R/greenyang.reliability.R

Defines functions greenyang.reliability

Documented in greenyang.reliability

## File Name: greenyang.reliability.R
## File Version: 1.236


#---- reliability from a multidimensional nonlinear SEM for dichotomous data
greenyang.reliability <- function( object.tetra, nfactors)
{
    cat("Reliability Estimation Based on a Nonlinear SEM\n\n")
    cat("Green & Yang (2009, Psychometrika). Reliability of summed item scores\n")
    cat("  using structural equation modeling: An alternative to coefficient alpha\n\n")
    mod.omega1 <- sirt_import_psych_omega( m=object.tetra$rho, nfactors=1)

    facloadings <- matrix( mod.omega1$schmid$sl[,1], ncol=1 )
    thresh <- object.tetra$tau

    # reliability for one factor
    rel1 <- reliability.nonlinearSEM( facloadings=facloadings, thresh=thresh )$omega.rel
    # reliability for f factors
    mod.omega <- sirt_import_psych_omega( m=object.tetra$rho, nfactors=nfactors)
    rel1h <- reliability.nonlinearSEM( facloadings=facloadings, thresh=thresh )
    relf <- reliability.nonlinearSEM( facloadings=mod.omega$schmid$orthog,
                                        thresh=thresh )

    #--- calculate Omega Hierarchical Asymptotic
        pthresh <- relf$pthresh
        I <- length(pthresh)
        # create matrix of multiplied facloadings (expected correlation)
        rho.exp <- matrix( 0, nrow=I, ncol=I )
        # colnames(rho.exp) <- rownames(rho.exp) <- rownames(facloadings)
        # reliability matrix
        rel.matrix3 <- rel.matrix2 <- rel.matrix <- rho.exp
        for (ii1 in 1:I){
            for (ii2 in 1:ii1){
                rho.exp[ii1,ii2] <- rel1h$rho.exp[ii1,ii2]
                rho.exp[ii2,ii1] <- rho.exp[ii1,ii2]
                r1 <- rho.exp[ii1,ii2]
                rel.matrix[ii1,ii2] <- sirt_pmvnorm( c(-Inf,-Inf), pthresh[c(ii1,ii2)],
                                         corr=matrix( c( 1, r1, r1, 1),2,2 ) ) - stats::pnorm( pthresh[ii1] ) * pnorm( pthresh[ii2] )
                rel.matrix[ii2,ii1] <- rel.matrix[ii1,ii2]
                # multidimensional analysis
                rho.exp[ii1,ii2] <- relf$rho.exp[ii1,ii2]
                rho.exp[ii2,ii1] <- rho.exp[ii1,ii2]
                r1 <- rho.exp[ii1,ii2]
                rel.matrix2[ii1,ii2] <- sirt_pmvnorm( c(-Inf,-Inf), pthresh[c(ii1,ii2)],
                                         corr=matrix( c( 1, r1, r1, 1),2,2 ) ) -
                                         stats::pnorm( pthresh[ii1] ) * stats::pnorm( pthresh[ii2] )
                rel.matrix2[ii2,ii1] <- rel.matrix2[ii1,ii2]
                rel.matrix3[ii1,ii2] <- rel.matrix2[ii1,ii2]
                rel.matrix3[ii2,ii1] <- rel.matrix3[ii1,ii2]
                if (ii1==ii2 ){
                    r1 <- 1
                    rel.matrix3[ii1,ii2] <- sirt_pmvnorm( c(-Inf,-Inf), pthresh[c(ii1,ii2)],
                                             corr=matrix( c( 1, r1, r1, 1),2,2 ) ) -
                                                 stats::pnorm( pthresh[ii1] ) * stats::pnorm( pthresh[ii2] )
                    rel.matrix3[ii2,ii1] <- rel.matrix3[ii1,ii2]
                                        }
                                }
                        }
        # calculation of reliability
        omega.relha <- sum( rel.matrix ) / sum( rel.matrix2 )
       rel1h <- sum( rel.matrix ) / sum( rel.matrix3 )
    #'''''''''
    # eigenvalue decomposition
    eigenval.rho <- svd( object.tetra$rho )$d
    #'''''''''
#    rel1h <- rel1h$omega.rel
    relf <- relf$omega.rel
    dfr <- data.frame( "coefficient"=c( "omega_1", "omega_h", "omega_t",
                                        "omega_ha","ECV", "ExplVar", "EigenvalRatio"),
                    "dimensions"=c(1,nfactors, nfactors,nfactors,
                        nfactors, NA, NA),
                    "estimate"=c( rel1, rel1h, relf, omega.relha,
                       mod.omega$omega.lim,
                       round( 100*eigenval.rho[1]/sum(eigenval.rho),3),
                       round( eigenval.rho[1]/eigenval.rho[2],3)
                                )
                        )
     dfr <- dfr[ c(1,3,2,4,5,6,7), ]
    rownames(dfr)[1] <- c("Omega Total (1D)")
    rownames(dfr)[2] <- paste("Omega Total (",nfactors,"D)",sep="")
    rownames(dfr)[3] <- paste("Omega Hierarchical (",nfactors,"D)",sep="")
    rownames(dfr)[4] <- paste("Omega Hierarchical Asymptotic (",nfactors,"D)",sep="")
    rownames(dfr)[5] <- paste("Explained Common Variance (",nfactors,"D)",sep="")
    rownames(dfr)[6] <- "Explained Variance (First Eigenvalue)"
    rownames(dfr)[7] <- "Eigenvalue Ratio (1st to 2nd Eigenvalue)"

    dfr1 <- dfr
    dfr1[,"estimate"] <- round( dfr1[,"estimate"], 3)
    cat( paste( rep( "-", 70), collapse="") )
    cat("\n\n")
    print(dfr1)
    cat("\n\n")
    cat( paste( rep( "-", 70), collapse="") )
    cat("\n\nOutput from Hierarchical Factor Analysis (psych package)\n\n" )
    cat( paste( nfactors, "-dimensional model\n\n",sep="") )
    print(mod.omega)
    cat("\n\n")
    cat( paste( rep( ".", 45), collapse="") )
    cat( paste( "\n", 1, "-dimensional model\n\n",sep="") )
    mod.omega1 <- sirt_import_psych_omega( m=object.tetra$rho, nfactors=1)
    print(mod.omega1)
    invisible(dfr1)
}
alexanderrobitzsch/sirt documentation built on March 18, 2024, 1:29 p.m.