scaledTMix.null: Fit the null model to t-statistics

In gitlongor/hisemi: Hierarchical Semiparametric Regression of Test Statistics

Description

This function fits a null model to t-statistics, i.e., a two-component mixture, with one component being central t-distribution, the other component being scaled central t-distribution with scale parameter larger than 1. The mixing proportion for the central t-distribution is pi0.

Usage

scaledTMix.null(tstat, df, starts = 
        list(pi0 = seq(0.1, 0.99, length = 20), 
             scale = 2^seq(0.01, log2(max(abs(tstat))),
                length = 20)))

Arguments

tstat	A numeric vector of t-statistics.
df	A numeric scalar or vector of the same length as tstat, giving the degrees of freedom for the tstat.
starts	A list of two components, pi0 and scale. Each being a numeric vector, which defines the grid for searching starting values.

Details

The function finds maximum likelihood estimates of pi0 and scale. pi0 should lie between 0 and 1. scale should be larger than 1. The L-BFGS-B method is used in optimization function optim. The parncp function with zeromean=TRUE fits the same model, except that it returns ncpest object instead of hisemit object.

Value

A hesimit object.

Author(s)

Long Qu [e-mail]

References

Qu, Nettleton, Dekkers (2012) A Hierarchical Semiparametric Model for Incorporating Inter-gene Information for Analysis of Genomic Data. Biometrics, 68(4):1168-1177

See Also

parncp, scaledTMix.sat, scaledTMix.psat

Examples

set.seed(99927220)
(tstat=rt(5,1))
scaledTMix.null(tstat,1)

gitlongor/hisemi documentation built on May 17, 2019, 5:28 a.m.