crashfrog writes:
Samples of populations, analyzed for certain traits. So long as we're able to conclude that the distribution of the sample has a 5% chance or better of being a random deviation from the "expected" distribution of the population, we conclude that it is.
Like, if we have a sample of people and we're looking at their height, and we find that there's a 5% chance or greater that the difference between the distribution in our sample and the distribution in the whole is due to nothing more than chance, then we conclude that the sample we have is truly random, and not the result of some kind of selection for heights.
Again, unless I'm way off base, here, we take anything over 5% confidence. Yeah, I was surprised, too.
No, you're way off base - it's just the opposite. We consider it significant if there's less than a 0.05 probability (assuming the experiment was designed with a p<0.05 significance level) that the null hypothesis was rejected when it was in fact true (Type I error). Confidence intervals of 95% or 99% are commonly used.
Edit: I totally misread your posts, please disregard the above. Upon re-reading, if I'm seeing it right this time, you're saying that they do the same thing I described; they consider the difference to be statistically insignificant if there is a probability of 0.05 or more that it is due to random chance. But that means that they're working with a 95% CI.
This message has been edited by Belfry, 02-11-2006 05:54 PM
This message has been edited by Belfry, 02-11-2006 06:50 PM