R/BIFIE.mva.R

In BIFIEsurvey: Tools for Survey Statistics in Educational Assessment

## File Name: BIFIE.mva.R
## File Version: 0.331


#######################################################################
# Missing value analysis
BIFIE.mva <- function( BIFIEobj, missvars, covariates=NULL, se=TRUE )
{
    s1 <- Sys.time()
    bifieobj <- BIFIEobj
    cl <- match.call()
    if ( ! bifieobj$cdata ){
        varnames <- unique( c(missvars, covariates ) )
        bifieobj <- BIFIE.BIFIEdata2BIFIEcdata( bifieobj, varnames )
    }

    if ( is.null(covariates) ){
        covariates <- "one"
    }


    if ( is.null(covariates) ){
        N <- bifieobj$N
        transform.formula <-  paste0( "~ 0 + I ( stats::runif( ", N, ", 0, 1E-10) ) " )
        bifieobj <- BIFIE.data.transform( bifieobj, transform.formula,  "_null" )
        covariates <- bifieobj$varnames.added
        se <- FALSE
    }

    FF <- Nimp <- bifieobj$Nimp
    N <- bifieobj$N
    dat1 <- bifieobj$dat1
    wgt <- bifieobj$wgt
    wgtrep <- bifieobj$wgtrep
    varnames <- bifieobj$varnames
    RR <- bifieobj$RR
    datalistM <- bifieobj$datalistM
    fayfac <- bifieobj$fayfac

    # start with a compact BIFIEdata object
    varnames <- unique( c(missvars, covariates ) )

    # assume that bifieobj is already in cdata format
    # define selected response indicators
    missvars_index <- match( missvars, colnames(bifieobj$dat1) )
    datalistM_ind_sel <- bifieobj$datalistM_ind[, missvars_index, drop=FALSE ]

    respvars <- paste0("resp_", missvars )
    colnames(datalistM_ind_sel) <- respvars
    VVadd <- length(respvars )
    varnames1 <- c( varnames, respvars )

    bifieobj$datalistM_ind <- cbind( bifieobj$datalistM_ind, datalistM_ind_sel )
    bifieobj$dat1 <- cbind( bifieobj$dat1, datalistM_ind_sel )
    bifieobj$Nvars <- bifieobj$Nvars + VVadd
    bifieobj$varnames <- c( bifieobj$varnames, respvars )
    # select dataset
    bifieobj <- BIFIE.BIFIEcdata2BIFIEdata( bifieobj, varnames=varnames1 )

    RR <- bifieobj$RR
    do_test <- TRUE
    if (RR < 2 ){
        wgtrep <- bifieobj$wgtrep
        do_test <- FALSE
        wgtrep <- cbind( wgtrep, wgtrep + stats::runif( nrow(wgtrep), 0, 1E-4) )
        bifieobj$wgtrep <- wgtrep
        bifieobj$RR <- 2
    }

    if ( is.null( covariates) ){
        do_test <- FALSE
        se <- FALSE
        covariates <- "one"
    }

    if (RR==1){ RR <- 0 }
    if ( ! se ){
        N11 <- length(wgt)
        wgtrep <- matrix( NA, nrow=N11, ncol=2 )
        wgtrep[,1] <-wgt
        eps <- 1E-8
        wgtrep[,2] <- wgt + stats::runif(N11, -eps,eps)
        RR <- 0
        bifieobj$wgtrep <- wgtrep
        bifieobj$RR <- ncol(wgtrep)
    }

    #*****
    # collect results
    VV <- length(respvars)
    CVV <- length(covariates)
    res_list <- list(1:VV)
    dfr <- NULL
    RR <- bifieobj$RR

    for (vv in 1:VV){
        rvv <- respvars[vv]
        res.vv <- as.list(1:2)
        names(res.vv) <- c("stat", "dstat")
        res <- BIFIE.univar( bifieobj, vars=covariates, group=rvv,
                    group_values=0:1 )
        res.vv$stat <- res$stat
        res1 <- BIFIE.univar.test( res, wald_test=FALSE )
        res.vv$dstat <- res1$stat.dstat
        res_list[[vv]] <- res.vv
        # collect results
        dfr.vv <- data.frame( "respvar"=rep(rvv,CVV) )
        dfr.vv$missprop <- res$stat$Nweight[1] / ( res$stat$Nweight[1] + res$stat$Nweight[2] )
        dfr.vv$covariate <- covariates
        if ( do_test ){
            dfr.vv$d <- res1$stat.dstat$d
            dfr.vv$d_SE <- res1$stat.dstat$d_SE
            dfr.vv$t <- res1$stat.dstat$t
            dfr.vv$p <- res1$stat.dstat$p
        }
        dfr.vv$M_resp <- res$stat$M[ seq(2,2*CVV, 2 ) ]
        dfr.vv$M_miss <- res$stat$M[ seq(1,2*CVV, 2 ) ]
        dfr.vv$SD_resp <- res$stat$SD[ seq(2,2*CVV, 2 ) ]
        dfr.vv$SD_miss <- res$stat$SD[ seq(1,2*CVV, 2 ) ]
        dfr <- rbind( dfr, dfr.vv )
    }
    if ( covariates[1]=="_null" ){ se <- FALSE }

    if ( ! do_test ){
        RR <- 0
    }

    if ( ( ! se ) |  ( RR==0 ) ){
        dfr$t <- dfr$p <- dfr$d_SE <-  NULL
    }

    if ( covariates[1] %in% c("_null","one") ){
        dfr$covariate <- dfr$d <- dfr$M_resp  <- NULL
        dfr$M_miss <- dfr$SD_resp <- dfr$SD_miss <- NULL
    }

    #***** OUTPUT
    s2 <- Sys.time()
    timediff <- c( s1, s2 ) #, paste(s2-s1 ) )
    res1 <- list( "stat.mva"=dfr, "res_list"=res_list,
            "timediff"=timediff,
            "N"=N, "Nimp"=Nimp, "RR"=RR, "fayfac"=fayfac,
            "NMI"=BIFIEobj$NMI, "Nimp_NMI"=BIFIEobj$Nimp_NMI,
            "CALL"=cl )
    class(res1) <- "BIFIE.mva"
    return(res1)
        }
###################################################################################

####################################################################################
# summary for BIFIE.mva function
summary.BIFIE.mva <- function( object, digits=4, ... )
{
    BIFIE.summary(object)
    cat("Missing Value Analysis \n")
    obji <- object$stat.mva
    print_object_summary( obji, digits=digits )
}