Which P should I consider?

[Mariana Mira]

I run U-CARE and I am confused about which p value should I consider and report in my paper. For example: performing the test 3.SR I got the following results:

Code: Select all

N(0,1) statistic for transient(>0) =1.1866
    P-level, two-sided test =0.2354
    P-level, one-sided test for transience =0.1177
Log-Odds-Ratio (LOR) =3.512
N(0,1) standardized LOR statistic for transience (>0) =1.9902
    P-level, two-sided test =0.046566
    P-level, one-sided test for transience =0.023283
Overall test df =12
Quadratic Chi2 =6.4151
    P-level =0.89373
G2 =6.4151
    P-level =0.89373

So, I don't know if I consider the statistic for transient with P-level, one-sided test for transience=0.1177 or the standardized LOR statistic for transience with P-level one-sided test for transience=0.023. If I consider a 0.05 significance level, one test is significant but the other isn't. Also, I don't understand the other two tests (quadratic Chi2 and G2). Why their P-level are so high? If someone could help me to solve this fundamental question, I would really appretiate.
Thanks in advance
Mariana

[CHOQUET]

Hello,

In reply to:

> So, I don't know if I consider the statistic for transient with P-level,
> one-sided test for transience=0.1177 or the standardized LOR statistic > for transience with P-level one-sided test for transience=0.023.
> If I consider a 0.05 significance level, one test is significant but the
> other isn't.

Although the z-test and the LOR tests are asymptotically equivalent.
The one side z-test has a lot of power sor I recommend to used this test.

> Also, I don't understand the other two tests (quadratic Chi2 and G2).
> Why their P-level are so high?

Definitively, there is no transient effect in our case.

Sincerely,

Rémi

[Mariana Mira]

Thank you Rémi! I still have another group with similar results, but at time, the z test is <0,05 for one-sided test for transience and LOR statistic for transience is >0,05. Also the P-level for the other tests are not so high. So, should I consider a transient effect or not for this group?

Code: Select all

N(0,1) statistic for transient(>0) =1.6502
    P-level, two-sided test =0.098893
    P-level, one-sided test for transience =0.049446
Log-Odds-Ratio (LOR) =2.6666
N(0,1) standardized LOR statistic for transience (>0) =1.6164
    P-level, two-sided test =0.10601
    P-level, one-sided test for transience =0.053005
Overall test df =9
Quadratic Chi2 =10.9662
    P-level =0.27804
G2 =11.1402
    P-level =0.26622

[CHOQUET]

To my opinion, it is not necessary for this group to consider a transient effect because on one hand the chi2 is far from being significant, on the other hand the one sided z-test is very close to 0.05.

Rémi