| |
 |
|
|
Science Forum Index » Statistics - Math Forum » The Matemathicias concernes
Page 1 of 1
|
| Author |
Message |
| Luis A. Afonso |
 Posted: Fri May 02, 2008 5:08 am |
|
|
|
Guest
|
In fact these 3 guys better to play GOLF than to try to say the most ridiculous even thought.
B. V. North,1 D. Curtis,1 and P. C. Sham2
1Joint Academic Department of Psychological Medicine, St Bartholomew’s and Royal London School of Medicine and Dentistry, and 2Department of Psychological Medicine, Institute of Psychiatry, London
Let´s appreciate:
*** . . . Typically, the estimate of the P value is obtained as p^ = r/n where n is the number of replicate samples that have been simulated and r is the number of these replicates that produce a test statistic. . .
. . Although use of (r+1)/(n+1) produces an unbiased estimate of the true P value (in contrast to use of r/n), this procedure will consistently overestimate small P values but will underestimate large P values. In fact, the expectation of (r+1)/(n+1) is (np+1)/(n+1), so that the bias is (1-P)/(n+1). Once again, when n is large, this overestimation is unlikely to be important.(. . . ) ***
************************
My comment
Unlike to be important, they say? Surely of no importance I add.
____Let be a0 = (r+1)/ (n+1) , a= n/ r
____Then a0/a = 1 + 1/ r approximately
Since n is never less (a somewhat imprecise MC) than 400´000 and p(W<=r) >= 0.005 (alpha=1%, 2-tail test) the difference is 0.05%..
Lis Amaral Afonso (The Moderator Detroyer) |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Tue May 13, 2008 8:19 pm
|
|
