ICCfit: Empirical item characteristic curve plot for the Rasch Model~

In david-ti/wrightmap: IRT Item-Person Map with 'ConQuest' Integration

Empirical item characteristic curve plot for the Rasch Model~

Description

The ICCfit function is intended for contrasting a Rasch model's expected item characteristic curve against the empirical data from dichotomous responses. The ICCfit function displays a confidence interval for the model based curve and plots the confidence interval for the empirical proportions.

Usage

ICCfit(itemNumber, observedResponses, personEstimates, 
    itemParameters, xlim = c(-4, 4), method = "Quantile", NQtiles = 10)

Arguments

itemNumber	The position of the item in the test. This position is used to select the column of observed responses and the item difficulty among the item parameters.
observedResponses	Data frame or matrix with observed responses. The data frame or matrix should be of size N * I, where N is the number of respondents and I is the number of items in the model.
personEstimates	A vector of length N containing the model based person estimates or predictions.
itemParameters	A data frame or matrix of dimensions I * 2 containing the model based item difficulty estimates in the firs column and the parameter standard error in the second column.
xlim	Vector with two values indicating the minimum and maximum values to be used when plotting the item characteristic curve.
method	Selects the method used to group the respondents: Quantile (default), ByPersonEstimate, and Histogram (see ‘Details’).
NQtiles	When using the Quantile method this value controls how many grouping will be used: 4 groups cases groups respondents by quartiles, 5 by quintiles, 10 by deciles, etc.

Details

The function uses the item difficulty parameter to generate the model based curve and the item difficulty parameter standard error to plot a confidence interval around the curve. The observed responses are then grouped using the selected method in order to contrast the model predicted response probability with the observed proportion. By default the function uses deciles to generate the respondent groups. The function allows the method ByPersonEstimate in order to make a different group for each observed person estimate (potentially useful when analyzing test data with large numbers with no missing data), and the Histogram method, which uses the Freedman-Diaconis algorithm to select the width of the bands used for grouping.

Author(s)

David Torres Irribarra

See Also

CCCfit

Examples

##---- Should be DIRECTLY executable !! ----
##-- ==>  Define data, use random,
##--	or do  help(data=index)  for the standard data sets.

## The function is currently defined as
function (itemNumber, observedResponses, personEstimates, itemParameters, 
    xlim = c(-4, 4), method = "Quantile", NQtiles = 10) 
{
    propCI <- function(propVector, nVector, alpha = 0.05) {
        propSE <- sqrt(propVector * (1 - propVector)/nVector)
        propLB <- propVector - (propSE * qnorm(1 - (alpha/2)))
        propUB <- propVector + (propSE * qnorm(1 - (alpha/2)))
        data.frame(propSE = propSE, propLB = propLB, propUB = propUB)
    }
    plotICC <- function(difficulty, range = xlim) {
        invlogit <- function(x) {
            1/(1 + exp(-x))
        }
        params <- data.frame(ability = seq(-4, 4, length.out = 1000), 
            difficulty = difficulty)
        probs <- invlogit(params[, 1] - params[, 2])
        lines(params[, 1], probs)
    }
    plotICCerrors <- function(difficulty, dSE, range = xlim) {
        invlogit <- function(x) {
            1/(1 + exp(-x))
        }
        params <- data.frame(ability = seq(-4, 4, length.out = 1000), 
            lb = difficulty - 1.96 * dSE, ub = difficulty + 1.96 * 
                dSE)
        probslb <- invlogit(params[, 1] - params[, 2])
        probsub <- invlogit(params[, 1] - params[, 3])
        xCoords <- c(ability = seq(-4, 4, length.out = 1000), 
            ability = seq(4, -4, length.out = 1000))
        yCoords <- c(probslb, rev(probsub))
        polygon(xCoords, yCoords, col = "grey75", border = NA)
    }
    if (method == "ByPersonEstimate") {
        aggdata <- aggregate(observedResponses, by = list(round(personEstimates, 
            2)), FUN = mean, na.rm = TRUE)
        aggdata[, -1][aggdata[, -1] == 1] <- 0.999
        aggdata[, -1][aggdata[, -1] == 0] <- 0.001
        sampleSizeAggdata <- aggregate(is.na(observedResponses), 
            by = list(round(personEstimates, 2)), FUN = length)
    }
    if (method == "Quantile") {
        cutPoints <- quantile(personEstimates, seq(0, 1, length = NQtiles + 
            1))
        aggdata <- aggregate(observedResponses, by = list(cut(personEstimates, 
            cutPoints)), FUN = mean, na.rm = TRUE)
        breakMeans <- aggregate(personEstimates, by = list(cut(personEstimates, 
            cutPoints)), FUN = mean, na.rm = TRUE)
        aggdata[, 1] <- breakMeans[, 2]
        aggdata[, -1][aggdata[, -1] == 1] <- 0.999
        aggdata[, -1][aggdata[, -1] == 0] <- 0.001
        sampleSizeAggdata <- aggregate(is.na(observedResponses), 
            by = list(cut(personEstimates, cutPoints)), FUN = length)
        sampleSizeAggdata[, 1] <- breakMeans[, 2]
    }
    if (method == "Histogram") {
        histData <- hist(personEstimates, breaks = "FD", plot = FALSE)
        cutPoints <- histData$breaks
        aggdata <- aggregate(observedResponses, by = list(cut(personEstimates, 
            cutPoints)), FUN = mean, na.rm = TRUE, drop = FALSE)
        breakMeans <- aggregate(personEstimates, by = list(cut(personEstimates, 
            cutPoints)), FUN = mean, na.rm = TRUE, drop = FALSE)
        aggdata[, 1] <- histData$mids
        aggdata[, -1][aggdata[, -1] == 1] <- 0.999
        aggdata[, -1][aggdata[, -1] == 0] <- 0.001
        sampleSizeAggdata <- aggregate(is.na(observedResponses), 
            by = list(cut(personEstimates, cutPoints)), FUN = length, 
            drop = FALSE)
        sampleSizeAggdata[, 1] <- histData$mids
    }
    plot(aggdata[, 1], aggdata[, itemNumber + 1], type = "n", 
        axes = FALSE, , ylab = "Proportion", xlab = "Proficiency", 
        ylim = c(0, 1), xlim = xlim)
    plotICCerrors(itemParameters[itemNumber, 1], itemParameters[itemNumber, 
        2])
    plotICC(itemParameters[itemNumber, 1])
    cbind(aggdata[, 1], propCI(aggdata[, itemNumber + 1], sampleSizeAggdata[, 
        itemNumber + 1]))
    apply(cbind(aggdata[, 1], propCI(aggdata[, itemNumber + 1], 
        sampleSizeAggdata[, itemNumber + 1])), 1, function(x) segments(x0 = x[1], 
        y0 = x[3], x1 = x[1], y1 = x[4], col = "#31333450"))
    points(aggdata[, 1], aggdata[, itemNumber + 1], pch = 18, 
        cex = 0.75, col = "#31333450")
    axis(2, las = 1)
    axis(1)
    title(paste("Item", rownames(itemParameters)[itemNumber]))
    print(sampleSizeAggdata)
  }

david-ti/wrightmap documentation built on May 24, 2022, 3:53 p.m.