R/R2noharm.R

Defines functions R2noharm

Documented in R2noharm

## File Name: R2noharm.R
## File Version: 2.24

#------------------------------------------------------------------
# NOHARM exploratory factor analysis
R2noharm <- function( dat=NULL, pm=NULL, n=NULL,model.type, weights=NULL,
            dimensions=NULL, guesses=NULL, noharm.path,
                            F.pattern=NULL, F.init=NULL, P.pattern=NULL, P.init=NULL,
                            digits.pm=4, writename=NULL,
                            display.fit=5, dec=".", display=TRUE  ){

    # INPUT:
    # dat ... dataframe
    # model.type ... CFA or EFA
    # dimensions ... dimensions for exploratory factor analysis
    # guesses ... fixed guessing paramters
    # noharm.path ... path for NOHARM console (console must have the name NOHARM87.exe)
    # digits.pm ... digits for calculation of product moment matrix
    # writename ... name of NOHARM Output file
    # dec ... "." or "," for decimal separator
    # display ... display noharm settings (TRUE or FALSE)
    #............................................................................
    # The matrices F.pattern, F.init, P.pattern and P.init correspond to the
    # definition in the NOHARM manual
#    noharm.path <- shQuote(noharm.path)
    if (model.type=="CFA" ){ dimensions <- ncol(F.pattern) }
    if ( display ){
        cat("Multidimensional Normal Ogive by Harmonic Analysis (NOHARM 4) \n" )
        cat("C. Fraser & R. P. McDonald (2012)\n" )
        cat("For more informations please look at \n ")
#        cat( "    http://people.niagaracollege.ca/cfraser/download/nhCLwindl.html\n\n")
        cat( "  http://noharm.niagararesearch.ca/nh4cldl.html\n\n")
        cat( paste( "Path of NOHARMCL file: \n  ** ", noharm.path, "\n" ) )
        cat( paste( "Path of NOHARM input and output files: \n  ** ", getwd(), "\n\n" ) )
        if (model.type=="EFA"){
                cat(paste( "Exploratory Item Factor Analysis with", dimensions, "Dimensions" ), "\n\n" )
                } else {             cat("Confirmatory Item Factor Analysis \n\n" ) }
        flush.console()
                }
    ########################################################
    # data input
    if ( ! is.null(dat) ){
        # allow also for input of moment matrix data
        dat <- as.matrix(dat)
        I <- ncol(dat)     # number of items
        n <- nrow(dat)     # number of subjects
        if ( is.null(weights) ){
            weights <- rep(1,n) } else {
            weights <- weights / sum(weights) * n
                }

        # calculate raw product moment matrix
        dat9 <- dat
        dat.resp <- is.na( dat)
        dat9[ dat.resp ] <- 9
        # pairwise product moment matrix
    #    BM <- t( dat9 * (1- dat.resp ) ) %*% ( dat9 * (1- dat.resp  ) )
         BM <- crossprod( dat9 * (1- dat.resp  )*weights  )
        # number of persons on an item pair
    #    NM <- t((1- dat.resp ) ) %*% (  (1- dat.resp  ) )
        NM <- crossprod(  (1- dat.resp)*weights  )
        BM <- round( BM/NM, digits.pm )
                    }
    inputdat <- TRUE
    ########################################################
    if ( ! is.null(pm) ){
      if ( is.vector(pm) ){
        I2 <- length(pm)
        I <- sqrt( 2*I2+1/4 ) - .5
        BM <- matrix( 0, I, I )
        colnames(BM) <- paste0("I",1:I)
        vv <- 0
        for (ii in 1:I){
#        ii <- 1
            BM[ ii, 1:ii ] <- pm[ seq(vv+1,vv+ii) ]
            vv <- vv + ii
                    }
        BM <-  BM + t(BM)
        diag(BM) <- diag(BM)/2
             } else {
                pm <- BM
                I <- ncol(pm)
                    }
        weights <- rep(1,n)
        NM <- matrix(n, I,I)
        dat <- BM[1:2,,drop=FALSE]
        inputdat <- FALSE
                        }
    # Exploratory analysis requested?
    EX <- 1 * ( model.type=="EFA" )
    # supply starting values
    IV <- 1 * (model.type=="CFA" )

    if ( is.null(guesses) ){
        guesses <- rep(0, I )
                    }

    # arrange NOHARM Input Files
    s1 <- Sys.time()
    noharm.input <- c( paste( "R2noharm Input file", s1 ),
                        paste( I, dimensions, n, 1,  EX, IV, 0, 0,  sep=" ")
                                        )
    # add guessing parameter
    noharm.input <- c( noharm.input, "   ", guesses, "  " )

    if (model.type=="CFA"){
        # pattern matrices
        noharm.input <- c( noharm.input,
            apply( F.pattern, 1, FUN=function(ll){ paste( ll, collapse=" " ) } ), " ",
            sapply( 1:dimensions, FUN=function(ss){ paste( P.pattern[ ss, 1:ss ], collapse=" " ) }  ),
            "   "     )
        # add initial matrices
        if ( is.null(F.init) ){ F.init <- .5*F.pattern }
        if ( is.null(P.init) ){ P.init <- 0.1  + diag( .9, dimensions) }
        noharm.input <- c( noharm.input,
                apply( F.init, 1, FUN=function(ll){ paste( ll, collapse=" " ) } ), " ",
                sapply( 1:dimensions, FUN=function(ss){ paste( P.init[ ss, 1:ss ], collapse=" " ) }  )
                        )
         }

        # product moment matrix
        pm <- unlist( sapply( 1:I, FUN=function( ii){  BM[ ii, 1:ii ]  } ) )
        LPM <- length(pm)
        h1 <- seq( 1, LPM, 10 )
        h2 <- c( seq( 10, LPM, 10 ) )
        if (length(h1) !=length(h2) ){ h2 <- c( h2, LPM ) }
        h3 <- data.frame( h1, h2 )
        pm <- apply( h3, 1, FUN=function(ll){ paste( pm[ ll[1]:ll[2] ], collapse=" " ) } )


    # add product moment matrix to NOHARM input
    noharm.input <- c( noharm.input, " ", pm   )
    if (dec=="," ){ noharm.input <- gsub( "\\.", ",", noharm.input ) }
    # path specifications and starting NOHARM
    current.path <- getwd()
    setwd( noharm.path )
    writeLines( noharm.input, "mymodel.inp" )
#    writeLines( "NoharmCL mymodel.inp mymodel.out 800 0.00001", "noharm_analysis.bat" )
    writeLines( "NoharmCL mymodel.inp mymodel.out", "noharm_analysis.bat" )
    # working without DOSBox
    system( "noharm_analysis.bat", show.output.on.console=F  )
    # read NOHARM output
    noharmout0 <- readLines( "MYMODEL.out" )
    if (dec=="," ){ noharmout0 <- gsub( ",", "\\.", noharmout0 ) }
    noharmout1 <- noharmout0
    noharmout1 <- c( noharmout1, rep("",3), "ENDE" )
    # set current working directory
    setwd( current.path )
    if (!is.null( writename ) ){
                writeLines( noharmout0, paste( writename, ".out", sep="") )
                writeLines( noharm.input, paste( writename, ".inp", sep="") )
                }
    # results for 1-dimensional IRT analysis
    if (dimensions==1 & model.type=="EFA" ){
            modtype <- 1
        res <- list( "tanaka"=.noharm.tanaka( noharmout1 ),
            "rmsr"=.noharm.rmsr( noharmout1 ),
            "N.itempair"=NM,
            "pm"=BM,
            "weights"=weights,
            "guesses"=guesses,
            "residuals"=.noharm.residuals( noharmout1, I=I, dat=dat),
            "final.constants"=.noharm.itemlevel( noharmout1, "Final Constants", I=I, dat=dat),
            "thresholds"=.noharm.itemlevel( noharmout1, "Threshold Values", I=I, dat=dat),
            "uniquenesses"=.noharm.itemlevel( noharmout1, "Unique Variances", I=I, dat=dat),
            "difficulties"=.noharm.itemlevel( noharmout1, "Vector B", I=I, dat=dat),
            "discriminations"=.noharm.itemlevel( noharmout1, "Vector A", I=I, dat=dat)
                )  }
    # results for noharm.efa with more than one dimension
    if (dimensions > 1 & model.type=="EFA" ){
            modtype <- 2
        res <- list( "tanaka"=.noharm.tanaka( noharmout1 ),
                "rmsr"=.noharm.rmsr( noharmout1 ),
                "N.itempair"=NM,
                "pm"=BM,
                "guesses"=guesses,
                "residuals"=.noharm.residuals( noharmout1, I=I, dat=dat),
                "final.constants"=.noharm.itemlevel( noharmout1, "Final Constants",I=I, dat=dat),
                "factor.cor"=.noharm.correlations( noharmout1, "Factor Correlations",
                                "Promax Rotated", dimensions=dimensions, dat=dat),
                "thresholds"=.noharm.itemlevel( noharmout1, "Threshold Values", I=I, dat=dat),
                "uniquenesses"=.noharm.itemlevel( noharmout1, "Unique Variances", I=I, dat=dat),
                "unrotated"=.noharm.loadings( noharmout1, "Factor Loadings",
                                "Varimax Rotated Factor Loadings",  dimensions=dimensions, I=I, dat=dat),
                "varimax"=.noharm.loadings( noharmout1, "Varimax Rotated Factor Loadings",
                                "Varimax Rotated Coefficients of Theta",   dimensions=dimensions, I=I, dat=dat),
                "varimax.theta"=.noharm.loadings( noharmout1, "Varimax Rotated Coefficients of Theta",
                                    "oblique",   dimensions=dimensions, I=I, dat=dat),
                "promax"=.noharm.loadings( noharmout1, "oblique",
                                "Factor Correlations",   dimensions=dimensions, I=I, dat=dat),
                "promax.theta"=.noharm.loadings( noharmout1, "Promax Rotated Coefficients of Theta",
                                        "ENDE",   dimensions=dimensions, I=I, dat=dat)
                )
            }

    # results for noharm.cfa with more than one dimension
    if (dimensions > 1 & model.type=="CFA" ){
            modtype <- 3
        res <- list( "tanaka"=.noharm.tanaka( noharmout1 ),
                "rmsr"=.noharm.rmsr( noharmout1 ),
                "N.itempair"=NM,
                "pm"=BM,
                "guesses"=guesses,
                "residuals"=.noharm.residuals( noharmout1, I=I, dat=dat),
                "final.constants"=.noharm.itemlevel( noharmout1, "Final Constants",I=I, dat=dat),
                "factor.cor"=.noharm.correlations( noharmout1, "Final Correlations", "Residual", dimensions=dimensions, dat=dat),
                "thresholds"=.noharm.itemlevel( noharmout1, "Threshold Values", I=I, dat=dat),
                "uniquenesses"=.noharm.itemlevel( noharmout1, "Unique Variances", I=I, dat=dat),
                "loadings"=.noharm.loadings( noharmout1, "Factor Loadings",
                                "ENDE",  dimensions=dimensions, I=I, dat=dat),
                "loadings.theta"=.noharm.loadings( noharmout1, "Final Coefficients of Theta",
                                "Final Correlations of Theta",   dimensions=dimensions, I=I, dat=dat)
                    )


            if( !is.null( colnames(F.pattern) ) ){
                    colnames(res$loadings) <- colnames(F.pattern)
                    colnames(res$loadings.theta) <- colnames(F.pattern)
                    colnames(res$factor.cor) <- rownames(res$factor.cor) <- colnames(F.pattern)
                                }
                 }
    # CFA 1 dimension
    if ( ( dimensions==1 ) & ( model.type=="CFA" ) ){
                    modtype <- 4
            if ( is.null(colnames(F.pattern) ) ){ colnames(F.pattern) <- "F1" }
        res <- list( "tanaka"=.noharm.tanaka( noharmout1 ),
                "rmsr"=.noharm.rmsr( noharmout1 ),
                "N.itempair"=NM,
                "pm"=BM,
                "guesses"=guesses,
                "residuals"=.noharm.residuals( noharmout1, I=I, dat=dat),
                "final.constants"=.noharm.itemlevel( noharmout1, "Final Constants",I=I, dat=dat),
                "thresholds"=.noharm.itemlevel( noharmout1, "Threshold Values", I=I, dat=dat),
                "uniquenesses"=.noharm.itemlevel( noharmout1, "Unique Variances", I=I, dat=dat),
                "loadings.theta"=.noharm.loadings( noharmout1, "Final Coefficients of Theta",
                                "Residual",   dimensions=dimensions, I=I, dat=dat),
                "factor.cor"=as.data.frame(matrix(1,1, 1 )),
                "difficulties"=.noharm.itemlevel( noharmout1, "Vector B", I=I, dat=dat),
                "discriminations"=.noharm.itemlevel( noharmout1, "Vector A", I=I, dat=dat),
                "loadings"=.noharm.loadings( noharmout1, "Factor Loadings",
                                "LORD",   dimensions=dimensions, I=I, dat=dat)
                #                ,
#                "loadings"=.noharm.loadings( noharmout1, "Factor Loadings",
#                                "LORD",  dimensions=dimensions, I=I, dat=dat)
                )
            if( !is.null( colnames(F.pattern) ) ){
                    colnames(res$loadings) <- colnames(F.pattern)
                    rownames(res$factor.cor) <- colnames(res$factor.cor) <- colnames(F.pattern)
                    colnames(res$loadings.theta) <- colnames(F.pattern)
                              } }
    # collect arguments in output
    res$model.type <- model.type
#    res$Nobs <- nrow(dat)
#    res$Nitems <- ncol(dat)
    res$Nobs <- n
    res$Nitems <- I
    res$modtype <- modtype
    res$F.init <- F.init
    res$F.pattern <- F.pattern
    res$P.init <- P.init
    res$P.pattern <- P.pattern
    res$dat <- dat
    res$systime <- s1
    res$guesses <- guesses
    res$noharm.path <- noharm.path
    res$digits.pm <- digits.pm
    res$dec <- dec
    res$display.fit <- display.fit
    if ( modtype %in% c(1,4)){ res$dimensions <- 1 }
    if ( modtype %in% c(2,3)){ res$dimensions <- ncol(res$factor.cor) }
    #***
    # calculate fit statistic
    if ( modtype %in% 2:4){
        RM <- res$residuals
        PV <- diag( res$pm )
        g1 <- sqrt( outer( PV * (1-PV), PV * ( 1-PV ) ) )
        rM <- ( RM / g1  )
        zM <- 0.5 * log( 1 + rM ) - 0.5 * log( 1 - rM )
        # chi square
        res$chisquare <- X2 <- ( res$Nobs - 3 ) * sum( zM^2 )
        # calculate number of estimated parameters and degrees of freedom
        I <- res$Nitems
        if (modtype %in% 3:4){
            Nestpars <- I + sum( F.pattern==1 ) + length( unique( intersect( F.pattern, 2:9999 ) ) )
            Nestpars <- Nestpars + sum( diag(P.pattern)==1 ) +
                            length( unique( diag(P.pattern)[ diag(P.pattern) > 1] ) )
            g2 <- P.pattern[ lower.tri(P.pattern) ]
            Nestpars <- Nestpars + sum( g2==1 ) + length( unique( intersect( g2, 2:9999 ) ))
            res$Nestpars <- Nestpars
                            }
        if ( modtype %in% 2){
            cs <- 1:(dimensions-1)
            res$Nestpars <- Nestpars <- I + I*dimensions - sum(cs)
                    }
        res$df <- df <- 0.5*I*(I+1) - Nestpars
        res$chisquare_df  <- res$chisquare / res$df
        # calculate RMSEA
        res$rmsea <- rmsea <- sqrt(max(c( (X2 / res$Nobs ) / df - 1/res$Nobs, 0)))
        # calculate p values
        res$p.chisquare <- 1 - pchisq( res$chisquare, df=res$df )
            }
    # display
    if (display){
        cat( paste( "Tanaka Index=", round(res$tanaka,display.fit), sep=""), "\n" )
        cat( paste( "RMSR=", round(res$rmsr,display.fit), sep=""), "\n" )
        if ( ! inputdat ){
           cat("\n**** Note that Jackknife does not work for pm input! **** \n")
                }
            }
    if ( ! inputdat ){
           res$dat <- NULL
            }
    res$inputdat <- inputdat
    res$upper <- 1+0*res$guess
    res$lower <- res$guess
    class(res) <- "R2noharm"
    return( res )
    }
#----------------------------------------------------------
alexanderrobitzsch/sirt documentation built on March 18, 2024, 1:29 p.m.