grm-methods {plink}

Graded Response Model Probabilities

Description

This function computes the cumulative probability of responding within or above a certain category or the probability of responding in a specific category for one or more items for a given set of theta values using the graded response model or multidimensional graded response model.

Usage

grm(x, cat, theta, dimensions = 1, catprob = FALSE, D = 1, 
  location = FALSE, items, information = FALSE, angle, ...)

## S4 method for signature 'matrix', 'numeric'
grm(x, cat, theta, dimensions, catprob, D, location, items, information, angle, ...)

## S4 method for signature 'data.frame', 'numeric'
grm(x, cat, theta, dimensions, catprob, D, location, items, information, angle, ...)

## S4 method for signature 'list', 'numeric'
grm(x, cat, theta, dimensions, catprob, D, location, items, information, angle, ...)

## S4 method for signature 'irt.pars', 'ANY'
grm(x, cat, theta, dimensions, catprob, D, location, items, information, angle, ...)

## S4 method for signature 'sep.pars', 'ANY'
grm(x, cat, theta, dimensions, catprob, D, location, items, information, angle, ...)

Arguments

x

ans R object containing item parameters.

cat

vector identifying the number of response categories (not the number of threshold parameters) for each item.

theta

vector, matrix, or list of theta values for which probabilities will be computed. If theta is not specified, an equal interval range of values from -4 to 4 is used with an increment of 0.5. See details below for more information.

dimensions

number of modeled dimensions

catprob

if TRUE, compute category probabilities instead of cumulative probabilities

D

scaling constant. The default value assumes that the parameters are already in the desired metric. If the parameters are in the logistic metric, they can be transformed to a normal metric by setting D = 1.7

location

if TRUE, the step parameters are deviations from a difficulty parameter

items

numeric vector identifying the items for which probabilities should be computed

information

logical value. If TRUE compute item information. In the multidimensional case, information will be computed in the directions specified by angle or default angles of 0 - 90 in increments of 10 degrees.

angle

vector or matrix of angles between the dimension 1 axis and the corresponding axes for each of the other dimensions for which information will be computed. When there are more than two dimensions and angle is a vector, the same set of angles will be used relative to each of the corresponding axes.

...

further arguments passed to or from other methods

Details

theta can be specified as a vector, matrix, or list. For the unidimensional case, theta should be a vector. If a matrix or list of values is supplied, they will be converted to a single vector of theta values. For the multidimensional case, if a vector of values is supplied it will be assumed that this same set of values should be used for each dimension. Probabilities will be computed for each combination of theta values. Similarly, if a list is supplied, probabilities will be computed for each combination of theta values. In instances where probabilities are desired for specific combinations of theta values, a j x m matrix should be specified for j ability points and m dimensions where the columns are ordered from dimension 1 to m.

Value

Returns an object of class irt.prob

Methods

x = "matrix", cat = "numeric"

This method allows one to specify an n x k matrix for n items. The number of columns can vary depending on the number of dimensions and whether a location parameter is included. The first m columns, for m dimensions, are for item slopes and the remaining columns are for the threshold/threshold deviation parameters.

Threshold parameters can be characterized in two ways: the actual thresholds or deviations from an overall item difficulty (location). In the deviation scenario the location argument should be TRUE and the location parameters should be in the m+1 column. The columns for the threshold/threshold deviation parameters will always follow the slope column(s) and, if applicable, the location column. The number of threshold/threshold deviation parameters can vary for each item. In these instances, all cells with missing values should be filled with NAs. For example, for a unidimensional model with no location parameter, if one item has five categories (four threshold parameters) and another item has three categories (two threshold parameters), there should be five columns. The first column includes the slope parameters and columns 2-5 include the threshold parameters. The values in the last two columns for the item with three categories should be NA.

x = "data.frame", cat = "numeric"

See the method for x = "matrix"

x = "list", cat = "numeric"

This method is for a list with two elements. The first list element is for item slopes and the second for the threshold/threshold deviation parameters. For the unidimensional case, the slope values should be a vector of length n or an n x 1 matrix for n items. For the multidimensional case, the slopes should be specified in an n x m matrix. For both the unidimensional and multidimensional cases, the threshold/threshold deviation parameters should be formatted as an n x k matrix. If the thresholds are deviations from a location parameter, the argument location should be TRUE and the location parameters should be in the first column. The number of threshold/threshold deviation parameters can vary for each item. In these instances, all cells with missing values should be filled with NAs (See the example in the method for x = "matrix").

x = "irt.pars", cat = "ANY"

This method can be used to compute probabilities for the grm items in an object of class "irt.pars". If x contains dichotomous items or items associated with another polytomous model, a warning will be displayed stating that probabilities will be computed for the grm items only. If x contains parameters for multiple groups, a list of "irt.prob" objects will be returned. The argument dimensions does not need to be included for this method.

x = "sep.pars", cat = "ANY"

This method can be used to compute probabilities for the grm items in an object of class sep.pars. If x contains dichotomous items or items associated with another polytomous model, a warning will be displayed stating that probabilities will be computed for the grm items only. The argument dimensions does not need to be included for this method.

Author(s)

Jonathan P. Weeks [e-mail]

References

Embretson, S. E., & Reise, S. P. (2000) Item Response Theory for Psychologists. Mahwah, New Jersey: Lawrence Erlbaum Associates.

Kolen, M. J. & Brennan, R. L. (2004) Test Equating, Scaling, and Linking (2nd ed.). New York: Springer.

Muraki, E., & Carlson, J. E. (1995). Full-information factor analysis for polytomous item responses. Applied Psychological Measurement, 19(1), 73-90.

Samejima, F. (1969) Estimation of latent ability using a response pattern of graded scores. Psychometrika Monograph, No. 17.

Samejima, F. (1996) The graded response model. In W. J. van der Linden & Hambleton, R. K. (Eds.) Handbook of Modern Item Response Theory (pp. 85-100). New York: Springer-Verlag.

Weeks, J. P. (2010) plink: An R package for linking mixed-format tests using IRT-based methods. Journal of Statistical Software, 35(12), 1–33. URL http://www.jstatsoft.org/v35/i12/

See Also

mixed: compute probabilities for mixed-format items
plot: plot item characteristic/category curves
irt.prob, irt.pars, sep.pars: classes

Examples

###### Unidimensional Examples ######
## Cumulative probabilities 
## Item parameters from Embretson & Reise (2000, p. 101) items 9-11
# Includes a location parameter
a <- c(2.09,1.18,1.69)
b <- matrix(c(-1.07,-1.03,.39,.86,1.78,1.85,-.87,-.39,.31,NA,
  -1.4,-.42,NA,NA,-1.74),3,5) 
pars <- cbind(a,b)
rownames(pars) <- paste("Item",9:11,sep=" ")
colnames(pars) <- c("a","b",paste("c",1:4,sep=""))
cat <- c(3,4,5)
x <- grm(pars,cat,location=TRUE)
plot(x)

## Item parameters from Kolen & Brennan (2004, p. 210)
# Use theta values from -3 to 3 with an increment of 0.5
pars <- t(c(1.2,-.5,.6,1.1,1.3))
x <- grm(pars,5,theta=seq(-3,3,.05))
plot(x,item.lab=FALSE)


## Category probabilities
## Single item
pars <- t(c(1.2,-.5,.6,1.1,1.3))
x <- grm(pars,5,seq(-3,3,.05),catprob=TRUE)
plot(x,item.lab=FALSE)

# Items without location parameter
a <- c(2.09,1.18,1.69)
b <- matrix(c(-1.93,-2.81,-1.46,-.2,-.64,.08,NA,.37,.81,NA,NA,2.13),3,4) 
pars <- cbind(a,b)
rownames(pars) <- paste("Item",9:11,sep=" ")
colnames(pars) <- c("a",paste("b",1:4,sep=""))
cat <- c(3,4,5)
x <- grm(pars,cat,catprob=TRUE)
plot(x)

## Create sep.pars object then compute category probabilities
a <- c(2.09,1.18,1.69)
b <- matrix(c(-1.93,-2.81,-1.46,-.2,-.64,.08,NA,.37,.81,NA,NA,2.13),3,4) 
pars <- cbind(a,b)
cat <- c(3,4,5)
pm <- as.poly.mod(3,"grm")
out <- sep.pars(pars,cat,pm)
x <- grm(out,catprob=TRUE)
plot(x)


###### Multidimensional Examples ######
## Cumulative probabilities
a <- matrix(c(.873, .226, .516, .380, .613, .286 ),3,2,byrow=TRUE)
d <- matrix(c(2.255, 1.334, -.503, -2.051, -3.082,
1.917, 1.074, -.497, -1.521, -2.589,
1.624, .994, -.656, -1.978, NA),3,5,byrow=TRUE)
pars <- cbind(a,d)
cat <- c(6,6,5)
x <- grm(pars,cat,dimensions=2)
plot(x)

# Plot separated response surfaces
plot(x, separate=TRUE, drape=TRUE)

## Category Probabilities
## Use {pars} an {cat} from the example above
x <- grm(pars,cat,dimensions=2, catprob=TRUE)

# plot combined item category surfaces
# The screen argument adjusts the orientation of the axes
plot(x,screen=list(z=-40,x=-60), auto.key=list(space="right"))