Rich Ulrich <[EMAIL PROTECTED]> wrote:
> On 21 Oct 2003 05:44:47 GMT, David Duffy <[EMAIL PROTECTED]>
> wrote:

>> Rich Ulrich <[EMAIL PROTECTED]> wrote:
>  [ ... ]
>> > On 17 Oct 2003 06:05:19 -0700, [EMAIL PROTECTED] (Enda Kelly)
>  [ editing and rearranging the table ]
>> >> Locus pair        P-value         Median          Lower           Upper
>> >> 1         0.002414        0.002422        0.000055        0.037401
>> >> 4         0.018936        0.016100        0.001082        0.193285

>> >> 2         0.971621        0.512296        0.181935        0.850761
>> >> 5         0.832001        0.505662        0.173286        0.857703
> [ ... ]

> DD > 
>> They are actually what you expect, I think, from a noncentral
>> chi-square.  If sampling from a 1-df central chi-square, for example, one would

> David, you mis-read, I hope.  I can see that your
> description fits what I said about #1 and #4.  Yes, 
> I was not surprised by the wide, multiplicative range 
> of the p-levels.  The original poster called  those  *narrow*  
> since the whole action was within a percentage point;
> I objected.

> It was just the #2 and #5  that were not 'useful' results,
> since the point estimate is outside the CI, for #2, and
> nearly so for #5.  That can happen, but it is pathological.  

> Agreed?
Surely.

I would guess that in #2 and #5, the table is sparse, so that the "observed" P-value
hinges on one cell of size 1-2.  I am always surprised by how much the
(Fisher) P-value moves around when one combines cells of small size
(plus people complain when their publishable P-value disappears ;)).  And
I'm personally still not sure how to interpret such results,
specifically when looking at association with a rare risk factor/haplotype, given that
adding just a few more observations to a study can make a fair
difference.  The problem, in this particular domain, is that there can
be many rare haplotypes, some of which merely represent measurement (genotyping) 
errors.
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to