Differential Item Functioning (DIF)

DESCRIPTION

@@ -20,11 +20,8 @@ Imports:
     jaspBase,
     jaspGraphs,
     jaspIrtStanModels,
-    ltm,
     mirt,
-    moments,
     parallel,
-    reshape2,
     rstan (>= 2.18.1),
     truncdist
 Remotes:

NAMESPACE

@@ -1,5 +1,4 @@
 import(jaspBase)
-export(classicalTestTheory)
 export(itemResponseTheoryDichotomous)
 export(itemResponseTheoryPolytomous)
 export(itemResponseTheoryDichotomousBayesian)

R/itemresponsetheorycommon.R

@@ -408,8 +408,8 @@
     state <- .irtIRTState(dataset, options, jaspResults)
   }
   plotdata <- state[["plotDataTestInformation"]]
-  xBreaks <- jaspGraphs::getPrettyAxisBreaks(c(-6, 6), min.n = 4)
-  yBreaks <- jaspGraphs::getPrettyAxisBreaks(c(0, subset(plotdata$y, plotdata$x >= -6 & plotdata$x <= 6)), min.n = 4)
+  xBreaks <- jaspGraphs::getPrettyAxisBreaks(c(-3, 3), min.n = 4)
+  yBreaks <- jaspGraphs::getPrettyAxisBreaks(c(0, subset(plotdata$y, plotdata$x >= -3 & plotdata$x <= 3)), min.n = 4)
   p <- ggplot2::ggplot(data = plotdata, mapping = ggplot2::aes(x = x, y = y, color = type)) +
     ggplot2::geom_line() +
     ggplot2::scale_x_continuous(name = "\u03B8", breaks = xBreaks, limits = range(xBreaks)) +
@@ -444,8 +444,8 @@
   }
   plotdata <- state[["plotDataItemInformation"]]
   plotdata <- subset(plotdata, plotdata$item %in% match(options[["plotItemInformationItems"]], options[["items"]]))
-  xBreaks <- jaspGraphs::getPrettyAxisBreaks(c(-6, 6), min.n = 4)
-  yBreaks <- jaspGraphs::getPrettyAxisBreaks(c(0, subset(plotdata$y, plotdata$x >= -6 & plotdata$x <= 6)), min.n = 4)
+  xBreaks <- jaspGraphs::getPrettyAxisBreaks(c(-3, 3), min.n = 4)
+  yBreaks <- jaspGraphs::getPrettyAxisBreaks(c(0, subset(plotdata$y, plotdata$x >= -3 & plotdata$x <= 3)), min.n = 4)
   colors <- colorspace::qualitative_hcl(length(options[["plotItemInformationItems"]]))
   p <- ggplot2::ggplot(data = plotdata, mapping = ggplot2::aes(x = x, y = y, col = factor(item))) +
     ggplot2::geom_line() +
@@ -482,7 +482,7 @@
     object <- createJaspContainer(title = gettext("Item Characteristic Curves"))
   }
   object$position <- position + 1
-  object$dependOn(options = c(.irtCommonDeps(type = "irt"), "plotItemCharacteristicItems", "plotItemCharacteristicLabels"))
+  object$dependOn(options = c(.irtCommonDeps(type = "irt"), "plotItemCharacteristicItems", "plotItemCharacteristicLabels", "plotItemCharacteristicFromX", "plotItemCharacteristicToX"))
   jaspResults[["plotItemCharacteristic"]] <- object
   if (!ready) {
     return()
@@ -492,7 +492,7 @@
   } else {
     state <- .irtIRTState(dataset, options, jaspResults)
   }
-  xBreaks <- jaspGraphs::getPrettyAxisBreaks(c(-6, 6), min.n = 4)
+  xBreaks <- jaspGraphs::getPrettyAxisBreaks(c(options[["plotItemCharacteristicFromX"]], options[["plotItemCharacteristicToX"]]), min.n = 4)
   if (options[["plotItemCharacteristicGroup"]]) {
     plotdata <- state[["plotDataItemCharacteristic"]]
     plotdata <- subset(plotdata, plotdata$item %in% match(options[["plotItemCharacteristicItems"]], options[["items"]]))
@@ -904,3 +904,95 @@
   result <- list(model = model, data = data, pars = pars)
   return(result)
 }
+
+.irtDifAnalysisTable <- function(dataset, options, jaspResults, ready, position) {
+  if (options[["explanatoryText"]] && options[["tableDifAnalysis"]]) {
+    text <- createJaspHtml(gettext("<h3>Explanatory Text: Differential Item Functioning (DIF)</h3> The table below presents the results of a Differential Item Functioning (DIF) analysis. A DIF-analysis is conducted to assess whether specific items perform differently across groups while controlling for overall ability or trait levels. This ensures that the items are fair and not biased toward any particular group."))
+    text$position <- position
+    text$dependOn(options = c("explanatoryText", "tableDifAnalysis"))
+    jaspResults[["tableDifAnalysisText"]] <- text
+  }
+  if (!is.null(jaspResults[["tableDifAnalysis"]]) || !options[["tableDifAnalysis"]]) {
+    return()
+  }
+  tb <- createJaspTable(title = gettext("Differential Item Functioning (DIF)"))
+  tb$position <- position + 1
+  tb$addColumnInfo(name = "item", title = gettext("Item"), type = "string")
+  tb$addColumnInfo(name = "aic", title = gettext("AIC"), type = "number")
+  tb$addColumnInfo(name = "sabic", title = gettext("SABIC"), type = "number")
+  tb$addColumnInfo(name = "hq", title = gettext("HQ"), type = "number")
+  tb$addColumnInfo(name = "bic", title = gettext("BIC"), type = "number")
+  tb$addColumnInfo(name = "x2", title = gettext("X2"), type = "number")
+  tb$addColumnInfo(name = "df", title = gettext("df"), type = "integer")
+  tb$addColumnInfo(name = "p", title = gettext("p"), type = "pvalue")
+  tb$dependOn(options = c(.irtCommonDeps(type = "irt"), "tableDifAnalysis", "groupingVariable", "tableDifAnalysisDifficulty", "tableDifAnalysisDiscrimination", "tableDifAnalysisGuess", "tableDifAnalysisSlip"))
+  tb$addFootnote(gettext("For each item, the null hypothesis specifies that there is no DIF between the groups."))
+  tb$addFootnote(gettext("p-values are not adjusted for multiple comparisons."), colName = "p")
+  if (length(options[["covariates"]]) > 0) {
+    tb$addFootnote(gettext("The latent regressions present in the ungrouped model are not included in this analysis."))
+  }
+  jaspResults[["tableDifAnalysis"]] <- tb
+  if (!ready || options[["groupingVariable"]] == "") {
+    return()
+  }
+  state <- .irtIRTStateBayesian(dataset, options, jaspResults)
+  parameters <- character()
+  if (options[["dichotomous"]]) {
+    if (options[["tableDifAnalysisDifficulty"]]) {
+      parameters <- c(parameters, "d")
+    }
+    if (options[["tableDifAnalysisDiscrimination"]] && options[["model"]] %in% c("2PL", "3PL", "4PL")) {
+      parameters <- c(parameters, "a1")
+    }
+    if (options[["tableDifAnalysisGuess"]] && options[["model"]] %in% c("3PL", "4PL")) {
+      parameters <- c(parameters, "g")
+    }
+    if (options[["tableDifAnalysisSlip"]] && options[["model"]] == "4PL") {
+      parameters <- c(parameters, "u")
+    }
+  } else {
+    all_parameters <- colnames(state[["coefficients"]])
+    if (options[["tableDifAnalysisDifficulty"]]) {
+      if (options[["model"]] %in% c("rsm", "grsm")) {
+        parameters <- c(parameters, "c")
+      } else if (options[["model"]] == "nominal") {
+        colnames_diff <- colnames(state[["coefficients"]])[grep("c", colnames(state[["coefficients"]]))]
+        colnames_diff <- gsub("c", "", colnames_diff)
+        colnames_diff <- paste0("d", as.numeric(colnames_diff) - 1)
+        parameters <- c(parameters, colnames_diff)
+      } else {
+        colnames_diff <- colnames(state[["coefficients"]])[grep("b", colnames(state[["coefficients"]]))]
+        colnames_diff <- gsub("b", "d", colnames_diff)
+        parameters <- c(parameters, colnames_diff)
+      }
+    }
+    if (options[["tableDifAnalysisDiscrimination"]] && options[["model"]] %in% c("gpcm", "grsm", "graded", "nominal")) {
+      if (options[["model"]] == "nominal") {
+        colnames_disc <- colnames(state[["coefficients"]])[grep("a", colnames(state[["coefficients"]]))]
+        colnames_disc <- gsub("a", "", colnames_disc)
+        colnames_disc <- paste0("ak", as.numeric(colnames_disc) - 1)
+        parameters <- c(parameters, colnames_disc)
+      } else {
+        parameters <- c(parameters, "a1")
+      }
+    }
+    if (options[["tableDifAnalysisThreshold"]] && options[["model"]] %in% c("rsm", "grsm")) {
+      colnames_tresh <- colnames(state[["coefficients"]])[grep("b", colnames(state[["coefficients"]]))]
+      parameters <- c(parameters, colnames_tresh)
+    }
+  }
+  if (length(parameters) == 0) {
+    tb$setError(gettext("DIf-analysis not possible: Select at least one parameter to test."))
+    return()
+  }
+  fit <- mirt::multipleGroup(data = state[["items"]], model = 1, itemtype = options[["model"]], group = dataset[[options[["groupingVariable"]]]], SE = FALSE, verbose = FALSE, TOL = options[["emTolerance"]], technical = list(NCYCLES = options[["emIterations"]], set.seed = options[["seed"]]))
+  dif <- mirt::DIF(fit, which.par = parameters)
+  tb[["item"]] <- options[["items"]]
+  tb[["aic"]] <- dif[["AIC"]]
+  tb[["sabic"]] <- dif[["SABIC"]]
+  tb[["hq"]] <- dif[["HQ"]]
+  tb[["bic"]] <- dif[["BIC"]]
+  tb[["x2"]] <- dif[["X2"]]
+  tb[["df"]] <- dif[["df"]]
+  tb[["p"]] <- dif[["p"]]
+}

R/itemresponstheorydichotomous.R

@@ -30,17 +30,20 @@ itemResponseTheoryDichotomous <- function(jaspResults, dataset, options, ...) {
   # Create the item fit statistics table
   .irtIRTItemFitStatisticsTable(dataset, options, jaspResults, ready, position = 5)
 
+  # Create the DIF-analysis table
+  .irtDifAnalysisTable(dataset, options, jaspResults, ready, position = 7)
+
   # Create the histogram of latent ability
-  .irtIRTHistogram(dataset, options, jaspResults, ready, position = 7)
+  .irtIRTHistogram(dataset, options, jaspResults, ready, position = 9)
 
   # Create the test information function
-  .irtIRTTestInfoCurve(dataset, options, jaspResults, ready, position = 9)
+  .irtIRTTestInfoCurve(dataset, options, jaspResults, ready, position = 11)
 
   # Create the item information curves
-  .irtIRTItemInfoCurve(dataset, options, jaspResults, ready, position = 11)
+  .irtIRTItemInfoCurve(dataset, options, jaspResults, ready, position = 13)
 
   # Create the item information curves
-  .irtIRTItemCharCurve(dataset, options, jaspResults, ready, position = 13)
+  .irtIRTItemCharCurve(dataset, options, jaspResults, ready, position = 15)
 }
 
 .irtIRTState <- function(dataset, options, jaspResults) {
@@ -55,12 +58,11 @@ itemResponseTheoryDichotomous <- function(jaspResults, dataset, options, ...) {
     } else {
       covariates <- NULL
     }
-    # Model fit (DIF analysis possible using multipleGroup model)
     fit <- mirt::mirt(data = items, model = 1, itemtype = options[["model"]], covdata = covariates, formula = ~., SE = FALSE, verbose = FALSE, TOL = options[["emTolerance"]], technical = list(NCYCLES = options[["emIterations"]], set.seed = options[["seed"]]))
-    if (options[["model"]] == "grsm") {
-      thetaRange <- seq(-10, 10, by = 0.1)
+    if (options[["model"]] %in% c("rsm", "grsm")) {
+      thetaRange <- seq(-6, 6, by = 0.05)
     } else {
-      thetaRange <- seq(-25, 25, by = 0.01) # Takes too long with grsm
+      thetaRange <- seq(-25, 25, by = 0.01) # Takes too long with grsm or rsm
     }
     latentScores <- as.numeric(mirt::fscores(fit))
     # Tables

R/itemresponstheorypolytomous.R

@@ -30,15 +30,18 @@ itemResponseTheoryPolytomous <- function(jaspResults, dataset, options, ...) {
   # Create the item fit statistics table
   .irtIRTItemFitStatisticsTable(dataset, options, jaspResults, ready, position = 5)
 
+  # Create the DIF-analysis table
+  .irtDifAnalysisTable(dataset, options, jaspResults, ready, position = 7)
+
   # Create the histogram of latent ability
-  .irtIRTHistogram(dataset, options, jaspResults, ready, position = 7)
+  .irtIRTHistogram(dataset, options, jaspResults, ready, position = 9)
 
   # Create the test information function
-  .irtIRTTestInfoCurve(dataset, options, jaspResults, ready, position = 9)
+  .irtIRTTestInfoCurve(dataset, options, jaspResults, ready, position = 11)
 
   # Create the item information curves
-  .irtIRTItemInfoCurve(dataset, options, jaspResults, ready, position = 11)
+  .irtIRTItemInfoCurve(dataset, options, jaspResults, ready, position = 13)
 
   # Create the item information curves
-  .irtIRTItemCharCurve(dataset, options, jaspResults, ready, position = 13)
+  .irtIRTItemCharCurve(dataset, options, jaspResults, ready, position = 15)
 }

inst/Description.qml

@@ -33,11 +33,6 @@ Description
 		title:		qsTr("Polytomous Item Response Theory")
 		func:		"itemResponseTheoryPolytomous"
 	}
-	Analysis
-	{
-		title:		qsTr("Classical Test Theory")
-		func:		"classicalTestTheory"
-	}
 
 	GroupTitle
 	{