beta==0),adj=0.5) legend(y=0.38,x=-3,lty=c(1,1,2,2),lwd=c(2,2,2,2),cex=1,col=c('#204776', '#ca346d','#204776', '#ca346d'),c("Control group (true)","Treatment group (true)","Control group (observed)","Treatment group (observed)")) lines(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 1,sd=1),type="l",lwd=2,col="#ca346d",lty=2) arrows(x0=0,x1=1,y0=.4125,length = 0.1,lwd = 2,col="black") text(x=0.5,y=0.425,expression(widehat(beta)>0),adj=0.5,col="black") segments(x0=0,x1=0,y0=0.4,y1=0.4125,lty=3,col="black") segments(x0=1,x1=1,y0=0.4,y1=0.4125,lty=3,col="black") arrows(x0=-2.65,x1=-3,y0=-0.05,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.05,length = 0.15,lwd = 2) text(x=-2.55,y=-0.05,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.05,"Better\nmental\nhealth",adj=1) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() #### Densité theta mask png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/densite_theta_masks.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) plot(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 0,sd=1),type="l",lwd=2,col="#204776",axes=F, xlab="",ylab="Probability density") axis(1) axis(2,las=2) title(main='B - true vs. observed latent variable density \n Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2,line = 3.4) lines(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 1,sd=1),type="l",lwd=2,col="#ca346d",lty=1) lines(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 0.5,sd=1),type="l",lwd=2,col="#ca346d",lty=2) arrows(x0=0,x1=1,y0=.4325,length = 0.1,lwd = 2,col="black") text(x=0.5,y=0.4425,expression(beta),adj=0.5) segments(x0=0,x1=0,y0=0.4,y1=0.4325,lty=3) segments(x0=1,x1=1,y0=0.4,y1=0.4325,lty=3) arrows(x0=1,x1=0.51,y0=.415,length = 0.1,lwd = 2,col="black") text(x=0.75,y=0.4225,"DIF",adj=0.5,col="black") segments(x0=0.5,x1=0.5,y0=0.4,y1=.415,lty=3,col="black") arrows(x0=0,x1=0.49,y0=.405,length = 0.1,lwd = 2,col="black") text(x=0.25,y=0.4175,expression(widehat(beta)),adj=0.5,col="black") segments(x0=0,x1=0,y0=0.4,y1=.405,lty=3,col="black") segments(x0=0.5,x1=0.5,y0=0.4,y1=.405,lty=3,col="black") title(xlab=expression("Latent variable " * theta * "\n (mental health)")) legend(y=0.38,x=-3,lty=c(1,1,2,2),lwd=c(2,2,2,2),cex=1,col=c('#204776', '#ca346d','#204776', '#ca346d'),c("Control group (true)","Treatment group (true)","Control group (observed)","Treatment group (observed)")) arrows(x0=-2.65,x1=-3,y0=-0.05,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.05,length = 0.15,lwd = 2) text(x=-2.55,y=-0.05,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.05,"Better\nmental\nhealth",adj=1) par(xpd=F) dev.off() #### Densité theta amplify png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/densite_theta_amplify.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) plot(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 0,sd=1),type="l",lwd=2,col="#204776",axes=F, xlab="",ylab="Probability density") title(xlab=expression("Latent variable " * theta * "\n (mental health)")) axis(1) axis(2,las=2) title(main='B - true vs. observed latent variable density \n Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2,line = 3.4) lines(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 1,sd=1),type="l",lwd=2,col="#ca346d",lty=1) lines(seq(-3,3,0.001),dnorm(seq(-3,3,0.001),mean = 1.5,sd=1),type="l",lwd=2,col="#ca346d",lty=2) arrows(x0=0,x1=0.98,y0=.4325,length = 0.1,lwd = 2,col="black") text(x=0.5,y=0.4425,expression(beta),adj=0.5) segments(x0=0,x1=0,y0=0.4,y1=0.4325,lty=3) segments(x0=1,x1=1,y0=0.4,y1=0.4325,lty=3) arrows(x0=1.02,x1=1.5,y0=.4325,length = 0.1,lwd = 2,col="black") text(x=1.25,y=0.4425,"DIF",adj=0.5,col="black") segments(x0=1.5,x1=1.5,y0=0.4,y1=.4325,lty=3,col="black") arrows(x0=0,x1=1.5,y0=.41,length = 0.1,lwd = 2,col="black") text(x=0.75,y=0.4215,expression(widehat(beta)),adj=0.5,col="black") segments(x0=0,x1=0,y0=0.4,y1=.41,lty=3,col="black") segments(x0=1.5,x1=1.5,y0=0.4,y1=.41,lty=3,col="black") legend(y=0.38,x=-3,lty=c(1,1,2,2),lwd=c(2,2,2,2),cex=1,col=c('#204776', '#ca346d','#204776', '#ca346d'),c("Control group (true)","Treatment group (true)","Control group (observed)","Treatment group (observed)")) arrows(x0=-2.65,x1=-3,y0=-0.05,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.05,length = 0.15,lwd = 2) text(x=-2.55,y=-0.05,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.05,"Better\nmental\nhealth",adj=1) par(xpd=F) dev.off() #### ICC DIF SUPPL RESIDIF 1 zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N300/scenario_XB_300.csv") zaza <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza2 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) zaza3 <- tam.mml(resp = zaz[,paste0("item",1:4)]) icc.tam.sup <- function(x, items=1:x$nitems, type="expected",TT=F, low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#200c23', '#62403d', '#a87b5e', '#e9bf98' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(-3, 3, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) if (TT==0) { graphics::plot(theta, expScore[,i]-0.2, type="l", lwd=3, las=1, ylab="Expected item response", main=NULL,xlab="", col="#204776",ylim=ylim2,xlim= ... ) } else { if (TT==1) { lines(theta,expScore[,i]+0.2, type="l", lwd=3, las=1,col="#ca346d") } else { lines(theta,expScore[,i], type="l", lwd=2, las=1,col="#999999",lty=2) } } } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } #graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab=expression(theta), col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/icc_residif_1.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.sup(zaza,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=0) icc.tam.sup(zaza2,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=1) icc.tam.sup(zaza3,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=2) axis(1) axis(2,las=2) title(main='Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) arrows(x0=-2.65,x1=-3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=-2,x1=-2,y1=0.375,y0=0.575,length = 0.15,lwd = 2,col="#ca346d") arrows(x0=0,x1=0,y1=1.6,y0=1.8,length = 0.15,lwd = 2,col="#ca346d") arrows(x0=2,x1=2,y1=2.7,y0=2.9,length = 0.15,lwd = 2,col="#ca346d") text(x=-2,y=0.725,col="#ca346d","-",cex=1.5) text(x=0,y=2.0,col="#ca346d","-",cex=1.5) text(x=2,y=3,col="#ca346d","-",cex=1.5) text(x=-2,y=0.0,col="#204776","+",cex=1.5) text(x=0,y=0.975,col="#204776","+",cex=1.5) text(x=2,y=2.3,col="#204776","+",cex=1.5) arrows(x0=-2,x1=-2,y1=0.305,y0=0.105,length = 0.15,lwd = 2,col="#204776") arrows(x0=0,x1=0,y1=1.4,y0=1.2,length = 0.15,lwd = 2,col="#204776") arrows(x0=2,x1=2,y1=2.625,y0=2.425,length = 0.15,lwd = 2,col="#204776") text(x=-2.55,y=-0.45,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.45,"Better\nmental\nhealth",adj=1) legend(cex=0.9,y=3,x=-3,lty=c(1,1,2,NA,NA),lwd=c(2,2,2,NA,NA),pch = c(NA,NA,NA,NA,NA),col=c('#204776', '#ca346d',"#999999",NA,NA),c("Control group ICC","Treatment group ICC","Overall ICC","Person-item residuals - Control","Person-item residuals - Treatment")) par(font=5) legend(cex=0.9,y=3,x=-3,lty=c(NA,NA,NA,NA,NA),lwd=c(NA,NA,NA,NA,NA),pt.cex=c(NA,NA,NA,1.5,1.5),pch = c(NA,NA,NA,174,174),col=c(NA,NA,NA,'#204776', '#ca346d'),c(NA,NA,NA,NA,NA),bty="n") par(font=1) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() #### ICC DIF SUPPL RESIDIF 2 zaz <- read.csv("/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N300/scenario_YA_300.csv") zaza <- tam.mml(resp = zaz[zaz$TT==0,paste0("item",1:4)]) zaza2 <- tam.mml(resp = zaz[zaz$TT==1,paste0("item",1:4)]) zaza3 <- tam.mml(resp = zaz[,paste0("item",1:4)]) icc.tam.sup <- function(x, items=1:x$nitems, type="expected",TT=F, low=-3, high=3, ngroups=6, groups_by_item=FALSE, wle=NULL, export=TRUE, export.type="png", export.args=list(), observed=TRUE, overlay=FALSE, ask=FALSE, package="lattice", fix.devices=TRUE, nnodes=100, ...) { require_namespace_msg("grDevices") if ( package=="lattice"){ require_namespace_msg("lattice") } # device.Option <- getOption("device") time1 <- NULL pall <- c('#200c23', '#62403d', '#a87b5e', '#e9bf98' ) if ( fix.devices ){ old.opt.dev <- getOption("device") old.opt.err <- c( getOption("show.error.messages")) old.par.ask <- graphics::par("ask") # remember new pars' values old.par.xpd <- graphics::par("xpd") old.par.mar <- graphics::par("mar") on.exit( options("device"=old.opt.dev)) on.exit( options("show.error.messages"=old.opt.err), add=TRUE) on.exit( graphics::par("ask"=old.par.ask), add=TRUE) # restore new pars' values on.exit( graphics::par("xpd"=old.par.xpd), add=TRUE) on.exit( graphics::par("mar"=old.par.mar), add=TRUE) } tamobj <- x ndim <- tamobj$ndim tammodel <- "mml" if(is.null(ndim)) { ndim <- 1 tammodel <- "jml" } if (ndim > 1 ) { if ( type=="expected"){ stop ("Expected scores curves are only available for uni-dimensional models") } } nitems <- tamobj$nitems if (ndim==1 ){ theta <- matrix(seq(low, high, length=nnodes), nrow=nnodes, ncol=ndim) } else { nodes <- seq(-3, 3, length=nnodes) theta <- as.matrix( expand.grid( as.data.frame( matrix( rep(nodes, ndim), ncol=ndim ) ) ) ) nnodes <- nrow(theta) B <- tamobj$B } iIndex <- 1:nitems A <- tamobj$A B <- tamobj$B if (tammodel=="mml") { xsi <- tamobj$xsi$xsi } else { xsi <- tamobj$xsi } maxK <- tamobj$maxK resp <- tamobj$resp resp.ind <- tamobj$resp.ind resp[resp.ind==0] <- NA AXsi <- matrix(0,nrow=nitems,ncol=maxK ) res <- tam_mml_calc_prob(iIndex=iIndex, A=A, AXsi=AXsi, B=B, xsi=xsi, theta=theta, nnodes=nnodes, maxK=maxK, recalc=TRUE ) rprobs <- res[["rprobs"]] AXsi <- res[["AXsi"]] cat <- 1:maxK - 1 #@@@ define initial empty objects expScore <- obScore <- wle_intervals <- NULL theta2 <- NULL #**** type='expected' if ( type=="expected" ){ expScore <- sapply(1:nitems, function(i) colSums(cat*rprobs[i,,], na.rm=TRUE)) #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups wle_intervals <- res$wle_intervals #-- compute observed scores obScore <- apply(d2,2, function(x){ stats::aggregate(x, list(groupnumber), mean, na.rm=TRUE) } ) } #---------------------------------------------------- # adds observed score for type="items" if (type=="items") { require_namespace_msg("plyr") #-- compute WLE score groups res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle, ngroups=ngroups, resp=resp ) wle <- res$wle theta2 <- res$theta2 d <- res$d d1 <- res$d1 d2 <- res$d2 groupnumber <- res$groupnumber ngroups <- res$ngroups obScore <- lapply(d2, function(item) { comp_case=stats::complete.cases(item) item=item[comp_case] uniq_cats=sort(unique(item)) plyr::ldply(split(item, groupnumber[comp_case]), .id="group", function (group) { ngroup=length(group) cat_freq=list() for (catt in uniq_cats) { cat_freq[[paste0("cat_", catt)]]=sum(group==catt)/ngroup } data.frame(cat_freq) }) }) } #************************************************* # begin plot function probs_plot <- as.list(1:nitems) names(probs_plot) <- items for (i in (1:nitems)[items]) { #*********************************************************** #** expected item response curves if ( type=="expected"){ if (i==1 || !overlay) { ylim2 <- c(0,max( tamobj$resp[,i], na.rm=TRUE ) ) if (TT==0) { graphics::plot(theta, expScore[,i]-0.2, type="l", lwd=3, las=1, ylab="Expected item response", main=NULL,xlab="", col="#204776",ylim=ylim2,xlim= ... ) } else { if (TT==1) { lines(theta,expScore[,i]+0.2, type="l", lwd=3, las=1,col="#ca346d") } else { lines(theta,expScore[,i], type="l", lwd=2, las=1,col="#999999",lty=2) } } } else { graphics::lines(theta, expScore[,i],type="l", col=i, lwd=3, pch=1) } if (observed){ theta2_i <- theta2 obScore_i <- obScore[[i]]$x if (groups_by_item){ ind_i <- ! is.na(resp[,i]) resp_i <- resp[ind_i, i, drop=FALSE] wle_i <- wle[ ind_i ] res <- plot_tam_grouped_wle( tamobj=tamobj, tammodel=tammodel, wle=wle_i, ngroups=ngroups, resp=resp_i ) theta2_i <- res$theta2 groupnumber_i <- res$groupnumber aggr <- stats::aggregate(resp_i, list(groupnumber_i), mean, na.rm=TRUE ) obScore_i <- aggr[,2] } #graphics::lines(theta2_i, obScore_i, type="o", lwd=2, pch=1) } } #*********************************************************** if ( ndim==1 ){ theta0 <- theta } if ( type=="items"){ rprobs.ii <- rprobs[i,,] rprobs.ii <- rprobs.ii[ rowMeans( is.na(rprobs.ii) ) < 1, ] K <- nrow(rprobs.ii) dat2 <- NULL #************ if ( ndim > 1 ){ B.ii <- B[i,,] ind.ii <- which( colSums( B.ii ) > 0 )[1] rprobs0.ii <- rprobs.ii rprobs0.ii <- stats::aggregate( t(rprobs0.ii), list( theta[,ind.ii] ), mean ) theta0 <- rprobs0.ii[,1,drop=FALSE] rprobs.ii <- t( rprobs0.ii[,-1] ) } probs_plot[[i]] <- rprobs.ii #************** for (kk in 1:K){ dat2a <- data.frame( "Theta"=theta0[,1], "cat"=kk, "P"=rprobs.ii[kk,] ) dat2 <- rbind(dat2, dat2a) } auto.key <- NULL simple.key <- paste0("Cat", 1:K - 1) auto.key <- simple.key dat2$time <- dat2$cat dat2$time1 <- paste0("Cat", dat2$time ) simple.key <- FALSE Kpercol <- K # package graphics if ( package=="graphics" ){ kk <- 1 dfr <- dat2 dfr1a <- dfr[ dfr$cat==kk, ] graphics::plot( dfr1a$Theta, dfr1a$P, ylim=c(-.1,1.1), xlab=expression(theta), col=pall[kk], type="l", xpd=TRUE,axes=F, ... ) axis(1) axis(2) grid(nx = NA, ny = NULL, lty = 3, col = "lightgray", lwd = 1) graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk], type="l", ... ) for (kk in seq(2,K) ){ dfr1a <- dfr[ dfr$cat==kk, ] graphics::lines( dfr1a$Theta, dfr1a$P, col=pall[kk] ) # graphics::points( dfr1a$Theta, dfr1a$P, pch=kk, col=kk+1 ) } } #*************************************** } #*************** graphics::par(ask=ask) } # end item ii #************************************************* } png(file = '/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Figures/PNG/icc_residif_2.png',width=800,height=800) par(xpd=T,mar=c(6.1,5.1,7.6,2.1)) icc.tam.sup(zaza,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=0) icc.tam.sup(zaza2,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=1) icc.tam.sup(zaza3,type = "expected",export=F,package = "graphics",items = 3,axes=F,TT=2) axis(1) axis(2,las=2) title(main='Item: \n "Have you been able to enjoy your normal daily activities?" ', font.main=2) arrows(x0=-2.65,x1=-3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=2.65,x1=3,y0=-0.45,length = 0.15,lwd = 2) arrows(x0=-2,x1=-2,y1=0.375,y0=0.675,length = 0.15,lwd = 2,col="#ca346d") arrows(x0=0,x1=0,y1=1.9,y0=2.45,length = 0.15,lwd = 2,col="#ca346d") arrows(x0=2,x1=2,y1=2.875,y0=3.075,length = 0.15,lwd = 2,col="#ca346d") text(x=-2,y=0.85,col="#ca346d","-",cex=1.5) text(x=0,y=2.65,col="#ca346d","-",cex=1.5) text(x=2,y=3.15,col="#ca346d","-",cex=1.5) text(x=-2,y=0.0,col="#204776","+",cex=1.5) text(x=0,y=0.975,col="#204776","+",cex=1.5) text(x=2,y=2.3,col="#204776","+",cex=1.5) arrows(x0=-2,x1=-2,y1=0.305,y0=0.105,length = 0.15,lwd = 2,col="#204776") arrows(x0=0,x1=0,y1=1.75,y0=1.15,length = 0.15,lwd = 2,col="#204776") arrows(x0=2,x1=2,y1=2.8,y0=2.425,length = 0.15,lwd = 2,col="#204776") text(x=-2.55,y=-0.45,"Worse\nmental\nhealth",adj=0) text(x=2.5,y=-0.45,"Better\nmental\nhealth",adj=1) legend(cex=0.8,y=3,x=-3,lty=c(1,1,2,NA,NA),lwd=c(2,2,2,NA,NA),pch = c(NA,NA,NA,NA,NA),col=c('#204776', '#ca346d',"#999999",NA,NA),c("Control group ICC","Treatment group ICC","Overall ICC","Person-item residuals - Control","Person-item residuals - Treatment")) par(font=5) legend(cex=0.8,y=3,x=-3,lty=c(NA,NA,NA,NA,NA),lwd=c(NA,NA,NA,NA,NA),pt.cex=c(NA,NA,NA,1.5,1.5),pch = c(NA,NA,NA,174,174),col=c(NA,NA,NA,'#204776', '#ca346d'),c(NA,NA,NA,NA,NA),bty="n") par(font=1) title(xlab=expression("Latent variable " * theta * "\n (mental health)")) par(xpd=F) dev.off() ########################## # DETECTION PERF NODIF ########################## #### res.dat ROSALI res.dat.sup.rosali <- res.dat.dif.rosali[,c("N","J","eff.size","nb.dif","dif.size", "prop.perfect","flexible.detect","moreflexible.detect","dif.detected")] colnames(res.dat.sup.rosali)[3] <- "true.beta" colnames(res.dat.sup.rosali)[5] <- "true.gamma" colnames(res.dat.sup.rosali)[6] <- "perfect" colnames(res.dat.sup.rosali)[7] <- "flexible" colnames(res.dat.sup.rosali)[8] <- "mostflexible" res.dat.sup.rosali[res.dat.sup.rosali$nb.dif==0,"true.gamma"] <- NA res.dat.sup.rosali[is.na(res.dat.sup.rosali)] <- " " res.dat.sup.rosali.w <- reshape(res.dat.sup.rosali, direction = "wide", idvar = c("J","true.beta","nb.dif",'true.gamma'),timevar = "N" ) res.dat.sup.rosali.nodif <- res.dat.sup.rosali.w[res.dat.sup.rosali.w$nb.dif==0,] res.dat.sup.rosali.dif <- res.dat.sup.rosali.w[res.dat.sup.rosali.w$nb.dif!=0,] #### res.dat RESIDIF res.dat.sup.residif <- res.dat.dif.resali[,c("N","J","eff.size","nb.dif","dif.size", "prop.perfect","flexible.detect","moreflexible.detect","dif.detected")] colnames(res.dat.sup.residif)[3] <- "true.beta" colnames(res.dat.sup.residif)[5] <- "true.gamma" colnames(res.dat.sup.residif)[6] <- "perfect" colnames(res.dat.sup.residif)[7] <- "flexible" colnames(res.dat.sup.residif)[8] <- "mostflexible" res.dat.sup.residif[res.dat.sup.residif$nb.dif==0,"true.gamma"] <- NA res.dat.sup.residif[is.na(res.dat.sup.residif)] <- " " res.dat.sup.residif.w <- reshape(res.dat.sup.residif, direction = "wide", idvar = c("J","true.beta","nb.dif",'true.gamma'),timevar = "N" ) res.dat.sup.residif.nodif <- res.dat.sup.residif.w[res.dat.sup.residif.w$nb.dif==0,] res.dat.sup.residif.dif <- res.dat.sup.residif.w[res.dat.sup.residif.w$nb.dif!=0,] #### Sortie table nodif write.csv(cbind(res.dat.sup.rosali.nodif[,c(1:3,8,12,16)],res.dat.sup.residif.nodif[,c(8,12,16)]),"/home/corentin/Documents/These/Valorisation/Articles/Simulations 1/Tables/nodif_detect.csv") ########################## # DETECTION PERF DIF ##########################