Square the signed interval as a set, with lower limit zero if it contains zero. This differs from an interval of squared draws. No draws are generated and no endpoints are clipped to one. Failed signed endpoints remain missing.
Arguments
- object
Context accepted by
SCorNB().- level
Single confidence level in (0, 1).
- type
Signed bootstrap interval: pc, bc, or bca.
Value
Data frame with est, lower and upper columns and predictor row names. The point estimate is the squared signed estimate. No standard error is implied by this interval transformation.
Details
The semipartial correlation \(sr_{j}\) (see SCorNB()) is signed,
and its confidence interval \(\left[ l, u \right]\) may contain zero.
The interval for \(sr^{2}_{j}\) is the image of that interval under
squaring.
The squared interval is
\(\left[ l^{2}, u^{2} \right]\) if \(0 < l\),
\(\left[ u^{2}, l^{2} \right]\) if \(u < 0\),
and \(\left[ 0, \max \left( l^{2}, u^{2} \right) \right]\)
if the signed interval contains zero.
This is not the interval of squared draws, and it is not an interval for
DeltaRSqNB().
Examples
# Data ---------------------------------------------------------------------
data("nas1982", package = "betaNB")
# Fit Model in lm ----------------------------------------------------------
object <- stats::lm(
formula = QUALITY ~ NARTIC + PCTGRT + PCTSUPP,
data = nas1982
)
# NB -----------------------------------------------------------------------
nb <- NB(
object = object,
R = 100, # use a large value e.g., 5000L for actual research
seed = 0508
)
# SCorSqNB -----------------------------------------------------------------
SCorSqNB(object = nb, level = 0.95, type = "pc")
#> est lower upper
#> NARTIC 0.1858925 0.079616147 0.2940437
#> PCTGRT 0.1176542 0.037356988 0.2216212
#> PCTSUPP 0.0568722 0.008838188 0.1394616