 |
 |
The p value has the same representative role as CI to the statistical result. The former tell us that we accept for example a 5% of alpha error, meaning that we might have 5% of chance of accepting the alternative Hypothesis as true when in fact it is not. The CI, represents the interval in which we can accept the alternative hypothesis with a 95% of chance of being the truth.
When we get a paper, we can infer the CI if this is not presented.
For this, we have only to apply the formula:
SE = SD/√N
Then:
95% CI (lowest value) = mean –(1,96 X SE)
95% CI (highest value) = mean + (1,96 X SE)
Wishes,
Marcelo Katz, M.D PhD
Albert Einstein Hospital
Hi – The interpretation of whether a finding is likely to clinically significant or not, is one of key arguments for using CIs rather than p-values. I find it easy to teach my students the difference between clinical significance and statistical significance using CIs. If the confidence interval includes the no effect point, then the findings are not statistically significant whereas, if you would make a different clinical decision if the true effect was close to the lower confidence limit compared to the clinical decision you would make if the true effect was close to the upper confidence limit, then the findings are not clinically significant. You cannot easily teach this lesson with p-values.
regards
Rod Jackson
Professor of Epidemiology
Section of Epidemiology and Biostatistics
School of Population Health, Tamaki Campus
Faculty of Medical & Health Sciences, University of Auckland
Private Bag 92019
Auckland, New Zealand
On 10/01/11 11:38 PM, "Aicken, Catherine" <[log in to unmask]" target="_blank">[log in to unmask]> wrote:
Hi Anoop,
Smaller p-values can indicate statistical significance - but this is very
different from clinical significance!
For instance, if I have an intervention for obese people, and I've run a
well-designed RCT and found evidence for a tiny reduction in BMI in favour
of the intervention - but with a p-value of 0.01 - I would conclude that the
difference is *statistically* significant.
In order to find out whether this finding is *clinically* significant, I'd
want to find out whether there is any evidence that such a small reduction
in BMI is going to have a clinically meaningful effect on participants'
health - is it enough to reduce their risk of diabetes? CHD? Other
morbidities? Mortality?
If you are writing up any research, it's a good idea to distinguish between
clinical and statistical significance.
Best wishes,
Catherine
-----Original Message-----
From: Evidence based health (EBH)
[mailto:[log in to unmask]] On Behalf Of Anoop
Balachandran
Sent: 08 January 2011 16:33
To: [log in to unmask]" target="_blank">[log in to unmask]
Subject: Re: P - Values
Thanks for the detailed response.
I understand that with smaller p-values you can look to see if it is
clinically significant, but how do we know if this clinically significant
difference is near to the true effect. It could be more than our clinically
significant value or less. ( which we can only find from CI). Makes sense?
In the exercise science field, studies still use the p-value concept.