censored.fit {fdrtool} | R Documentation |
censored.fit
fits the scale parameter of a null distribution with mean zero
to censored data.
censored.fit(x, statistic=c("normal", "correlation", "studentt"), pct0=0.75, method=c("MM", "ML"), diagnostic.plot=FALSE)
x |
vector of test statistics. |
statistic |
type of statistic - normal, correlation, or student t. |
pct0 |
fraction of data used to fit null distribution. If multipe pct0 are specified the fit is performed for each value, and the null distribution with smallest variance is chosen. |
method |
procedure used to fit null distibution, either ML (maximum likelihood) or MM (=median matching, the default). |
diagnostic.plot |
for length(pct0) > 1 plot the estimated
variances in dependence of the pct0 parameter. |
For median matching, the median of the truncated half-normal distribution (or the equivalent for correlation and student t distribution) is matched to the median of the absolute values of the censored sample. Note that if length(x) is small then median matching is less efficient than maximum likelihood.
For maximum likelihood the truncated density is fit to the censored sample. As this can be quite time consuming for large data set (and as there may be convergence problems) this is not the default method.
The estimate scale parameter of the null distribution. This depends on the type of statistic:
normal |
sd value. |
correlation |
kappa value (see dcor0 ). |
studentt |
degrees of freedom (df). |
# load "fdrtool" library library("fdrtool") # simulate normal data sd.true = 2.232 n = 5000 z = rnorm(n, sd=sd.true) censored.fit(z, statistic="normal") # simulate contaminated mixture of correlation distribution r <- rcor0(700, kappa=10) u1 <- runif(200, min=-1, max=-0.7) u2 <- runif(200, min=0.7, max=1) rc <- c(r, u1, u2) censored.fit(r, statistic="correlation") censored.fit(rc, statistic="correlation")