mirt (version 1.17.1)

mirt.model: Specify model loadings

Description

The mirt.model function scans/reads user input to specify the confirmatory model. Item locations must be used in the specifications if no itemnames argument is supplied. This is called implicitly by estimation functions when a string is passed to the model argument.

Usage

mirt.model(input = NULL, itemnames = NULL, file = "", COV = NULL,
  quiet = TRUE, ...)

Arguments

input

input for writing out the model syntax. Can either be a string declaration of class character or the so-called Q-matrix or class matrix that specifies the model either with integer or logical values. If the Q-matrix method is chosen covarianc

itemnames

a character vector or factor indicating the item names. If a data.frame or matrix object is supplied the names will be extracted using colnames(itemnames). Supplying this input allows the syntax to be specified with the raw item names rather

file

a input specifying an external file that declares the input.

COV

a symmetric, logical matrix used to declare which covariance terms are estimated

quiet

logical argument passed to scan() to suppress console read message

...

additional arguments for scan()

Value

• Returns a model specification object to be used in mirt, bfactor, multipleGroup, or mixedmirt

Details

Factors are first named and then specify which numerical items they affect (i.e., where the slope is not equal to 0), separated either by commas or by - to indicate a range of items. Products between factors may be specified by enclosing the left hand term within brackets. To finish the declaration of a model simply enter a blank line with only a carriage return (i.e., the 'enter' or 'return' key), or instead read in an input version of the model syntax.

There is an optional keyword for specifying the correlation between relationships between factors called COV, and non-linear factor products can be included by enclosing the product combination on the left hand side of the declaration (e.g., (F1*F1) would create a quadratic factor for F1).

[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object]

Examples

# interactively through the console (not run)
#model <- mirt.model()
#  F1 = 1,2,3,4-10
#  F2 = 10-20
#  (F1*F2) = 1,2,3,4-10
#  COV = F1*F2


#Or alternatively with a string input
s <- 'F1 = 1,2,3,4-10
      F2 = 10-20
      (F1*F2) = 1,2,3,4-10
      COV = F1*F2'
model <- mirt.model(s)

# strings can also be passed to the estimation functions directly,
#   which silently calls mirt.model(). E.g., using the string above:
# mod <- mirt(data, s)


#Q-matrix specification
Q <- matrix(c(1,1,1,0,0,0,0,0,0,1,1,1), ncol=2, dimnames = list(NULL, c('Factor1', 'Factor2')))
COV <- matrix(c(FALSE, TRUE, TRUE, FALSE), 2)
model <- mirt.model(Q, COV=COV)

## constrain various items slopes and all intercepts in single group model to be equal,
#   and use a log-normal prior for all the slopes
s <- 'F = 1-10
      CONSTRAIN = (1-3, 5, 6, a1), (1-10, d)
      PRIOR = (1-10, a1, lnorm, .2, .2)'
model <- mirt.model(s)


## constrain various items slopes and intercepts across groups for use in multipleGroup(),
#  and constrain first two slopes within 'group1' to be equal
s <- 'F = 1-10
      CONSTRAIN = (1-2, a1)
      CONSTRAINB = (1-3, 5, 6, a1), (1-10, d)'
model <- mirt.model(s)


## specify model using raw item names
data(data.read, package = 'sirt')
dat <- data.read

# syntax with variable names
mirtsyn2 <- "
       F1 = A1,B2,B3,C4
       F2 = A1-A4,C2,C4
       MEAN = F1
       COV = F1*F1, F1*F2
       CONSTRAIN=(A2-A4,a2),(A3,C2,d)
       PRIOR = (C3,A2-A4,a2,lnorm, .2, .2),(B3,d,norm,0,.0001)"
# create a mirt model
mirtmodel <- mirt.model(mirtsyn2, itemnames=dat)
# or equivelently:
# mirtmodel <- mirt.model(mirtsyn2, itemnames=colnames(dat))

# mod <- mirt(dat , mirtmodel)