R/frm_prepare_data_fb.R

In mdmb: Model Based Treatment of Missing Data

## File Name: frm_prepare_data_fb.R
## File Version: 0.20

frm_prepare_data_fb <- function(dat, dep, ind, weights0, dat0, data_init )
{
    NM <- attr(ind,"NM")
    NM1 <- NM + 1
    N <- nrow(dat)
    dat$case <- 1:N
    dat$weights0 <- weights0
    dat$weights <- 1
    dat$resp_all <- 1
    # vector of dependent variables
    dv_vars <- c()
    ind00 <- ind
    ind[[ dep$dv_vars ]] <- dep

    for (mm in 1:NM1){
        ind_mm <- ind[[mm]]
        var_mm <- ind_mm$dv_vars
        dv_vars <- c( dv_vars, var_mm)
        r_mm <- paste0("resp_", var_mm)
        dat[, r_mm] <- 1 * ( 1 - is.na( dat[,var_mm] )    )
        dat$resp_all <- dat$resp_all * dat[, r_mm ]
        # include initial values for dat[,var_mm]
        M_r_mm <- mean( dat[,r_mm] )
        N1 <- sum( dat[,r_mm] )
        N0 <- sum( 1 - dat[,r_mm] )
        y <- dat[,var_mm]
        ind1_r_mm <- which( dat[, r_mm]==1 )
        ind0_r_mm <- which( dat[, r_mm]==0 )
        if ( M_r_mm > 0 ){    # some observations
            y <- y[ ind1_r_mm ]
            y0 <- sample(y, N0, replace=TRUE )
            dat[ ind0_r_mm, var_mm] <- y0
        }
        if ( M_r_mm==0 ){    # some observations
            y0 <- data_init[, var_mm]
            if ( length(y0)==0 ){
                stop_mess <- paste0("Provide initial values for variable ",
                        var_mm, " in argument 'data_init'!\n")
                stop( stop_mess )
            }
        }
    }
    #** prepare dependent variables
    dat$weights <- dat$weights * dat$weights0

    #--- include data_init if it is not NULL
    if ( ! is.null( data_init ) ){
        cn <- colnames(data_init)
        dat[, cn ] <- data_init
    }

    dep <- ind[[ dep$dv_vars ]]
    ind[[ dep$dv_vars ]] <- NULL

    # extract matrix with response indicators
    dat_resp <- dat[, paste0( "resp_", dv_vars ) ]
    ind_resp <- as.list( dv_vars )
    names(ind_resp) <- dv_vars
    ind_miss <- ind_resp
    freq_miss_values <- rep(0,NM+1)
    names(freq_miss_values) <- dv_vars
    for (mm in 1:(NM+1) ){
        ind_resp[[mm]] <- which( dat_resp[,mm]==1)
        ind_miss[[mm]] <- which( dat_resp[,mm]==0)
        freq_miss_values[mm] <- sum( is.na( dat0[, dv_vars[mm] ] ) )
    }
    #*** variables which should be imputed
    impute_vars <- names(freq_miss_values)[ freq_miss_values > 0]
    impute_vars_index <- match( impute_vars, dv_vars )
    #--- output
    res <- list( dat=dat, dv_vars=dv_vars, dat_resp=dat_resp, ind_resp=ind_resp,
                ind_miss=ind_miss, freq_miss_values=freq_miss_values,
                impute_vars=impute_vars, impute_vars_index=impute_vars_index)
    return(res)
}