If the confidence interval includes or crosses (1), then there is insufficient evidence to conclude that the groups are statistically significantly different (there is no difference between arms of the study).

Essam -

Stick with confidence intervals (prediction intervals for regression).  P-values are often misleading.  You can make your p-value smaller by increasing the sample size, or larger by decreasing it, with the same effect size. 

Note the following:

Press release for the American Statistical Association:

http://www.amstat.org/newsroom/pressreleases/P-ValueStatement.pdf

And related: 

 https://www.researchgate.net/publication/262971440_Practical_Interpretation_of_Hypothesis_Tests_-_letter_to_the_editor_-_TAS

Cheers - Jim

PS -  Confidence intervals are more practically interpretable.  

Article Practical Interpretation of Hypothesis Tests - letter to the...

I agree with Dr. Knaub.  The point estimate with its confidence interval is more useful, especially for  estimating an effect in a meta-analysis.  In fact, I would remove the p values from your forest plot, if I were you.  They do not add any relevant information.

Thanks for all above answers, but unfortunately, the majority of answers didn't understand my question EXCEPT,  the answers of James R Knaub  and Keturah Faurot. Thus, that is why I deleted the above figure.

My question was, when there is conflicting between value of CI and p-value , as in my analysis, CI showing no significant (because CI crosses the null hypothesis (1), whereas ,p-value indicating it was significant (less than 5%)

Again , my question in other words: What is the value (P value or CI (that can be relied upon to take a decision in respect of the significances ?

Hi. Dr. Essam.

In my own opinion the confidence interval is the value you can rely upon over the p- value...

Have a look at the attached file.

Regards.

Yes Dr Mohammed, that what I mean

Thanks

Dear Sofia Cividini,

Thank  you for your clarification 

how ! 

CI crossing the null hypothesis (0 in a scale or 1 in a ratio) must have p value more than 0.05 

> 1 to > 1 means a positive significant association

< 1 to < 1 is means a negative significant association

< 1 to > 1 means an insignificant association

Explaining the reason behind considering a value of 1 in CI as statistically insignificant , in cases of Relative risk and odds ratio.

If the ratio = 1 this means that the incidence of

outcome in the intervention (or exposed) group is

equal to that in the control group i.e. there is no

difference in the outcome between the intervention

(or exposed) group and the control group.

Can someone explain why it is possible for the p-trend to be statistically significant if all the individual OR values are non-significant?

I don’t have an answer. However, for those who said discrepancy is not possible I have an example for a 2x2 contingency table (24,1,86,25) when Fisher’s p=0.046 while 95% CI range is 0.9-54.0.

I am also facing the similar problem, what should I do if my P value rejects the null hypothesis but test statistics (beta coefficient) falls inside confidence interval? I got different answers from different statisticians.

What is the reason behind a result HR 1 (95% CI 1-1) with p value =0.0049 considered as significant?

Since there are still questions on the problem: p-value indicates significance - but the 95%-CI does not.

I like to push the answer of Pavel Ruzankin . There are several different possibilities to calculate a confidence interval, depending on the distribution that one assumes under the null hypothesis. For instance, the 'typical' calculated 95%-CI of regression parameters in any program of your choice is based on quantiles of a normal distribution (Wald confidence interval). This assumption of normality does not have to be true.

Now for the crosstab example of Valentina Pilipenko . There the two-sided Fisher's exact test gives a p-value of 0.046 (using a hypergeometric distribution) and the 95%-Wald(!)-CI gives 0.89 to 54.16. Calculating the 95%-Ci by means of the minimum likelihood one obtains: 1.11 to 147.0. This kind of calculation respects the used test procedure. So that should be the appropriate confidence interval.

I really recommend reading the manual of the R package 'exact2x2', there you will find the different 'kind' of Fisher's exact test and the appropriate 95%-CI of the odds ratios.

Nevertheless I would be careful with a 'significant conclusion' for 2x2 table like this.

Greetings,

Markus

Jifmi JOSE Manjali it considered significant because you rounded the 95% CI with one digit only. Your 95% CI, for example, can be ranged from 1.000001 to 1.000002

Upendra Raj Dhakal your findings should be found under the range of CI. For example, the odds ratio of 2.5 with 95% CI ranged from 1.5 to 4.5. This considered a significant odds ratio. BTW, your question is so confusing.

Valentina Pilipenko In your case, the 95% CI is not crossed "0", so it is significant. It's not the odds ratio.

Rishika Jain sometimes we don't have enough samples to find the significance.

Essam A. Al-Moraissi If possible, you may share the exact values with us. Because your question was unclear.

What I tell people is that they are looking at two measures of 'significance'. Although it may have met the threshold for p value 'statistical significance' what a small beta and CI touching or crossing 0 are telling you is that the effect itself is very small, not 'clinically significant'. I have also seen this in multivariate regressions with too many variables.

Article Explaining Odds Ratio

No,it is not significant

Would offer a question that might be an answer. Contemplate what is the significance (value) of a study where the CI includes (crosses) both "0" and "1"? An example would be the Global mortality from outdoor fine particle pollution generated by

fossil fuel combustion: Results from GEOS-Chem study....where there the CI for excess death due to carbon fuel combustion is 95% CI: -47.1 to 33 17.0) million premature deaths annually attributable to the fossil-fuel component of PM2.5. As a professional, I would have to reject this study if prescriptive authority were to be exercised. Any comments would be appreciated.