[BioC] Multtest distributions

[Ken Termiso]

Hi all,

I was working through the multtest tutorial pdf that comes with the package, 
and also perusing the archive here of multtest threads, and for the 
non-statistician, it is unclear as to which distributions to use for each 
test. As an excercise, I have attempted to call each statistic implemented 
in multtest. Hopefully this will prove helpful for others, too.

These are the tests implemented in multtest:

teststat_t_para <- mt.teststat(m,m.cl,test="t",nonpara="n")
teststat_t_nonpara <- mt.teststat(m,m.cl,test="t",nonpara="y")
teststat_te_para <- mt.teststat(m,m.cl,test="t.equalvar",nonpara="n")
teststat_te_nonpara <- mt.teststat(m,m.cl,test="t.equalvar",nonpara="y")
teststat_w_para <- mt.teststat(m,m.cl,test="wilcoxon",nonpara="n")
teststat_w_nonpara <- mt.teststat(m,m.cl,test="wilcoxon",nonpara="y")
teststat_f_para <- mt.teststat(m,m.cl,test="f",nonpara="n")
teststat_f_nonpara <- mt.teststat(m,m.cl,test="f",nonpara="y")
teststat_pt_para <- mt.teststat(m_pt,m_pt.cl,test="pairt",nonpara="n")
teststat_pt_nonpara <- mt.teststat(m_pt,m_pt.cl,test="pairt",nonpara="y")
teststat_bf_para <- mt.teststat(m_pt,m_pt.cl,test="blockf",nonpara="n")
teststat_bf_nonpara <- mt.teststat(m_pt,m_pt.cl,test="blockf",nonpara="y")

This assigns the raw P-value assuming two-tailed tests, and if I have, say, 
3 control and 3 experimental chips, I guess I should add "df=5" :

rawp0_teststat_t_para <- 2 * (1 - pt(abs(teststat_t_para), df =5))
rawp0_teststat_t_nonpara <- 2 * (1 - pt(abs(teststat_t_nonpara)), df=5)
rawp0_teststat_te_para <- 2 * (1 - pt(abs(teststat_te_para)), df=5)
rawp0_teststat_te_nonpara <- 2 * (1 - pt(abs(teststat_te_nonpara)), df=5)
rawp0_teststat_w_para <- 2 * (1 - pwilcox(abs(teststat_w_para)), df=5)
rawp0_teststat_w_nonpara <- 2 * (1 - pwilcox(abs(teststat_w_nonpara)), df=5)
rawp0_teststat_f_para <- 2 * (1 - pf(abs(teststat_f_para)), df=5)
rawp0_teststat_f_nonpara <- 2 * (1 - pf(abs(teststat_f_nonpara)), df=5)
rawp0_teststat_pt_para <- 2 * (1 - pt(abs(teststat_pt_para)), df=5)
rawp0_teststat_pt_nonpara <- 2 * (1 - pt(abs(teststat_pt_nonpara)), df=5)
rawp0_teststat_bf_para <- 2 * (1 - pf(abs(teststat_bf_para)), df=5)
rawp0_teststat_bf_nonpara <- 2 * (1 - pf(abs(teststat_bf_nonpara)), df=5)

But again, I am not sure that I am using the correct distribution for each 
test...and also am not sure if the degrees of freedom needs to be calculated 
specially for any test.

Any corrections would be much appreciated.

Thanks in advance,
Ken