R/lpcm_inits.R

In immer: Item Response Models for Multiple Ratings

## File Name: lpcm_inits.R
## File Version: 0.132

########################################################
# lpcm initial parameters
# This function is mainly copied from the
# pcmodel function from the psychotools package
lpcm_inits <- function( dat, weights, maxK, b_const, W,
        irtmodel, normalization )
{
    I <- ncol(dat)
    m <- I
    oj_vec <- sapply( 1L:I, FUN=function(ii){
                seq( 0, maxK[ii] ) }, simplify=FALSE )
    y <- dat
    ctot <- vector('list', length=m)
    for (j in seq_len(m)){
        ctot[[j]] <- as.vector( tapply( weights, factor(y[, j], levels=oj_vec[[j]]), sum))
    }
    start <- lapply(ctot, function(x){
                - cumsum(diff.default( log(x))) } )
            # delta_jk=log(x_j(k-1) - log(x_jk), beta_jk=cumsum_k=1^k(delta_jk)
    start <- unlist(start)
    start[ is.na(start) ] <- 0
    par_init <- ( start - b_const ) %*% W
    par_init <- par_init[1,]
    if ( ( irtmodel=='PCM2') & ( normalization=='sum') ){
        par_init <- 0 * par_init
    }
    return(par_init)
}
########################################################