Corrected pcmdif.do error

.gitignore

@@ -1,3 +1,5 @@
 *.csv
 *.xls
 .Rproj.user
+*.RData
+*.Rhistory

RProject/.Rhistory

@@ -1,262 +1,149 @@
-res.null0 <- res.dat[res.dat$eff.size==0 & res.dat$dif.size==0,]
+muB=10
-points(y=res.null0$h0.rejected.p,x=rep(1,nrow(res.null0)),col='gray',pch=3)
+kappa=1
-res.null3 <- res.dat[res.dat$eff.size==0 & res.dat$dif.size==0.3,]
+sd=10
-points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+alpha=0.05
-res.null5 <- res.dat[res.dat$eff.size==0 & res.dat$dif.size==0.5,]
+beta=0.20
-points(y=res.null5$h0.rejected.p,x=rep(3,nrow(res.null5)),col='#053305',pch=3)
+(nB=(1+1/kappa)*(sd*(qnorm(1-alpha/2)+qnorm(1-beta))/(muA-muB))^2)
-## Proportion of rejected h0 per dif value in h0 scenarios (A) by DIF size (1 item)
+ceiling(nB) # 63
-par(mfrow=c(2,2))
+z=(muA-muB)/(sd*sqrt((1+1/kappa)/nB))
-# 0 item
+(Power=pnorm(z-qnorm(1-alpha/2))+pnorm(-z-qnorm(1-alpha/2)))
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0,]
+muA=5
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
+muB=10
-ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+kappa=1
-points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+sd=10
-# 1 item
+alpha=0.05
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1,]
+beta=0.20
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
+(nB=(1+1/kappa)*(sd*(qnorm(1-alpha/2)+qnorm(1-beta))/(muA-muB))^2)
-ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
+ceiling(nB) # 63
-res.null3 <- res.null[res.null$dif.size==0.3,]
+z=(muA-muB)/(sd*sqrt((1+1/kappa)/nB))
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
+(Power=pnorm(z-qnorm(1-alpha/2))+pnorm(-z-qnorm(1-alpha/2)))
-res.null5 <- res.null[res.null$dif.size==0.5,]
+-1*(-0.49-0.47)
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
+-1*(-1.69-0.95)+(-0.49-0.47)
-# 2 items
+-1*(-0.49-0.47)-0.49
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2,]
+-1*(-0.49-0.47)-1.69
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
+-1*(-0.49-0.47)-(1.69-0.49)
-ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
+3.35-1.69
-res.null3 <- res.null[res.null$dif.size==0.3,]
+library(TAM)
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
+library(doMC)
-res.null5 <- res.null[res.null$dif.size==0.5,]
+library(parallel)
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
+library(pbmcapply)
-# 3 items
+library(funprog)
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3,]
+library(dplyr)
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
+lastChar <- function(str){
-ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
+substr(str, nchar(str)-2, nchar(str))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-par(mfrow=c(1,1))
-par(mfrow=c(2,2))
-# 0 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==4,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
-points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
-# 1 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1 & res.dat$J==4,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-# 2 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==4,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-par(mfrow=c(1,1))
-par(mfrow=c(2,2))
-# 0 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==7,]
-nrow(res.null)
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
-points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
-# 2 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==7,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-# 3 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$J==7,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-par(mfrow=c(1,1))
-par(mfrow=c(2,2))
-# 0 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==7,]
-nrow(res.null)
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
-points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
-# 2 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==7,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-# 3 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$J==7,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-par(mfrow=c(1,1))
-par(mfrow=c(2,2))
-# 0 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==4,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
-points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
-# 1 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1 & res.dat$J==4,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-# 2 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==4,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-par(mfrow=c(1,1))
-par(mfrow=c(2,2))
-# 0 item
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==7,]
-nrow(res.null)
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
-points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
-# 2 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==7,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-# 3 items
-res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$J==7,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2,3),xlab='DIF size',
-ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.null[res.null$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.null[res.null$dif.size==0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(2,nrow(res.null5)),col='#053305',pch=3)
-par(mfrow=c(1,1))
-# Overall
-boxplot(beta.same.sign.truebeta.p~dif.size,data=res.dat,col=c(2,3),xlab='DIF size',
-ylab='Proportion of estimates with same sign as true value in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.dat[res.dat$dif.size==-0.5,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0.5,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(5,nrow(res.null3)),col='#590b0c',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==-0.3,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(2,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0.3,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(4,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(3,nrow(res.null3)),col='gray',pch=3)
-#### Create data.frame
-results <- c(sapply(1:4,function(x) paste0(x,c('A','B','C','D','E'))),sapply(5:9,function(x) paste0(x,c('A','B','C','D','E','F','G'))))
-results2 <- c(sapply(10:20,function(x) paste0(x,c('A','B','C','D','E','F','G'))))
-results <- c(sapply(c(100,200,300),function(x) paste0(results,'_',x)))
-results2 <- c(sapply(c(100,200,300),function(x) paste0(results2,'_',x)))
-results <- sort(results)
-results2 <- sort(results2)
-results <- c(results,results2)
-#### Compiler function
-compile_simulation <- function(scenario) {
-name <- as.numeric(gsub("[^0-9.-]", "", substr(scenario,start=0,stop=2)))
-if (substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario))=="100" & name<=4) {
-s <- read.csv(paste0('/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N100/scenario_',scenario,'.csv'))
 }
-if (substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario))=="100" & name>4) {
+pcm_analysis <- function(df=NULL,treatment='TT',irtmodel='PCM2',method='MML') {
-s <- read.csv(paste0('/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N100/scenario_',scenario,'_nodif.csv'))
+nbitems <- sum(sapply(1:20,function(x) paste0('item',x)) %in% colnames(df))
+resp <- df[,sapply(seq(1,nbitems),function(x) paste0('item',x))]
+if (method=='MML') {
+tam1 <-  tam.mml(resp=resp,Y=df[,treatment],irtmodel = irtmodel,est.variance = T,verbose=F)
 }
-if (substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario))=="200" & name<=4) {
+if (method=='JML') {
-s <- read.csv(paste0('/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N200/scenario_',scenario,'.csv'))
+tam1 <- tam.jml(resp=resp,group=1+df[,treatment])
 }
-if (substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario))=="200" & name>4) {
+if (method!='MML' & method!='JML') {
-s <- read.csv(paste0('/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N200/scenario_',scenario,'_nodif.csv'))
+stop('Invalid method. Please choose among MML or JML')
 }
-if (substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario))=="300" & name<=4) {
+return(tam1)
-s <- read.csv(paste0('/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N300/scenario_',scenario,'.csv'))
 }
-if (substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario))=="300" & name>4) {
+replicate_pcm_analysis_m4 <- function(df=NULL,treatment='TT',irtmodel='PCM2',method='MML',sequence='replication',eff.size=0,difsize=NA,nbdif=0) {
-s <- read.csv(paste0('/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N300/scenario_',scenario,'_nodif.csv'))
+nbitems <- sum(sapply(1:20,function(x) paste0('item',x)) %in% colnames(df))
+resp <- df[,sapply(seq(1,nbitems),function(x) paste0('item',x))]
+truebeta <- eff.size
+if (method=='MML') {
+n <- max(df[,sequence])
+print(n)
+tam1 <- lapply(seq(1,n),
+function(x) pcm_analysis(df=df[df[,sequence]==x,],treatment=treatment,irtmodel=irtmodel)
+)
 }
-if (unique(s$J)==4) {
+listitems <- c(sapply(c('_1','_2','_3'),function(x) paste0(sapply(seq(1,nbitems),function(x) paste0('item',x)),x)))
-if (unique(s$M)==2) {
+returndat <- data.frame(matrix(nrow=max(df[,sequence]),ncol=length(listitems)))
-a <- data.frame(m.item1=mean(s$item1),m.item2=mean(s$item2),m.item3=mean(s$item3),m.item4=mean(s$item4))
+colnames(returndat) <- listitems
+for (s in seq(1,max(df[,sequence]))) {
+for (k in seq(1,nbitems)) {
+returndat[s,paste0('item',k,'_1')] <- tam1[[s]]$item[k,'AXsi_.Cat1']
+returndat[s,paste0('item',k,'_2')] <- tam1[[s]]$item[k,'AXsi_.Cat2']-tam1[[s]]$item[k,'AXsi_.Cat1']
+returndat[s,paste0('item',k,'_3')] <- tam1[[s]]$item[k,'AXsi_.Cat3']-tam1[[s]]$item[k,'AXsi_.Cat2']
+}
+}
+returndat <- returndat[,sort_by(listitems, lastChar)]
+returndat$beta <- sapply(seq(1,max(df[,sequence])),function(k) tam1[[k]]$beta[2])
+returndat$se.beta <- 1.413612*sapply(seq(1,max(df[,sequence])),function(k) tam.se(tam1[[k]])$beta$se.Dim1[2] )
+returndat$low.ci.beta <- returndat$beta-1.96*returndat$se.beta
+returndat$high.ci.beta <- returndat$beta+1.96*returndat$se.beta
+returndat$true.value.in.ci <- 1*(truebeta>returndat$low.ci.beta & truebeta<returndat$high.ci.beta)
+returndat$h0.rejected <- 1*(0<returndat$low.ci.beta | 0>returndat$high.ci.beta)
+if (truebeta==0) {
+returndat$beta.same.sign.truebeta <- NA
 } else {
-a <- data.frame(m.item1_1=mean(s$item1_1),m.item1_2=mean(s$item1_2),m.item1_3=mean(s$item1_3),
+returndat$beta.same.sign.truebeta <- 1*(sign(truebeta)==sign(returndat$beta))
-m.item2_1=mean(s$item2_1),m.item2_2=mean(s$item2_2),m.item2_3=mean(s$item2_3),
+}
-m.item3_1=mean(s$item3_1),m.item3_2=mean(s$item3_2),m.item3_3=mean(s$item3_3),
+returndat2 <- data.frame(J=rep(nbitems,max(df[,sequence])),
-m.item4_1=mean(s$item4_1),m.item4_2=mean(s$item4_2),m.item4_3=mean(s$item4_3)
+M=1+max(df$item1),
+N=nrow(df[df$replication==1,])/2,
+eff.size=truebeta,
+dif.size= difsize,
+nb.dif= nbdif
 )
+returndat <- cbind(returndat2,returndat)
+return(returndat)
 }
-} else {
+replicate_pcm_analysis_m2 <- function(df=NULL,treatment='TT',irtmodel='PCM2',method='MML',sequence='replication',eff.size=0,difsize=NA,nbdif=0) {
-if (unique(s$M)==2) {
+truebeta <- eff.size
-a <- data.frame(m.item1=mean(s$item1),m.item2=mean(s$item2),m.item3=mean(s$item3),m.item4=mean(s$item4),
+nbitems <- sum(sapply(1:20,function(x) paste0('item',x)) %in% colnames(df))
-m.item5=mean(s$item5),m.item6=mean(s$item6),m.item7=mean(s$item7))
+resp <- df[,sapply(seq(1,nbitems),function(x) paste0('item',x))]
-} else {
+if (method=='MML') {
-a <- data.frame(m.item1_1=mean(s$item1_1),m.item1_2=mean(s$item1_2),m.item1_3=mean(s$item1_3),
+n <- max(df[,sequence])
-m.item2_1=mean(s$item2_1),m.item2_2=mean(s$item2_2),m.item2_3=mean(s$item2_3),
+print(n)
-m.item3_1=mean(s$item3_1),m.item3_2=mean(s$item3_2),m.item3_3=mean(s$item3_3),
+tam1 <- lapply(seq(1,n),
-m.item4_1=mean(s$item4_1),m.item4_2=mean(s$item4_2),m.item4_3=mean(s$item4_3),
+function(x) pcm_analysis(df=df[df[,sequence]==x,],treatment=treatment,irtmodel=irtmodel)
-m.item5_1=mean(s$item5_1),m.item5_2=mean(s$item5_2),m.item5_3=mean(s$item5_3),
-m.item6_1=mean(s$item6_1),m.item6_2=mean(s$item6_2),m.item6_3=mean(s$item6_3),
-m.item7_1=mean(s$item7_1),m.item7_2=mean(s$item7_2),m.item7_3=mean(s$item7_3)
 )
 }
+listitems <- sapply(seq(1,nbitems),function(x) paste0('item',x))
+returndat <- data.frame(matrix(nrow=max(df[,sequence]),ncol=length(listitems)))
+colnames(returndat) <- listitems
+for (s in seq(1,max(df[,sequence]))) {
+for (k in seq(1,nbitems)) {
+returndat[s,paste0('item',k)] <- tam1[[s]]$xsi$xsi[k]
 }
-zz <- substr(scenario,start=0,stop=nchar(scenario)-4)
+}
-b <- data.frame(scenario=zz,
+returndat$beta <- sapply(seq(1,max(df[,sequence])),function(k) tam1[[k]]$beta[2])
-scenario.type=substr(zz,start=nchar(zz),stop=nchar(zz)),
+returndat$se.beta <- 1.413612*sapply(seq(1,max(df[,sequence])),function(k) tam.se(tam1[[k]])$beta$se.Dim1[2] )
-N=substr(scenario,start=nchar(scenario)-2,stop=nchar(scenario)),
+returndat$low.ci.beta <- returndat$beta-1.96*returndat$se.beta
-J=unique(s$J),
+returndat$high.ci.beta <- returndat$beta+1.96*returndat$se.beta
-M=unique(s$M),
+returndat$true.value.in.ci <- 1*(truebeta>returndat$low.ci.beta & truebeta<returndat$high.ci.beta)
-eff.size=unique(s$eff.size),
+returndat$h0.rejected <- 1*(0<returndat$low.ci.beta | 0>returndat$high.ci.beta)
-nb.dif=unique(s$nb.dif),
+if (truebeta==0) {
-dif.size=unique(s$dif.size)
+returndat$beta.same.sign.truebeta <- NA
+} else {
+returndat$beta.same.sign.truebeta <- 1*(sign(truebeta)==sign(returndat$beta))
+}
+returndat2 <- data.frame(J=rep(nbitems,max(df[,sequence])),
+M=1+max(df$item1),
+N=nrow(df[df$replication==1,])/2,
+eff.size=truebeta,
+dif.size= difsize,
+nb.dif= nbdif
 )
-z <- data.frame(m.beta=mean(s$beta),
+returndat <- cbind(returndat2,returndat)
-se.empirical.beta=sd(s$beta),
+return(returndat)
-se.analytical.beta=mean(s$se.beta),
-m.low.ci.beta=mean(s$low.ci.beta),
-m.high.ci.beta=mean(s$high.ci.beta),
-true.value.in.ci.p=mean(s$true.value.in.ci),
-h0.rejected.p=mean(s$h0.rejected),
-beta.same.sign.truebeta.p=mean(s$beta.same.sign.truebeta,na.rm=T),
-beta.same.sign.truebeta.signif.p=mean(s[s$h0.rejected==1,]$beta.same.sign.truebeta,na.rm=T))
-d <- cbind(b,a,z)
-d$prop.
-return(d)
 }
-#### Compiled results
+replicate_pcm_analysis<- function(df=NULL,treatment='TT',irtmodel='PCM2',method='MML',sequence='replication',eff.size=0,difsize=NA,nbdif=0) {
-res.dat <- compile_simulation('1A_100')
+j <- max(df$item1)
-for (x in results[seq(2,length(results))]) {
+if(j==1) {
-y <- compile_simulation(x)
+return(replicate_pcm_analysis_m2(df=df,treatment=treatment,irtmodel=irtmodel,method=method,sequence=sequence,eff.size=eff.size,difsize=difsize,nbdif=nbdif))
-res.dat <- bind_rows(res.dat,y)
+} else {
+return(replicate_pcm_analysis_m4(df=df,treatment=treatment,irtmodel=irtmodel,method=method,sequence=sequence,eff.size=eff.size,difsize=difsize,nbdif=nbdif))
 }
-library(TAM)
+}
-library(doMC)
+registerDoMC(4)
-library(parallel)
+dat1 <- read.csv(file = '/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N100/scenario_20A_100.csv')
-library(pbmcapply)
+res_nodif <- pbmclapply(c('dat1'),function(x) replicate_pcm_analysis(get(x),nbdif = 1,difsize = 0.5))
-library(funprog)
+write.csv(res_nodif[[1]],'/home/corentin/Documents/These/Recherche/Simulation/Analysis/DIF/N100/scenario_20A_100_nodif.csv')
-library(dplyr)
+res_nodif[[1]]
-#### Create data.frame
+write.csv(res_nodif[[1]],'/home/corentin/Documents/These/Recherche/Simulationss/Analysis/NoDIF/N100/scenario_20A_100_nodif.csv')
+write.csv(res_nodif[[1]],'/home/corentin/Documents/These/Recherche/Simulations/Analysis/NoDIF/N100/scenario_20A_100_nodif.csv')
 results <- c(sapply(1:4,function(x) paste0(x,c('A','B','C','D','E'))),sapply(5:9,function(x) paste0(x,c('A','B','C','D','E','F','G'))))
 results2 <- c(sapply(10:20,function(x) paste0(x,c('A','B','C','D','E','F','G'))))
 results <- c(sapply(c(100,200,300),function(x) paste0(results,'_',x)))
@@ -346,63 +233,188 @@ res.dat[300:396,'nb.dif'] <- 3
 res.dat.simple <- res.dat[,c(1:8,13,16:18)]
 res.dat.simple$m.beta <- round(res.dat.simple$m.beta,3)
 res.dat.simple
-boxplot(beta.same.sign.truebeta.signif.p~dif.size,data=res.dat,col=c(2,3),xlab='DIF size',
-ylab='Proportion of estimates with same sign as true value in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.dat[res.dat$dif.size==-0.5,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0.5,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(5,nrow(res.null3)),col='#590b0c',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==-0.3,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(2,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0.3,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(4,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0,]
-points(y=res.null3$beta.same.sign.truebeta.p,x=rep(3,nrow(res.null3)),col='gray',pch=3)
-boxplot(beta.same.sign.truebeta.signif.p~dif.size,data=res.dat,col=c(2,3),xlab='DIF size',
-ylab='Proportion of estimates with same sign as true value in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.dat[res.dat$dif.size==-0.5,]
-points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0.5,]
-points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(5,nrow(res.null3)),col='#590b0c',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==-0.3,]
-points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(2,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0.3,]
-points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(4,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$dif.size==0,]
-points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(3,nrow(res.null3)),col='gray',pch=3)
 ## Proportion of rejected h0 per dif value in h0 scenarios (A) by DIF size
-res.null <- res.dat[res.dat$eff.size!=0,]
+res.null <- res.dat[res.dat$eff.size==0,]
 boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,1),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
-res.null0 <- res.dat[res.dat$eff.size!=0 & res.dat$dif.size==0,]
+res.null0 <- res.dat[res.dat$eff.size==0 & res.dat$dif.size==0,]
 points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='gray',pch=3)
-res.null3 <- res.dat[res.dat$eff.size!=0 & res.dat$dif.size==-0.3,]
+res.null3 <- res.dat[res.dat$eff.size==0 & res.dat$dif.size==-0.3,]
 points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.dat[res.dat$eff.size!=0 & res.dat$dif.size==-0.5,]
+res.null5 <- res.dat[res.dat$eff.size==0 & res.dat$dif.size==-0.5,]
 points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
-res.null <- res.dat[res.dat$eff.size>0,]
+par(mfrow=c(2,2))
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,1),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
+# 0 item
-res.null0 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0,]
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0,]
-points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='gray',pch=3)
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
-res.null3 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==-0.3,]
+ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+# 1 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
 points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==-0.5,]
+res.null5 <- res.null[res.null$dif.size==-0.5,]
 points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
-res.null <- res.dat[res.dat$eff.size>0,]
+# 2 items
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2,]
-res.null0 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
-points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='gray',pch=3)
+ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
-res.null3 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==-0.3,]
+res.null3 <- res.null[res.null$dif.size==-0.3,]
 points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==-0.5,]
+res.null5 <- res.null[res.null$dif.size==-0.5,]
 points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
-res.null <- res.dat[res.dat$eff.size>0,]
+# 3 items
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3,]
-res.null0 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
-points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='darkblue',pch=3)
+ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null3 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==-0.3,]
+res.null3 <- res.null[res.null$dif.size==-0.3,]
 points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==-0.5,]
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+par(mfrow=c(1,1))
+par(mfrow=c(2,2))
+# 0 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==4,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+# 1 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1 & res.dat$J==4,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 2 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==4,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+par(mfrow=c(1,1))
+par(mfrow=c(2,2))
+# 0 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$J==7,]
+nrow(res.null)
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+# 2 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$J==7,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 3 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$J==7,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+par(mfrow=c(1,1))
+par(mfrow=c(2,2))
+# 0 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$N==100,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+# 1 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1 & res.dat$N==100,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 2 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$N==100,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 3 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$N==100,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+par(mfrow=c(1,1))
+####### N=200
+par(mfrow=c(2,2))
+# 0 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$N==200,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+# 1 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1 & res.dat$N==200,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 2 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$N==200,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 3 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$N==200,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+par(mfrow=c(1,1))
+####### N=300
+par(mfrow=c(2,2))
+# 0 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==0 & res.dat$N==300,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='No DIF',ylim=c(0,1))
+points(y=res.null$h0.rejected.p,x=rep(1,nrow(res.null)),col='#590b0c',pch=3)
+# 1 item
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==1 & res.dat$N==300,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 1 item',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 2 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==2 & res.dat$N==300,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 2 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
+points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+# 3 items
+res.null <- res.dat[res.dat$eff.size==0 & res.dat$nb.dif==3 & res.dat$N==300,]
+boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2),xlab='DIF size',
+ylab='H0 rejection proportion in target scenario',main='DIF on 3 items',ylim=c(0,1))
+res.null3 <- res.null[res.null$dif.size==-0.3,]
+points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null5 <- res.null[res.null$dif.size==-0.5,]
 points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+par(mfrow=c(1,1))
 res.null <- res.dat[res.dat$eff.size>0,]
 boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
 res.null0 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0,]
@@ -415,6 +427,7 @@ res.null3 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0.3,]
 points(y=res.null3$h0.rejected.p,x=rep(4,nrow(res.null3)),col='#590b0c',pch=3)
 res.null5 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0.5,]
 points(y=res.null5$h0.rejected.p,x=rep(5,nrow(res.null5)),col='#053305',pch=3)
+####### N=100
 res.null <- res.dat[res.dat$eff.size>0 & res.dat$N==100,]
 boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
 res.null0 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0 & res.dat$N==100,]
@@ -427,6 +440,7 @@ res.null3 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0.3 & res.dat$N==100
 points(y=res.null3$h0.rejected.p,x=rep(4,nrow(res.null3)),col='#590b0c',pch=3)
 res.null5 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0.5 & res.dat$N==100,]
 points(y=res.null5$h0.rejected.p,x=rep(5,nrow(res.null5)),col='#053305',pch=3)
+####### N=300 // DIF à 0.5 - QUELS SONT LES SCENARIOS EN HAUT
 res.null <- res.dat[res.dat$eff.size>0 & res.dat$N==300,]
 boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
 res.null0 <- res.dat[res.dat$eff.size>0 & res.dat$dif.size==0 & res.dat$N==300,]
@@ -449,10 +463,6 @@ res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.3 & res.dat$N==10
 points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
 res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.5 & res.dat$N==100,]
 points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0.3 & res.dat$N==100,]
-points(y=res.null3$h0.rejected.p,x=rep(4,nrow(res.null3)),col='#590b0c',pch=3)
-res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0.5 & res.dat$N==100,]
-points(y=res.null5$h0.rejected.p,x=rep(5,nrow(res.null5)),col='#053305',pch=3)
 ####### N=300 // DIF à 0.5 - QUELS SONT LES SCENARIOS EN HAUT
 res.null <- res.dat[res.dat$eff.size<0 & res.dat$N==300,]
 boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
@@ -462,51 +472,41 @@ res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.3 & res.dat$N==30
 points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
 res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.5 & res.dat$N==300,]
 points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0.3 & res.dat$N==300,]
+boxplot(true.value.in.ci.p~dif.size,data=res.dat,col=c(2,3),xlab='DIF size',
-points(y=res.null3$h0.rejected.p,x=rep(4,nrow(res.null3)),col='#590b0c',pch=3)
+ylab='Proportion of true beta value in CI in target scenario',main='DIF on 3 items',ylim=c(0,1))
-res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0.5 & res.dat$N==300,]
+res.null3 <- res.dat[res.dat$dif.size==-0.5,]
-points(y=res.null5$h0.rejected.p,x=rep(5,nrow(res.null5)),col='#053305',pch=3)
+points(y=res.null3$true.value.in.ci.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-res.null <- res.dat[res.dat$eff.size<0 & res.dat$N==100,]
+res.null3 <- res.dat[res.dat$dif.size==0.5,]
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
+points(y=res.null3$true.value.in.ci.p,x=rep(5,nrow(res.null3)),col='#590b0c',pch=3)
-res.null0 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0 & res.dat$N==100,]
+res.null3 <- res.dat[res.dat$dif.size==-0.3,]
-points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='darkblue',pch=3)
+points(y=res.null3$true.value.in.ci.p,x=rep(2,nrow(res.null3)),col='#053305',pch=3)
-res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.3 & res.dat$N==100,]
+res.null3 <- res.dat[res.dat$dif.size==0.3,]
-points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+points(y=res.null3$true.value.in.ci.p,x=rep(4,nrow(res.null3)),col='#053305',pch=3)
-res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.5 & res.dat$N==100,]
+res.null3 <- res.dat[res.dat$dif.size==0,]
-points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+points(y=res.null3$true.value.in.ci.p,x=rep(3,nrow(res.null3)),col='gray',pch=3)
-res.null <- res.dat[res.dat$eff.size<0 & res.dat$N==300,]
+# Overall
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
+boxplot(beta.same.sign.truebeta.p~dif.size,data=res.dat,col=c(2,3),xlab='DIF size',
-res.null0 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0 & res.dat$N==300,]
+ylab='Proportion of estimates with same sign as true value in target scenario',main='DIF on 3 items',ylim=c(0,1))
-points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='darkblue',pch=3)
+res.null3 <- res.dat[res.dat$dif.size==-0.5,]
-res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.3 & res.dat$N==300,]
+points(y=res.null3$beta.same.sign.truebeta.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+res.null3 <- res.dat[res.dat$dif.size==0.5,]
-res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.5 & res.dat$N==300,]
+points(y=res.null3$beta.same.sign.truebeta.p,x=rep(5,nrow(res.null3)),col='#590b0c',pch=3)
-points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+res.null3 <- res.dat[res.dat$dif.size==-0.3,]
-res.null <- res.dat[res.dat$eff.size<0 & res.dat$N==100,]
+points(y=res.null3$beta.same.sign.truebeta.p,x=rep(2,nrow(res.null3)),col='#053305',pch=3)
-boxplot(h0.rejected.p~dif.size,data=res.null,col=c(3,2,4,2,3),xlab='DIF size',ylab='H0 rejection proportion in target scenario')
+res.null3 <- res.dat[res.dat$dif.size==0.3,]
-res.null0 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==0 & res.dat$N==100,]
+points(y=res.null3$beta.same.sign.truebeta.p,x=rep(4,nrow(res.null3)),col='#053305',pch=3)
-points(y=res.null0$h0.rejected.p,x=rep(3,nrow(res.null0)),col='darkblue',pch=3)
+res.null3 <- res.dat[res.dat$dif.size==0,]
-res.null3 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.3 & res.dat$N==100,]
+points(y=res.null3$beta.same.sign.truebeta.p,x=rep(3,nrow(res.null3)),col='gray',pch=3)
-points(y=res.null3$h0.rejected.p,x=rep(2,nrow(res.null3)),col='#590b0c',pch=3)
+# Overall // H0 rejected
-res.null5 <- res.dat[res.dat$eff.size<0 & res.dat$dif.size==-0.5 & res.dat$N==100,]
+boxplot(beta.same.sign.truebeta.signif.p~dif.size,data=res.dat,col=c(2,3),xlab='DIF size',
-points(y=res.null5$h0.rejected.p,x=rep(1,nrow(res.null5)),col='#053305',pch=3)
+ylab='Proportion of estimates with same sign as true value in target scenario',main='When H0 rejected',ylim=c(0,1))
-muA=5
+res.null3 <- res.dat[res.dat$dif.size==-0.5,]
-muB=10
+points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(1,nrow(res.null3)),col='#590b0c',pch=3)
-kappa=1
+res.null3 <- res.dat[res.dat$dif.size==0.5,]
-sd=10
+points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(5,nrow(res.null3)),col='#590b0c',pch=3)
-alpha=0.05
+res.null3 <- res.dat[res.dat$dif.size==-0.3,]
-beta=0.20
+points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(2,nrow(res.null3)),col='#053305',pch=3)
-(nB=(1+1/kappa)*(sd*(qnorm(1-alpha/2)+qnorm(1-beta))/(muA-muB))^2)
+res.null3 <- res.dat[res.dat$dif.size==0.3,]
-ceiling(nB) # 63
+points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(4,nrow(res.null3)),col='#053305',pch=3)
-z=(muA-muB)/(sd*sqrt((1+1/kappa)/nB))
+res.null3 <- res.dat[res.dat$dif.size==0,]
-(Power=pnorm(z-qnorm(1-alpha/2))+pnorm(-z-qnorm(1-alpha/2)))
+points(y=res.null3$beta.same.sign.truebeta.signif.p,x=rep(3,nrow(res.null3)),col='gray',pch=3)
-muA=5
-muB=10
-kappa=1
-sd=10
-alpha=0.05
-beta=0.20
-(nB=(1+1/kappa)*(sd*(qnorm(1-alpha/2)+qnorm(1-beta))/(muA-muB))^2)
-ceiling(nB) # 63
-z=(muA-muB)/(sd*sqrt((1+1/kappa)/nB))
-(Power=pnorm(z-qnorm(1-alpha/2))+pnorm(-z-qnorm(1-alpha/2)))

Scripts/Analysis/DIF/pcm_dif.do

@@ -296,6 +296,8 @@ putexcel A1=matrix(outmat), colnames
 
 ****** Scenarios with DIF on 2 items
 
+* Load pcm.ado
+adopath+"/home/corentin/Documents/These/Recherche/Simulations/Modules/"
 
 ** Scenario 9: J = 4 items / M = 2 modalities / DIF size 0.3 x2
   local N = "100 200 300"
@@ -427,7 +429,7 @@ forvalues k=1/1000 {
       local mod = "`mod'"+"(1.item`i'<-THETA@1)(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)"
     }
   }
-  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2)"
+  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2 3 4)"
   qui `constrnt'
   qui `constrnt2'
   qui `constrnt3'
@@ -611,7 +613,7 @@ forvalues k=1/1000 {
       local mod = "`mod'"+"(1.item`i'<-THETA@1)(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)"
     }
   }
-  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2)"
+  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2 3 4)"
   qui `constrnt'
   qui `constrnt2'
   qui `constrnt3'
@@ -666,12 +668,14 @@ putexcel A1=matrix(outmat), colnames
 
 
 *==========================
-* Scenarios with : J=4
+* Scenarios with : J=7
 *==========================
 
 
 ****** Scenarios with DIF on 2 items
 
+* Load pcm.ado
+adopath+"/home/corentin/Documents/These/Recherche/Simulations/Modules/"
 
 ** Scenario 13: J = 7 items / M = 2 modalities / DIF size 0.3 x2
   local N = "100 200 300"
@@ -803,7 +807,7 @@ forvalues k=1/1000 {
       local mod = "`mod'"+"(1.item`i'<-THETA@1)(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)"
     }
   }
-  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2)"
+  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2 3 4)"
   qui `constrnt'
   qui `constrnt2'
   qui `constrnt3'
@@ -987,7 +991,7 @@ forvalues k=1/1000 {
       local mod = "`mod'"+"(1.item`i'<-THETA@1)(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)"
     }
   }
-  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2)"
+  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2 3 4)"
   qui `constrnt'
   qui `constrnt2'
   qui `constrnt3'
@@ -1041,6 +1045,8 @@ putexcel A1=matrix(outmat), colnames
 
 ***** Scenarios with DIF on 3 items
 
+* Load pcm.ado
+adopath+"/home/corentin/Documents/These/Recherche/Simulations/Modules/"
 
 ** Scenario 17: J = 7 items / M = 2 modalities / DIF size 0.3 x3
   local N = "100 200 300"
@@ -1207,7 +1213,7 @@ forvalues k=1/1000 {
       local mod = "`mod'"+"(1.item`i'<-THETA@1)(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)"
     }
   }
-  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2)"
+  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2 3 4 5 6)"
   qui `constrnt'
   qui `constrnt2'
   qui `constrnt3'
@@ -1440,7 +1446,7 @@ forvalues k=1/1000 {
       local mod = "`mod'"+"(1.item`i'<-THETA@1)(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)"
     }
   }
-  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2)"
+  local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(1 2 3 4 5 6)"
   qui `constrnt'
   qui `constrnt2'
   qui `constrnt3'