fwdmsa-package: Robust Mokken Scale Analysis by Means of the Forward Search...
In fwdmsa: Forward search for Mokken scale analysis

Description Details Author(s) References Examples

The package conducts the Forward Search on test and questionnaire data, and shows forward plots for the detection of outliers.

Package:	fwdmsa
Type:	Package
Version:	0.2
Date:	2011-07-26
License:	GPL Version 2 or later

The package includes the functions

`fs.MSA`	Computes the necessary input for forward plots
`plot.fs.class`	S3 method for forward plots
`fs.MSA.n1`	Computes n1
`plot.fs.n1.class`	S3 method for a plot showing graphically n1

and data set

`acs`	Autonomy-Connectedness Scale

Thanks are due to L. Andries van der Ark for contributing R code, and Marrie Bekker and Marcel van Assen for providing the data set.

Wobbe P. Zijlstra Maintainer: Wobbe P. Zijlstra <w.p.zijlstra@uvt.nl>.

Bekker M. H. J., and Van Assen, M. A. L. M. (2006). A short form of the autonomy scale: Properties of the autonomy-connectedness scale (ACS-30). Journal of Personality Assessment, 86, 51-60.

Van der Ark, L. A. (2007). Mokken scale analysis in R. Journal of Statistical Software. http://www.jstatsoft.org

Zijlstra, W. P., Van der Ark, L. A., and Sijtsma, K. (2011). Robust Mokken scale analysis by means of the forward search algorithm for outlier detection. Multivariate Behavioral Research, 46, 58-89.

## Not run: 
## Analyses of Zijlstra et al. (2010).
## First Forward Search Analysis
   library(fwdmsa)
   data(acs.cont)

 # Determining n1 = 292
 # Takes approximately 40 minutes
   fs1.1.n1 <- fs.MSA.n1(acs.cont, B=100)
   n1 <- fs1.1.n1$n1 

 # Figure 2: Plot of number unique subsamples
   plot(fs1.1.n1)

 # Running the forward search
   fs1.1 <- fs.MSA(acs.cont)

 # Figure 3: Plot of objective function
   plot(fs1.1, type="objective", observations=1:618, col="gray70", n0=TRUE, n1=fs.res.cont.n1$n1, xlim=c(0,650))
   plot(fs1.1, type="objective", id.observation=619, col=1, lwd=2, lty=2, add=TRUE)
   plot(fs1.1, type="objective", observations=589:618, lwd=2, add=TRUE)

 # Figure 4: Gap plot
   plot(fs1.1, type="gap", ylim=c(-10,12), n0=TRUE, n1=292)

 # Figure 5: Follow-up plots
   plot(fs1.1, type="followup", step=543:548, reference.step=543, n0=TRUE, n1=292)

## Remove influential observations from the data
   acs.sus <- acs.cont[-(589:618),]
 
 # Determining n1 = 296
   fs1.2.n1 <- fs.MSA.n1(acs.sus, B=100)
   n1 <- fs1.2.n1$n1 

 # Running the forward search
   fs1.2 <- fs.MSA(acs.sus)

 # Figure 6: Minexcl plot
   plot(fs1.2, type="minexcl", n0=TRUE, n1=296, n2=TRUE)

 # Figure 7: Plot of number of scales
   plot(fs1.2, type="num.scale", n0=TRUE, n1=296, n2=TRUE)

 # Figure 8: Item entry plot for the longest scale
   plot(fs1.2, type="scale", id.scale=1, n0=TRUE, n1=296, n2=TRUE)

## Second Forward Search Analysis
 # Remove bad items from the data
   acs.min.core <- acs.cont[-(589:618),-c(3,7,8,11,13,16)]
 
 # Determining n1 = 302
   fs2.n1 <- fs.MSA.n1(acs.min.core, B=100)
   n1 <- fs2.1.n1$n1 

 # Running the forward search
   fs2 <- fs.MSA(acs.min.core)

 # Figure 9: Plot of restscore regression of item 1 for steps 302 and 589
   plot(fs2, type="restscore", step=302, items=1, lty=2, ylim=c(0,4), n0=TRUE, n1=302, n2=TRUE)
   plot(fs2, type="restscore", step=589, items=1, lty=1, add=TRUE)

 # Figure 10: Plot of estimated IRF of item 1
   plot(fs2, type="IRF", items=1, n0=TRUE, n1=302, n2=TRUE)

 # Figure 11: Plot of coefH
   plot(fs2, type="coefH", n0=TRUE, n1=302, n2=TRUE, ylim=c(.1,.8))

## What if influential observations were not removed from the data
   acs.cont.core <- acs.cont[,-c(3,7,8,11,13,16)]
 # Determining n1 = 347
   fs3.n1 <- fs.MSA.n1(acs.cont.core, B=100)
   n1 <- fs3.n1$n1 

 # Running the forward search
   fs3 <- fs.MSA(acs.cont.core)

 # Figure 12a: Plot of estimated IRF of item 1 with influential observations
   plot(fs3, type="IRF", items=1, n0=TRUE, n1=347, n2=FALSE)

 # Figure 12b: Plot of coefH with influential observations
   plot(fs3, type="coefH", n0=TRUE, n1=347, n2=FALSE, ylim=c(.1,.8))

## End(Not run)