Skip to contents

Open One Side of a Confidence Interval

Source: R/applySides.R
applySides.Rd

Clamps a confidence interval to the range of the parameter and opens the side that a one-sided interval leaves free. One implementation for the whole suite, so that every function reports a one-sided bound the same way.

Usage

applySides(ci, sides = "two.sided", lo = -Inf, hi = Inf)

Arguments

ci

numeric vector of length two, the lower and upper bound in that order. NA bounds are passed through, and c(NA, NA) is accepted although it is logical rather than numeric - that is how an interval which could not be computed is usually written.

sides

character string, one of "two.sided" (default), "left" or "right". It names the side carrying the finite bound, so "left" corresponds to alternative = "greater" in a test. Callers are expected to have resolved the value with match.arg() already; an unmatched value is an error rather than a partial match.

lo, hi

the range of the parameter, not infinities by default in spirit but in signature. See Details.

Value

a named numeric vector with the elements lci and uci.

Details

sides names the side carrying the finite bound:

"left"

the informative bound is the lower one; the upper one is opened to hi.

"right"

the informative bound is the upper one; the lower one is opened to lo.

lo and hi are the parameter's range, not infinities. Most statistics are bounded, so reporting the open side at the boundary is the ordinary case rather than an exception: a correlation opens to \(\pm 1\), an association measure in \([0, 1]\) to 0 or 1, Pearson's \(C\) to \(\sqrt{(m-1)/m}\). Where the parameter really is unbounded, \(\pm\)Inf is passed and the usual half-line comes back. Some statistics need one of each: Cronbach's alpha takes lo = -Inf and hi = 1, a relative risk lo = 0 and hi = Inf.

The two-sided interval is clamped to \([lo, hi]\) as well, so an interval can never claim a value the statistic cannot take.

Why this is not written out per function

Five hand-written copies of the same three lines produced four different defects across one review: two functions had the sides inverted, one ignored them after adjusting the level, and one returned NA where a boundary belonged. The operation is short enough to retype and just subtle enough to retype wrongly.

See also

checkConfLevel(), checkFlag()

Examples

ci <- c(0.12, 0.58)

applySides(ci, "two.sided", lo = 0, hi = 1)
#>  lci  uci 
#> 0.12 0.58 
applySides(ci, "left",      lo = 0, hi = 1)   # uci opens to 1
#>  lci  uci 
#> 0.12 1.00 
applySides(ci, "right",     lo = 0, hi = 1)   # lci opens to 0
#>  lci  uci 
#> 0.00 0.58 

# an unbounded parameter opens to infinity
applySides(c(-1.4, 2.6), "left", lo = -Inf, hi = Inf)
#>  lci  uci 
#> -1.4  Inf 

# and one of each: Cronbach's alpha is bounded above only
applySides(c(0.61, 0.94), "right", lo = -Inf, hi = 1)
#>  lci  uci 
#> -Inf 0.94 

# the two-sided interval is clamped too
applySides(c(-0.2, 1.3), "two.sided", lo = 0, hi = 1)
#> lci uci 
#>   0   1 

# NA bounds survive
applySides(c(NA, NA), "left", lo = -1, hi = 1)
#> lci uci 
#>  NA   1