https://github.com/cran/coin
Tip revision: 0bbdd83f8d1d7537dda36e896e3390cd68a33eb7 authored by Torsten Hothorn on 21 December 2010, 00:00:00 UTC
version 1.0-18
version 1.0-18
Tip revision: 0bbdd83
ToDoList
o formula interface to wilcoxsign_test is brain damaged
y ~ 1 must be OK
bug:
y<-c(1.11,2.22,3.33,0,100,1)
x<-as.factor(c(1,1,0,0,1,0))
foo <- function(f)
.Call("R_split_up_2sample", f * y, as.integer(3), sum(f * y[x == "0"]), PACKAGE = "coin")
sapply(1:100, foo)
o implement `pperm(x, q, lower.tail = TRUE)'
o more checks on direct `ytrafo = foo' constructs
o improve error messages
o add checks on achieved alpha in confint (see wilcox.test changes by
Thomas L)
o some classes are the same in `coin' and `party': can they conflict?
o maxstat_tests scales poorly for very large problems (number of obs)
maybe we can look at each 2-sample problem separately?
- unify min/maxprop arguments to something like `quantiles = c(...)'
(either of length 2 or arbitrary)
o compare multiple test procedures with multtest and add more checks,
compare with Peter Westfalls example
[done]
o check spearman_test vs. cor.test
o definition of two-sided p-values in ansari.test contradicts the one
used here in some cases (-> regtest_2samples.R)
o confint() after contrast_test: how to check if ranks were used???
labels of `contrMat' and direction of differences don't match
o check storage.mode of all vars going into C code (MC)
o add wilcoxon-gehan scores, check logrank scores!
o check if trend tests for clustered data are part of the framework
o maximally selected McNemar statistics -> Betensky, Biometrics, 2000
o visualize deviations from H_0 for all problems (in a way
motivated by shaded mosaicplots)
o mh_test with weights
o free W1 and W2 in `vandeWiel.c' in case of an error
o enable confidence intervals in the presence of blocks or weights
o check whats happening here
d <- data.frame(y = rnorm(20), x = gl(2, 10))
d$y[d$x == "1"] <- d$y[d$x == "1"] * 2
wt <- ansari_test(y ~ x, data = d, di = "ex", conf.int = TRUE)
pvalue(wt)
confint(wt, level = 1 - pvalue(wt))
o Kurt: trafo as family of transformations (see glm(..., family,...)
o should we have `distribution = "none"'?
o some trafos need to take weights into account -> ranks?
o permutation p-value = 0 means pval < 1 / B, not < 1e-16 (as
format.pval reports)