ANCOVA for Pre-Posttest Design

Pre-Posttest
About

Comparing two independent conditions

Note: Input values must be separated by tabs. Copy and paste from Excel/Numbers.

Your data needs to have the header (variable names) in the first row.
The covariate must be placed in the second column of the dataframe.

Basic statistics

Histograms

Interaction plot

Scatterplot

The data above the diagonal of the plot show increase in the posttest. Wright (2003, p. 130) suggests that this is the only procedure that is always correct in the pre-posttest design.

t-test with the pretest data

Checking ANCOVA assumptions

If "Pr(>F)" is over .05 (p > .05) for both tests (pre:group for homogeneity of regression slopes), assumptions are met.

ANCOVA result

If "Pr(>F) of group[name]" is below .05 (p < .05), the posttest scores for the two groups are statistically different after controling for the pretest score difference.

Adjusted means of posttest scores

Statistical test of adjusted means

Effect size of adjusted means

R session info

Note

This web application is developed with Shiny.

List of Packages Used
library(shiny)
library(shinyAce)
library(psych)
library(lattice)
library(ggplot2)
library(compute.es)
library(car)
library(effects)
library(multcomp)

Code

Source code for this application is based on "The handbook of Research in Foreign Language Learning and Teaching" (Takeuchi & Mizumoto, 2014).

Citation in Publications

Mizumoto, A. (2015). Langtest (Version 1.0) [Web application]. Retrieved from https://langtest.jp

Article

Mizumoto, A., & Plonsky, L. (2015). R as a lingua franca: Advantages of using R for quantitative research in applied linguistics. Applied Linguistics, 37(2), 284–291. https://academic.oup.com/applij/article-abstract/37/2/284/1742739

Author

Atsushi MIZUMOTO, Ph.D.
Professor of Applied Linguistics
Faculty of Foreign Language Studies /
Graduate School of Foreign Language Education and Research,
Kansai University, Osaka, Japan

License

library(shiny) library(shinyAce) library(psych) library(lattice) library(ggplot2) library(compute.es) library(car) library (effects) library (multcomp) shinyServer(function(input, output) { #----------------------------------------------------------------- # ANCOVA for Pre-Posttest Design #----------------------------------------------------------------- # Basic statistics bs <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) describeBy(dat[,-1], dat[,1]) }) output$bs.out <- renderPrint({ bs() }) # Histograms hist <- function(){ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") histogram(~ post + pre | group, data=dat) } output$hist <- renderPlot({ print(hist()) }) # Interaction plot int.plot <- function(){ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") labelnames <- colnames(dat) groupnames <- unique(dat[,1]) dat <- reshape(dat, varying=2:length(dat), v.names="score", direction = "long") res <- dat[,3] names(dat) <- c("fac1", "fac2", "res") df <- with(dat, aggregate(res, list(fac1=fac1, fac2=fac2), mean)) ci <- with(dat , aggregate(res, list(fac1=fac1, fac2=fac2), function(x) qt(0.975,length(x)-1)*sd(x)/sqrt(length(x)))) df$ci <- ci[,3] df$fac1 <- factor(df[,1]) df$fac2 <- factor(df[,2]) opar <- theme_update(panel.grid.major = element_blank(), panel.grid.minor = element_blank(), panel.background = element_rect(colour = "black")) xgap <- position_dodge(0.15) gp <- ggplot(df, aes(x=fac2, y=x, colour=fac1, group=fac1)) gp + geom_line(aes(linetype=fac1), size=.6, position=xgap) + geom_point(aes(shape=fac1), size=6, position=xgap) + geom_errorbar(aes(ymax=x+ci, ymin=x-ci), width=.1, position=xgap) + ylim(min(dat$res), max(dat$res)) + labs(title = "Error bars: 95% CI", y = "Score" ) + theme(plot.title=element_text(size=14), axis.title.x = element_text(size=14), axis.title.y = element_text(size=14), axis.text.x = element_text(size=14), axis.text.y = element_text(size=14), legend.position = c(1, 1), legend.justification = c(1, 1), legend.key = element_rect(fill = "white",colour = "white"), legend.title = element_text(size=14), legend.text = element_text(size = 14)) + scale_colour_discrete(name ="Condition", labels = c("1" = as.character(groupnames[1]), "2" = as.character(groupnames[2]))) + scale_linetype_discrete(name ="Condition", labels = c("1" = as.character(groupnames[1]), "2" = as.character(groupnames[2]))) + scale_shape_discrete(name ="Condition", labels = c("1" = as.character(groupnames[1]), "2" = as.character(groupnames[2]))) + scale_x_discrete("", labels = c("1" = labelnames[2], "2"=labelnames[3])) } output$int.plot <- renderPlot({ print(int.plot()) }) # Scatterplot scatterplot <- function(){ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) allscore <- c(dat[,2], dat[,3]) minval <- round(min(allscore)*0.95,0) maxval <- round(max(allscore)*1.05,0) groupnames <- unique(dat[,1]) labelnames <- colnames(dat) dat2 <- split(dat, dat[,1]) group1 <- dat2[[1]] group2 <- dat2[[2]] par(mar = c(5, 5, 2, 2)) par(xaxs = "i", yaxs = "i") plot(group1[,2], group1[,3], pch = 16, cex = 1.5, col = 2, xlim = c(minval,maxval), ylim = c(minval,maxval), las = 1, xlab = labelnames[2], ylab = labelnames[3], cex.lab = 1.2) points(group2[,2], group2[,3], pch = 17, cex = 1.5, col = 1) lines(par()$usr[1:2], par()$usr[3:4], lty = 1, lwd = 0.5) legend("bottomright", legend = c(as.character(unique(group1[,1])),as.character(unique(group2[,1]))), col = c(2, 1), pch = c(16, 17), cex = 1.2) } output$scatterplot <- renderPlot({ print(scatterplot()) }) # t-test with pretest scores t_test <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") t.test(pre ~ group, data=dat) }) output$t_test.out <- renderPrint({ t_test() }) t_effect <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat2 <- split(dat, dat[,1]) group1 <- dat2[[1]] group2 <- dat2[[2]] m1 <- mean(group1[,2]) sd1 <- sd(group1[,2]) n1 <- length(group1[,2]) m2 <- mean(group2[,2]) sd2 <- sd(group2[,2]) n2 <- length(group2[,2]) esd <- mes(m1, m2, sd1, sd2, n1, n2, verbose=FALSE) cat("d [95 %CI] = ", esd[4]$d, "[", esd[6]$l.d, ",", esd[7]$u.d, "]", "\n", "var(d) = ", esd[5]$var.d, "\n", "p-value(d) = ", esd[11]$pval.d, "\n" ) }) output$t_effect.out <- renderPrint({ t_effect() }) # Checking ANCOVA assumptions levene <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") leveneTest(dat$post, dat$group, center = median) }) output$levene.out <- renderPrint({ levene() }) slope <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") summary(aov(post ~ pre * group, data = dat)) }) output$slope.out <- renderPrint({ slope() }) # ANCOVA result ancova.res <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") summary(lm(post ~ pre + group, data=dat)) }) output$ancova.res.out <- renderPrint({ ancova.res() }) # Adjusted means of posttest scores adj.m <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") model <- aov(post ~ pre + group, data = dat) adjustedMeans <- effect("group", model, se=TRUE) summary(adjustedMeans) }) output$adj.m.out <- renderPrint({ adj.m() }) adjustedplot <- function(){ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") model <- aov(post ~ pre + group, data = dat) adjustedMeans <- effect("group", model, se=TRUE) plot(adjustedMeans) } output$adjustedplot <- renderPlot({ print(adjustedplot()) }) # Statistical test of adjusted means adj.m.test <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") model <- aov(post ~ pre + group, data = dat) postHocs <- glht(model, linfct = mcp(group = "Tukey")) print(summary(postHocs)) cat("\n", "[95% CI]", "\n") print(head(confint(postHocs)$confint)) }) output$adj.m.test.out <- renderPrint({ adj.m.test() }) # Effect size of adjusted means adj.m.effect <- reactive({ dat <- read.csv(text=input$text1, sep="", na.strings=c("","NA",".")) dat[,1] <- as.factor(dat[,1]) colnames(dat) <- c("group","pre","post") model <- aov(post ~ pre + group, data = dat) adjustedMeans <- effect("group", model, se=TRUE) dat2 <- split(dat, dat[,1]) group1 <- dat2[[1]] group2 <- dat2[[2]] n1 <- length(group1[,2]) n2 <- length(group2[,2]) n <- c(n1, n2) sdadj <- adjustedMeans$se*sqrt(n) adjustedMeans esd <- mes(adjustedMeans$fit[1], adjustedMeans$fit[2], sdadj[1], sdadj[2], n1, n2, verbose=FALSE) cat("d [95 %CI] = ", esd[4]$d, "[", esd[6]$l.d, ",", esd[7]$u.d, "]", "\n", "var(d) = ", esd[5]$var.d, "\n", "p-value(d) = ", esd[11]$pval.d, "\n" ) }) output$adj.m.effect.out <- renderPrint({ adj.m.effect() }) # Info ancova.info <- reactive({ info1 <- paste("This analysis was conducted with ", strsplit(R.version$version.string, " \\(")[[1]][1], ".", sep = "") info2 <- paste("It was executed on ", date(), ".", sep = "") cat(sprintf(info1), "\n") cat(sprintf(info2), "\n") }) output$ancova.info.out <- renderPrint({ ancova.info() }) })

library(shiny) library(shinyAce) shinyUI(bootstrapPage( headerPanel("ANCOVA for Pre-Posttest Design"), ########## loading message ####################################### tags$head(tags$style(type="text/css", " #loadmessage { position: fixed; top: 0px; left: 0px; width: 100%; padding: 10px 0px 10px 0px; text-align: center; font-weight: bold; font-size: 100%; color: #000000; background-color: #CCFF66; z-index: 105; } ")), conditionalPanel(condition="$('html').hasClass('shiny-busy')", tags$div("Loading...",id="loadmessage")), ################################################################### mainPanel( tabsetPanel( # Rのバージョンが古いlangtest.jpのみ必要？ tabPanel("Pre-Posttest", h4("Comparing two independent conditions"), p('Note: Input values must be separated by tabs. Copy and paste from Excel/Numbers.'), p(HTML("<b><div style='background-color:#FADDF2;border:1px solid black;'>Your data needs to have the header (variable names) in the first row. <br> The covariate must be placed in the second column of the dataframe. </div></b>")), aceEditor("text1", value="Group\tPre\tPost\ntreatment\t31\t48\ntreatment\t39\t51\ntreatment\t56\t67\ntreatment\t47\t44\ntreatment\t29\t33\ntreatment\t37\t41\ntreatment\t46\t43\ntreatment\t44\t53\ntreatment\t52\t64\ntreatment\t41\t52\ntreatment\t40\t53\ntreatment\t38\t44\ntreatment\t31\t39\ntreatment\t25\t32\ntreatment\t51\t62\ntreatment\t31\t43\ntreatment\t56\t59\ntreatment\t18\t19\ntreatment\t35\t46\ntreatment\t30\t50\ntreatment\t55\t62\ntreatment\t35\t45\ntreatment\t48\t58\ntreatment\t40\t49\ntreatment\t46\t58\ntreatment\t50\t66\ntreatment\t39\t44\ntreatment\t52\t64\ntreatment\t35\t45\ntreatment\t36\t53\ncontrol\t32\t40\ncontrol\t35\t39\ncontrol\t35\t42\ncontrol\t38\t28\ncontrol\t13\t11\ncontrol\t34\t30\ncontrol\t35\t25\ncontrol\t31\t39\ncontrol\t35\t53\ncontrol\t25\t40\ncontrol\t28\t41\ncontrol\t49\t47\ncontrol\t40\t37\ncontrol\t41\t36\ncontrol\t36\t41\ncontrol\t29\t33\ncontrol\t20\t22\ncontrol\t38\t26\ncontrol\t31\t34\ncontrol\t41\t49\ncontrol\t32\t43\ncontrol\t36\t37\ncontrol\t11\t24\ncontrol\t33\t35\ncontrol\t17\t27\ncontrol\t48\t56\ncontrol\t31\t29\ncontrol\t51\t60\ncontrol\t51\t58\ncontrol\t55\t62", mode="r", theme="cobalt"), br(), h3("Basic statistics"), verbatimTextOutput("bs.out"), br(), h3("Histograms"), plotOutput("hist"), br(), h3("Interaction plot"), plotOutput("int.plot"), br(), h3("Scatterplot"), plotOutput("scatterplot"), p('The data above the diagonal of the plot show increase in the posttest.', a("Wright (2003, p. 130)", href="https://doi.org/10.1348/000709903762869950", target="_blank"), 'suggests that this is the only procedure that is always correct in the pre-posttest design.', br()), br(), h3("t-test with the pretest data"), verbatimTextOutput("t_test.out"), verbatimTextOutput("t_effect.out"), br(), h3("Checking ANCOVA assumptions"), verbatimTextOutput("levene.out"), verbatimTextOutput("slope.out"), p('If "Pr(>F)" is over .05 (p > .05) for both tests (pre:group for homogeneity of regression slopes), assumptions are met.',br()), br(), h3("ANCOVA result"), verbatimTextOutput("ancova.res.out"), p('If "Pr(>F) of group[name]" is below .05 (p < .05), the posttest scores for the two groups are statistically different after controling for the pretest score difference.',br()), br(), h3("Adjusted means of posttest scores"), verbatimTextOutput("adj.m.out"), plotOutput("adjustedplot"), br(), h3("Statistical test of adjusted means"), verbatimTextOutput("adj.m.test.out"), br(), h3("Effect size of adjusted means"), verbatimTextOutput("adj.m.effect.out"), br(), br(), strong('R session info'), verbatimTextOutput("ancova.info.out") ), # About tabPanel("About", strong('Note'), p('This web application is developed with', a("Shiny.", href="https://shiny.rstudio.com/", target="_blank"), ''), br(), strong('List of Packages Used'), br(), code('library(shiny)'),br(), code('library(shinyAce)'),br(), code('library(psych)'),br(), code('library(lattice)'),br(), code('library(ggplot2)'),br(), code('library(compute.es)'),br(), code('library(car)'),br(), code('library(effects)'),br(), code('library(multcomp)'),br(), br(), strong('Code'), p('Source code for this application is based on', a('"The handbook of Research in Foreign Language Learning and Teaching" (Takeuchi & Mizumoto, 2014).', href='http://mizumot.com/handbook/', target="_blank")), br(), strong('Citation in Publications'), p('Mizumoto, A. (2015). Langtest (Version 1.0) [Web application]. Retrieved from', a("https://langtest.jp", href="https://langtest.jp", target="_blank")), br(), strong('Article'), p('Mizumoto, A., & Plonsky, L. (2015). R as a lingua franca: Advantages of using R for quantitative research in applied linguistics.', em('Applied Linguistics,'),'37(2), 284–291.', a("https://academic.oup.com/applij/article-abstract/37/2/284/1742739", href="https://academic.oup.com/applij/article-abstract/37/2/284/1742739", target="_blank")), br(), strong('Author'), p(a("Atsushi MIZUMOTO,", href="https://mizumot.com", target="_blank"),' Ph.D.',br(), 'Professor of Applied Linguistics',br(), 'Faculty of Foreign Language Studies /',br(), 'Graduate School of Foreign Language Education and Research,',br(), 'Kansai University, Osaka, Japan'), br(), strong('License'), br(), a(img(src="https://mizumot.com/images/by-sa.png"), target="_blank", href="https://creativecommons.org/licenses/by-sa/4.0/deed"), p(br()) ) ) ) ))