Send output to:
Browser Blue - Charts White
Browser Black/White
CSV
Data X:
19 17 18 12 17 15 17 14 19 16 19 14 16 15 16 15 15 9 18 18 18 15 19 14 17 15 17 14 13 14 17 11 17 13 13 17 19 15 15 15 14 15 16 8 15 14 12 15 8 17 16 9 12 19 13 12 16 16 16 13 16 15 16 16 18 15 14 16 14 19 15 16 19 14 17 11 16 12 17 13 15 11 11 20 8 11 20 14 20 13 11 15 15 13 14 14 15 17 20 10 14 9 16 10 16 9 15 17 16 10 16 9 15 14 17 14 18 10 13 18 12 14 6 16 9 11 15 17 15 14 17 11 19 11 20 8 9 9 15 18 16 15 16 16 17 15 19 17 12 16 11 17 17 10 9 14 19 16 17 17 13 18 11 17 7 14 17 9 16 16 12 11 10 12 8 15 18 15 15 17 16 11 20 13 16 13 17 9 13 16 9 11 12 9 15 16 10 13 18 15 17 15 14 15 17 15 13 17 12 17 9 9 15 9 14 15 14 6 15 16 10 8 9 10 12 8 13 10 14 15 18 16 15 12 14 10 10 12 17 13 12 12 16 16 10 16 16 14 11 15 14 15 16 15 16 13 16 14 7 16 14 12 14 10 15 9 20 16 14 19 14 13 15 15 13 14 12 11 8 14 10 16 14 13 12 15 16 14 19 15 11 14 12 12 12 13 11 9 8 9 12 15 16 14 18 11 16 14 15 14 20 13 15 14 14 14 14 11 19 14 17 13 18 14 13 12 14 15 15 16 20 12 9 13 16 14 14 14 15 15 19 11 14 11 14 12 13 19 18 16 15 13 14 10 19 14 16 14 11 14 13 16 14 7 11 17 11 11 16 11 15 10 15 13 17 12 12 15 15 11 17 12 5 20 7 9 10 15 15 15 18 11 20 12 20 16 16 16 15 13 18 12 9 19 12 6 11 12 19 17 16 15 15 15 14 17 14 16 12 18 16 11 16 10 16 15 13 14 13 10 6 15 15 14 14 16 15 12 13 19 11 11 17 15 14 15
Names of X columns:
Man Vrouw
Column number of first sample
Column number of second sample
Confidence
Alternative
two.sided
two.sided
less
greater
Are observations paired?
unpaired
unpaired
paired
Null Hypothesis
Chart options
Title:
R Code
par1 <- as.numeric(par1) #column number of first sample par2 <- as.numeric(par2) #column number of second sample par3 <- as.numeric(par3) #confidence (= 1 - alpha) if (par5 == 'unpaired') paired <- FALSE else paired <- TRUE par6 <- as.numeric(par6) #H0 z <- t(y) if (par1 == par2) stop('Please, select two different column numbers') if (par1 < 1) stop('Please, select a column number greater than zero for the first sample') if (par2 < 1) stop('Please, select a column number greater than zero for the second sample') if (par1 > length(z[1,])) stop('The column number for the first sample should be smaller') if (par2 > length(z[1,])) stop('The column number for the second sample should be smaller') if (par3 <= 0) stop('The confidence level should be larger than zero') if (par3 >= 1) stop('The confidence level should be smaller than zero') (r.t <- t.test(z[,par1],z[,par2],var.equal=TRUE,alternative=par4,paired=paired,mu=par6,conf.level=par3)) (v.t <- var.test(z[,par1],z[,par2],conf.level=par3)) (r.w <- t.test(z[,par1],z[,par2],var.equal=FALSE,alternative=par4,paired=paired,mu=par6,conf.level=par3)) (w.t <- wilcox.test(z[,par1],z[,par2],alternative=par4,paired=paired,mu=par6,conf.level=par3)) (ks.t <- ks.test(z[,par1],z[,par2],alternative=par4)) m1 <- mean(z[,par1],na.rm=T) m2 <- mean(z[,par2],na.rm=T) mdiff <- m1 - m2 newsam1 <- z[!is.na(z[,par1]),par1] newsam2 <- z[,par2]+mdiff newsam2 <- newsam2[!is.na(newsam2)] (ks1.t <- ks.test(newsam1,newsam2,alternative=par4)) mydf <- data.frame(cbind(z[,par1],z[,par2])) colnames(mydf) <- c('Variable 1','Variable 2') bitmap(file='test1.png') boxplot(mydf, notch=TRUE, ylab='value',main=main) dev.off() bitmap(file='test2.png') qqnorm(z[,par1],main='Normal QQplot - Variable 1') qqline(z[,par1]) dev.off() bitmap(file='test3.png') qqnorm(z[,par2],main='Normal QQplot - Variable 2') qqline(z[,par2]) dev.off() load(file='createtable') a<-table.start() a<-table.row.start(a) a<-table.element(a,paste('Two Sample t-test (',par5,')',sep=''),2,TRUE) a<-table.row.end(a) if(!paired){ a<-table.row.start(a) a<-table.element(a,'Mean of Sample 1',header=TRUE) a<-table.element(a,r.t$estimate[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Mean of Sample 2',header=TRUE) a<-table.element(a,r.t$estimate[[2]]) a<-table.row.end(a) } else { a<-table.row.start(a) a<-table.element(a,'Difference: Mean1 - Mean2',header=TRUE) a<-table.element(a,r.t$estimate) a<-table.row.end(a) } a<-table.row.start(a) a<-table.element(a,'t-stat',header=TRUE) a<-table.element(a,r.t$statistic[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'df',header=TRUE) a<-table.element(a,r.t$parameter[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'p-value',header=TRUE) a<-table.element(a,r.t$p.value) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'H0 value',header=TRUE) a<-table.element(a,r.t$null.value[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Alternative',header=TRUE) a<-table.element(a,r.t$alternative) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'CI Level',header=TRUE) a<-table.element(a,attr(r.t$conf.int,'conf.level')) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'CI',header=TRUE) a<-table.element(a,paste('[',r.t$conf.int[1],',',r.t$conf.int[2],']',sep='')) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'F-test to compare two variances',2,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'F-stat',header=TRUE) a<-table.element(a,v.t$statistic[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'df',header=TRUE) a<-table.element(a,v.t$parameter[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'p-value',header=TRUE) a<-table.element(a,v.t$p.value) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'H0 value',header=TRUE) a<-table.element(a,v.t$null.value[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Alternative',header=TRUE) a<-table.element(a,v.t$alternative) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'CI Level',header=TRUE) a<-table.element(a,attr(v.t$conf.int,'conf.level')) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'CI',header=TRUE) a<-table.element(a,paste('[',v.t$conf.int[1],',',v.t$conf.int[2],']',sep='')) a<-table.row.end(a) a<-table.end(a) table.save(a,file='mytable.tab') a<-table.start() a<-table.row.start(a) a<-table.element(a,paste('Welch Two Sample t-test (',par5,')',sep=''),2,TRUE) a<-table.row.end(a) if(!paired){ a<-table.row.start(a) a<-table.element(a,'Mean of Sample 1',header=TRUE) a<-table.element(a,r.w$estimate[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Mean of Sample 2',header=TRUE) a<-table.element(a,r.w$estimate[[2]]) a<-table.row.end(a) } else { a<-table.row.start(a) a<-table.element(a,'Difference: Mean1 - Mean2',header=TRUE) a<-table.element(a,r.w$estimate) a<-table.row.end(a) } a<-table.row.start(a) a<-table.element(a,'t-stat',header=TRUE) a<-table.element(a,r.w$statistic[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'df',header=TRUE) a<-table.element(a,r.w$parameter[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'p-value',header=TRUE) a<-table.element(a,r.w$p.value) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'H0 value',header=TRUE) a<-table.element(a,r.w$null.value[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Alternative',header=TRUE) a<-table.element(a,r.w$alternative) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'CI Level',header=TRUE) a<-table.element(a,attr(r.w$conf.int,'conf.level')) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'CI',header=TRUE) a<-table.element(a,paste('[',r.w$conf.int[1],',',r.w$conf.int[2],']',sep='')) a<-table.row.end(a) a<-table.end(a) table.save(a,file='mytable1.tab') a<-table.start() a<-table.row.start(a) myWlabel <- 'Wilcoxon Signed-Rank Test' if (par5=='unpaired') myWlabel = 'Wilcoxon Rank-Sum Test (Mann–Whitney U test)' a<-table.element(a,paste(myWlabel,' with continuity correction (',par5,')',sep=''),2,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'W',header=TRUE) a<-table.element(a,w.t$statistic[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'p-value',header=TRUE) a<-table.element(a,w.t$p.value) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'H0 value',header=TRUE) a<-table.element(a,w.t$null.value[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Alternative',header=TRUE) a<-table.element(a,w.t$alternative) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Kolmogorov-Smirnov Test to compare <i>Distributions</i> of two Samples',2,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'KS Statistic',header=TRUE) a<-table.element(a,ks.t$statistic[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'p-value',header=TRUE) a<-table.element(a,ks.t$p.value) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Kolmogorov-Smirnov Test to compare <i>Distributional Shape</i> of two Samples',2,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'KS Statistic',header=TRUE) a<-table.element(a,ks1.t$statistic[[1]]) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'p-value',header=TRUE) a<-table.element(a,ks1.t$p.value) a<-table.row.end(a) a<-table.end(a) table.save(a,file='mytable2.tab')
Compute
Summary of computational transaction
Raw Input
view raw input (R code)
Raw Output
view raw output of R engine
Computing time
0 seconds
R Server
Big Analytics Cloud Computing Center
Click here to blog (archive) this computation