e-mail thread between Phil and I 5-17-12

2013-12-01

azim58 - e-mail thread between Phil and I 5-17-12

selecting p value cutoff
3 messages

===========================================================================

Kurt Whittemore <kurtwhittemore@gmail.com>
 Wed, May 16, 2012 at 4:29 PM

To: Phillip Stafford <phillip.stafford@asu.edu>

Hi Phil, how do I select a p value cutoff. Kathy made the following
statement which I don't think is quite correct.

"Why are you using a p value of 0.05 for the random array results? there
are 10,000 tests, so p= 1/10k"

The main reason I don't think it is correct is because my t tests did not
involve 10,000 samples. My tests involved comparing the intensity of 3
peptides in normal vs 3 peptides in condition X so there are 6 samples.
So given the fact that I have 6 samples or 3 in each group, how do I
choose an appropriate p value cutoff?

-Kurt

===========================================================================

Kurt Whittemore <kurtwhittemore@gmail.com>
 Wed, May 16, 2012 at 4:30 PM

To: Phillip Stafford <phillip.stafford@asu.edu>

Ah so I guess my real question is how to choose alpha not p.
[Quoted text hidden]

===========================================================================

Phillip Stafford <Phillip.Stafford@asu.edu>
 Wed, May 16, 2012 at 6:19 PM

To: Kurt Whittemore <Kurt.Whittemore@asu.edu>

Actually you did do 10,000 different tests - each binding event is a
test. For each peptide, you have 6 samples, but every time you test a
peptide for significance, it's a test. The 1/10000 is the p-value for
which you expect 1 false positive. At p=0.05, you expect 5% false
positives (500), so if you only have 10 peptides that are significant at
p=0.05, then you have 490 false positives for your 10 hits.

Alpha is the significance level at which you do a significance test, p is
the probability of making a false interpretation of your data. In this
case it's the same thing.

Phil

azim58wiki: