R Statistics Rasch analysis software and Winsteps

R command

R's response

Meaning

ls()

[1] "IFILE"

list of loaded datasets

help(ls)

Help windows opens

help for "ls" command

names(IFILE)

[1] "ENTRY" "NAME"

names of variables in the dataset "IFILE"

IFILE

(contents of IFILE)

displays the data in the IFILE dataset

attach(IFILE)

-

IFILE variables active without $ reference

plot (x, y)

scatterplot displays

variable y is plotted against variable x

q()

(R terminates)

quit (exit from) R Statistics

Maximize your R Console window: edit your "Rconsole" file in "C:\Program Files\R\R-...\etc" from "# MDIsize = 0*0+0+0" to "MDIsize = 0*0+0+0"

Action:

Screen:

Import R statistics data in a .rdata or .rda file.

Now follow the procedure at Excel/RSSST menu.

Export Winsteps output files as R statistics files: .rdata or ,rda

from the Output Files menu

IFILE=

PFILE=

IPMATRIX=

Code for missing data: NA is usually applied automatically

R Statistics screen

ls() at any time to see list of available objects, etc.

R functions

data[data=="."] <- NA # Convert missing data to NA (R Statistics missing data code)

Descriptive statistics

> install.packages('psych'')

> library (psych)

> des <- describe(data)

> print (des, digits=3)

  vars   n   mean    sd median trimmed   mad    min   max range   skew kurtosis    se  Q0.25 Q0.75

ZSCORE 476 -0.009 0.827  -0.01   0.006 0.304 -5.969 6.141 12.11 -0.985   21.018 0.038 -0.163  0.21 

Principal Components Analysis and

Factor analysis of standardized residuals

or other numbers

> install.packages('psych',''FactoMineR')

> library (psych)

> library (FactoMineR)

> result <- fa.parallel (data)

a Scree Plot displays of PCA and FA eigenvalues.

R Console window: "Parallel analysis suggests that the number of factors =  3"

Put 3 (or whatever) in the R "factors" instruction

Principal Components Analysis:

> pca <- PCA(data, graph=FALSE)

> pca$eig

> pca$var$coord

> head(pca$ind$coord)

Factor analysis:

> factors <- fa(data, 3)

> print (factors)

    the factor loadings are in columns MR1, MR2, ...

> plot(factors)

Simple plot with R

Winsteps Output Files menu: IFILE=, PFILE=. Output to R Statistics. In R Statistics:

res -> attach(IFILE)

res -> plot (IN.MSQ, OUT.MSQ)

3-D scatter plot

Use Winsteps Plots menu

Winsteps Output Files menu: IFILE=, PFILE=: Select any 3 variables, e.g. Measure, Infit MnSq, Outfit MnSQ. Output to R Statistics

In R Statistics:

> install.packages("scatterplot3d")

> library (scatterplot3d)

> res <- scatterplot3d(IFILE)

Mokken scaling:

Loevinger H Coefficient

Output files menu: IPMATRIX=

scored responses, no extreme scores. no person entry numbers, missing data = NA

OK

then when R displays:

install.packages("mokken")

library(mokken)

coefH(data)

Exploratory Factor Analysis (PCA) Functions for Assessing Dimensionality

Output files menu: IPMATRIX=

standardized residuals, no extreme scores, no person entry numbers, missing data = 0

install.packages("remotes")

remotes::install_github("bpoconnor/paramap")

library(paramap)

MAP(data)

CMLE Conditional Maximum Likelihood Estimation: Dichotomous

Output files menu: IPMATRIX=

scored responses, no extreme scores. no person entry numbers, missing data = NA

> install.packages("eRm")

> library(eRm)

# activate eRm

> res <- RM(data)  # CMLE estimation of item easinesses for dichotomies

                                         # RSM() and PCM() for polytomies

> coef(res)  # or  summary(res)  # report the items

> pres <- person.parameter(res)   # AMLE estimation of person abilities (thetas)

> coef(pres)  # or summary(pres)  # report the person estimates

> LRtest(res, splitcr = "median", se = TRUE) # Andersen's Likelihood Ratio Test (LRT)

also

> install.packages("RM.weights")

> library (RM.weights)

> res <- RM.w(data, .w = NULL, .d=NULL, country=NULL, se.control = TRUE,

quantile.seq = NULL, write.file = FALSE, max.it=100)    # dichotomies

res <- PC.w(data, wt=NULL, extr=NULL, maxiter=100,minconv=.00001,country=NULL, write.file=FALSE, recode = 0, write.iteration=FALSE) # partial credit

ERMA (Everyone's Rasch Measurement Analyzer)

> install.packages("shiny")

> library(shiny)

> runGitHub("ShinyERMA", "RaschModel")

Jue Wang (University of Miami) and George Engelhard Jr. (University of Georgia)

JMLE Joint Maximum Likelihood Estimation: TAM

Output files menu: IPMATRIX=

scored responses start from zero, no extreme item scores. no person entry numbers, missing data = NA

> install.packages("TAM")

> library(TAM)  # activate ltm

> res <- tam.jml(data)

> summary(res)  # report the items

MMLE Marginal Maximum Likelihood Estimation:

Dichotomous

Output files menu: IPMATRIX=

scored responses, no extreme item scores. no person entry numbers, missing data = NA

> install.packages("ltm")

> library(ltm)  # activate ltm

> res <- rasch(data =data, constraint = cbind(ncol(data) + 1, 1))

> coef(res)  # or  summary(res)  # report the items

Generalized Partial Credit Model:

> res <- gpcm(data, constraint = c("gpcm", "1PL", "rasch"), IRT.param = TRUE,

start.val = NULL, na.action = NULL, control = list())

MMLE Marginal Maximum Likelihood Estimation:

Dichotomous

Polytomous, Rating Scale

Output files menu: IPMATRIX=

scored responses start from zero, no extreme item scores. no person entry numbers, missing data = NA

> install.packages("TAM")

> library(TAM)  # activate TAM

> res <- TAM::tam.mml( data, irtmodel="1PL" ) # dichotomous

> res <- TAM::tam.mml( data, irtmodel="RSM" )

>summary(res)  # report the items and thresholds

>pers <- tam.mml.wle( res, score.resp=NULL, WLE=FALSE, adj=.3) # compute AMLE person thetas

pers$theta # list the thetas

PMLE Pair-wise Maximum Likelihood Estimation: dichotomous

Output files menu: IPMATRIX=

scored responses, no extreme scores. no person entry numbers, missing data = NA

> install.packages("pairwise")

> library(pairwise)  # activate pairwise

> res <- pair(data)

> summary(res)

and

> install.packages("sirt")

> library("sirt")

> res <- rasch.pairwise(data)

> summary(res)

other Rasch packages

> install.packages("mixRasch")

> install.packages("pcIRT")

>install.packages('mirt')

> res <- mirt (data, 1, 'Rasch') # for dichotomous and partial credit analysis

Wright Map

Use Winsteps Plots menu

Receiver Operating Curve

Area Under the Curve

ROC AUC

install.packages("PRROC")

library(PRROC)

res <- roc.curve(scores.class0 = XFILE$EXPECTATION, weights.class0=XFILE$ORDERED,curve=TRUE)

plot(res)

R Graphics

ggplot2

Use Winsteps Graph window or

> install.packages("ggplot2")

> library(ggplot2)

if this fails, then

> remove.packages("ggplot2")

then try again

> data (diamonds)

? ggplot(diamonds, aes(x=carat, y=price))

Cluster Analysis with complete data

install.packages ("pvclust")

library(pvclust)

data <- scale(data)

fit <- pvclust(data, method.hclust="ward.D2",    method.dist="euclidean")

plot(fit) # dendogram with p values

pvrect(fit, alpha=.95) # red boxes around significantly different clusters

Generalizability Theory

> remove.packages("Matrix")

> install.packages("Matrix")

> install.packages("gtheory")

> library(Matrix)

> library(gtheory)

#A univariate D study.

#Compare to results on page 116 of Brennan (2001).

data(Brennan.3.2)

formula.Brennan.3.2 <- "Score ~ (1 | Person) + (1 | Task) + (1 | Rater:Task) +

(1 | Person:Task)"

gstudy.out <- gstudy(data = Brennan.3.2, formula = formula.Brennan.3.2)

dstudy(gstudy.out, colname.objects = "Person", data = Brennan.3.2, colname.scores = "Score")

#A multivariate D study.

#Compare to results on pages 270-272 of Brennan (2001).

data(Rajaratnam.2)

formula.Rajaratnam.2 <- "Score ~ (1 | Person) + (1 | Item)"

gstudy.out <- gstudy(data = Rajaratnam.2, formula = formula.Rajaratnam.2,

colname.strata = "Subtest", colname.objects = "Person")

dstudy(gstudy.out, colname.objects = "Person", data = Rajaratnam.2, colname.scores = "Score",

colname.strata = "Subtest", weights = c(0.25, 0.5, 0.25))

Reading SAS file

> install.packages("haven")

> library(haven)

>  read_sas("c:/winsteps/examples/sastest.sas7bdat")

Reading Excel file

install.packages("readxl")

# Loading

library("readxl")

# xls files

my_data <- read_excel("my_file.xls")

write.table(my_data, file = "c:/filename.txt", sep = "\t", quote = FALSE)

Mokken scaling

Loevinger H Coefficient

install.packages("mokken")

library(mokken)

# no missing values: rating scales allowed

# Partition the the scale into mokken scales

aisp(data)

# Investigate the assumption of monotonicity

monotonicity.list <- check.monotonicity(data)

summary(monotonicity.list)

# plot(monotonicity.list)

# Investigate the assumption of non-intersecting ISRFs using method restscore

restscore.list <- check.restscore(data)

summary(restscore.list)

# plot(restscore.list)

# Investigate the assumption of non-intersecting ISRFs using method pmatrix

# pmatrix.list <- check.pmatrix(data)

# summary(pmatrix.list)

# plot(pmatrix.list)

# Investigate the assumption of IIO using method MIIO

iio.list <- check.iio(data)

summary(iio.list)

# plot(iio.list)

# Compute the reliability of the scale

check.reliability(data)

DIF / DPF analysis

> install.packages("difR")

> library(difR)

DIF: start with mantelhaenszel

DPF: transpose your data: t()

then mantelhaenszel

Nonparametric Rasch Model Tests

install.packages("eRm")

library(eRm)

# no missing values: must be binary

rmat <- as.matrix(data)

# T1: Checks for local dependence via increased inter-item correlations. For all item pairs, cases are counted with equal responses on both items.

t1 <- NPtest(rmat, n = 100, method = "T1")

print(t1, alpha = 0.01)

print(t1, alpha = 0.05)

# T1m: Checks for multidimensionality via decreased inter-item correlations. For all item pairs, cases are counted with equal responses on both items.

t1m <- NPtest(rmat, n = 100, method = "T1m")

print(t1m, alpha = 0.01)

print(t1m, alpha = 0.05)

# T10: Global test for subgroup-invariance. Checks for different item difficulties in two subgroups

t101 <- NPtest(rmat, method = "T10") # default split criterion is "median"

t101

t102 <- NPtest(rmat, method = "T10", splitcr = "mean")

t102

# T11: Global test for local dependence. The statistic calculates the sum of absolute deviations between the observed inter-item correlations and the expected correlations

t11 <- NPtest(rmat, method = "T11")

t11

Fit Analysis "Aberrance"

install.packages("aberrance"))

library(aberrance)

Latent Class Analysis LCA

install.packages("poLCA")

library(poLCA)

# observations must start at 1

data <- data+1

f <- cbind(X1,X2)~1

p <- loPCA(f, data, nclass=2)

# results difficult to interpret at the variable (item) level

Latent Profile Analysis LPA

library(tidyLPA)

library(dplyr)

data  %>%  estimate_profiles(n_profiles = 1:2, models = 1:2) %>% plot_profiles()

Note: AIC, BIC - the lower the better.

difR: Collection of Methods to Detect Dichotomous Differential Item Functioning (DIF)

install.packages("difR")

library(difR)