Adapt a cer design by dropping hypotheses, with a simplified strategy for redistributing weights

The weights of the dropped hypotheses are set to 0 and distributed proportionally to the previous weight to the other hypotheses. This means the weights "stay the same", they are only adapted to still sum up to 1. Note that this method does not lead to an adapted graph that is coherent with the actual weight distribution, this may lead to problems down the line

Usage

cer_alt_drop_hypotheses(design, hypotheses, adapt_bounds = TRUE)

Arguments

design

cer_design object

hypotheses

vector of booleans indicating for each hypotheses if it should be dropped

adapt_bounds

Adapt the bounds for rejecting a hypotheses to keep the FWER with the new adaptions. If doing multiple adaptions, it is enough to adapt bounds only for the last one, or call adapt_bounds() manually after.

Value

design with specified hypotheses dropped (so TRUE means the hypothesis is dropped)

Examples

as <- function(x,t) 2-2*pnorm(qnorm(1-x/2)/sqrt(t))
design <- cer_design(
 correlation=rbind(H1=c(1, NA),
                   H2=c(NA, 1)),
 weights=c(2/3, 1/3),
 alpha=0.05,
 test_m=rbind(c(0, 1),
              c(1, 0)),
 alpha_spending_f=as,
 t=0.5)

design <- cer_interim_test(design, c(0.1, 0.02))

design <- cer_alt_drop_hypotheses(design, c(TRUE, FALSE))
design
#> A CER Design object, for testing 2 hypotheses at FWER 0.05.
#> 
#> ── An interim test has been performed. ─────────────────────────────────────────
#> No Hypotheses were rejected at the interim.
#> ── The following characteristics have been adapted: ────────────────────────────
#> • Hypotheses weights
#> ── No final test has been performed yet ────────────────────────────────────────