Linear Regression

Computes linear regression.

BIFIE.linreg(BIFIEobj, dep=NULL, pre=NULL, formula=NULL,
    group=NULL, group_values=NULL, se=TRUE)

# S3 method for BIFIE.linreg
summary(object,digits=4,...)

# S3 method for BIFIE.linreg
coef(object,...)

# S3 method for BIFIE.linreg
vcov(object,...)

Arguments

BIFIEobj: Object of class BIFIEdata
dep: String for the dependent variable in the regression model
pre: Vector of predictor variables. If the intercept should be included, then use the variable one for specifying it (see Examples).
formula: An R formula object which can be applied instead of providing dep and pre. Note that there is additional computation time needed for model matrix creation.
group: Optional grouping variable(s)
group_values: Optional vector of grouping values. This can be omitted and grouping values will be determined automatically.
se: Optional logical indicating whether statistical inference based on replication should be employed.
object: Object of class BIFIE.linreg
digits: Number of digits for rounding output
...: Further arguments to be passed

Value

A list with following entries

stat: Data frame with unstandardized and standardized regression coefficients, residual standard deviation and \(R^2\)
output: Extensive output with all replicated statistics
...: More values

Examples

#############################################################################
# EXAMPLE 1: Imputed TIMSS dataset
#############################################################################

data(data.timss1)
data(data.timssrep)

# create BIFIE.dat object
bdat <- BIFIEsurvey::BIFIE.data( data.list=data.timss1, wgt=data.timss1[[1]]$TOTWGT,
             wgtrep=data.timssrep[, -1 ] )

#**** Model 1: Linear regression for mathematics score
mod1 <- BIFIEsurvey::BIFIE.linreg( bdat, dep="ASMMAT", pre=c("one","books","migrant"),
              group="female" )
summary(mod1)

if (FALSE) {
# same model but specified with R formulas
mod1a <- BIFIEsurvey::BIFIE.linreg( bdat, formula=ASMMAT ~ books + migrant,
               group="female", group_values=0:1 )
summary(mod1a)

# compare result with lm function and first imputed dataset
dat1 <- data.timss1[[1]]
mod1b <- stats::lm( ASMMAT ~ 0 + as.factor(female) + as.factor(female):books +
                              as.factor(female):migrant,
                         data=dat1,  weights=dat1$TOTWGT )
summary(mod1b)

#**** Model 2: Like Model 1, but books is now treated as a factor
mod2 <- BIFIEsurvey::BIFIE.linreg( bdat, formula=ASMMAT ~ as.factor(books) + migrant)
summary(mod2)

#############################################################################
# EXAMPLE 2: PISA data | Nonlinear regression models
#############################################################################

data(data.pisaNLD)
data <- data.pisaNLD

#--- Create BIFIEdata object immediately using BIFIE.data.jack function
bdat <- BIFIEsurvey::BIFIE.data.jack( data.pisaNLD, jktype="RW_PISA", cdata=TRUE)
summary(bdat)

#****************************************************
#*** Model 1: linear regression
mod1 <- BIFIEsurvey::BIFIE.linreg( bdat, formula=MATH ~ HISEI )
summary(mod1)

#****************************************************
#*** Model 2: Cubic regression
mod2 <- BIFIEsurvey::BIFIE.linreg( bdat, formula=MATH ~ HISEI + I(HISEI^2) + I(HISEI^3) )
summary(mod2)

#****************************************************
#*** Model 3: B-spline regression

# test with design of HISEI values
dfr <- data.frame("HISEI"=16:90 )
des <- stats::model.frame( ~ splines::bs( HISEI, df=5 ), dfr )
des <- des$splines
plot( dfr$HISEI, des[,1], type="l", pch=1, lwd=2, ylim=c(0,1) )
for (vv in 2:ncol(des) ){
    lines( dfr$HISEI, des[,vv], lty=vv, col=vv, lwd=2)
}

# apply B-spline regression in BIFIEsurvey::BIFIE.linreg
mod3 <- BIFIEsurvey::BIFIE.linreg( bdat, formula=MATH ~ splines::bs(HISEI,df=5) )
summary(mod3)

#*** include transformed HISEI values for B-spline matrix in bdat
bdat2 <- BIFIEsurvey::BIFIE.data.transform( bdat, ~ 0 + splines::bs( HISEI, df=5 ))
bdat2$varnames[ bdat2$varsindex.added ] <- paste0("HISEI_bsdes",
            seq( 1, length( bdat2$varsindex.added ) ) )

#****************************************************
#*** Model 4: Nonparametric regression using BIFIE.by

?BIFIE.by

#---- (1) test function with one dataset
dat1 <- bdat$dat1
vars <- c("MATH", "HISEI")
X <- dat1[,vars]
w <- bdat$wgt
X <- as.data.frame(X)
# estimate model
mod <- stats::loess( MATH ~ HISEI, weights=w, data=X )

# predict HISEI values
hisei_val <- data.frame( "HISEI"=seq(16,90) )
y_pred <- stats::predict( mod, hisei_val )
graphics::plot( hisei_val$HISEI, y_pred, type="l")

#--- (2) define loess function
loess_fct <- function(X,w){
    X1 <- data.frame( X, w )
    colnames(X1) <- c( vars, "wgt")
    X1 <- stats::na.omit(X1)
#    mod <- stats::lm( MATH ~ HISEI, weights=X1$wgt, data=X1 )
    mod <- stats::loess( MATH ~ HISEI, weights=X1$wgt, data=X1 )
    y_pred <- stats::predict( mod, hisei_val )
    return(y_pred)
}

#--- (3) estimate model
mod4 <-  BIFIEsurvey::BIFIE.by( bdat, vars, userfct=loess_fct )
summary(mod4)

# plot linear function pointwise and confidence intervals
graphics::plot( hisei_val$HISEI, mod4$stat$est, type="l", lwd=2,
        xlab="HISEI", ylab="PVMATH", ylim=c(430,670) )
graphics::lines( hisei_val$HISEI, mod4$stat$est - 1.96* mod4$stat$SE, lty=3 )
graphics::lines( hisei_val$HISEI, mod4$stat$est + 1.96* mod4$stat$SE, lty=3 )
}

Arguments

Value

See also

Examples