R/likelihood_adjustment_aux.R

Defines functions trait_normal_approx likelihood_adjustment_tuning like_adj_EAP_reliability likelihood_adjustment_compute

## File Name: likelihood_adjustment_aux.R
## File Version: 0.19


#######################################################
# compute adjusted likelihood
likelihood_adjustment_compute <- function( likelihood, theta,
            thetaM, adjfac, tuningfac=1)
{
    TP <- length(theta)
    # compute weighted mean and weighted SD
    res0 <- likelihood_moments( likelihood, theta=theta )
    M1 <- res0$M
    SD1 <- res0$SD
    w1 <- rowSums(likelihood)
    # compute adjusted likelihood
    like2 <- 0*likelihood
    for (tt in 1:TP){
        like2[,tt] <- sirt_dnorm( theta[tt], mean=M1, sd=SD1*adjfac*tuningfac )
    }
    like2 <- like2 / rowSums(like2) * w1
    return(like2)
}
#############################################################
# EAP reliability
like_adj_EAP_reliability <- function( M, SD )
{
    stats::var(M) / ( stats::var(M) + mean( SD^2 ) )
}
#########################################################
# likelihood adjustment with a tuning parameter
likelihood_adjustment_tuning <- function( likelihood, theta, thetaM, adjfac,
        tuningfac, probs, maxiter=100, conv=5*1E-5, normal_approx=TRUE )
{

    like2 <- likelihood_adjustment_compute( likelihood, theta, thetaM, adjfac,
                tuningfac=tuningfac)
    change <- 1
    iter <- 0

    if (normal_approx){
        probs <- trait_normal_approx( probs, theta )
    }

    while( ( iter < maxiter ) & ( change > conv ) ){

        probsM <- matrix( probs, nrow(likelihood), ncol(likelihood), byrow=TRUE )
        post <- like2 * probsM
        post <- post / rowSums(post)
        probs_new <- colMeans( post )
        if (normal_approx){
            probs_new <- trait_normal_approx( probs_new, theta )
        }

        change <- max( abs( probs_new - probs ))
        probs <- probs_new
        iter <- iter + 1
    }

    res0 <- likelihood_moments( likelihood=like2 * probsM, theta=theta  )
    EAP.rel <- like_adj_EAP_reliability( res0$M, res0$SD )
    res <- list( "likelihood"=like2, "EAP.rel"=EAP.rel )
    return(res)
}

#*** normal approximation of trait distribution
trait_normal_approx <- function( probs, theta )
{
    M <- sum( theta*probs )
    SD <- sqrt( sum( theta^2*probs ) - M^2 )
    probs1 <- sirt_dnorm_discrete( theta, mean=M, sd=SD)
    return(probs1)
}
alexanderrobitzsch/sirt documentation built on March 18, 2024, 1:29 p.m.