Quote:
Originally Posted by Robert Tulip
Correct me if I am wrong, but I would expect the products of these results would give combined probability: ie the likelihood of both the Uranus Rainest Day and Uranus Driest day occurring by chance would be 3.65% x 18.8% = 0.686%. Combining all six results above in this way give a result with likelihood of one in 9000. Significant.
|
OK, Robert, I'll correct you.
You can't cherry-pick the results you like and disregard the others, and you certainly can't just multiply the probabilities together like that. Especially when they're drawn from the same raw data, so the same outliers are probably responsible for multiple "effects" - in other words, your probabilities are not independent. If you want to look at combined effects, you'll need to do a multivariate regression or at least an analysis of variance.
Your best result, assuming your assumption of normality is correct (and I'm sceptical - did you test this assumption?), and assuming all your numbers here are kosher, gives p=0.04 - we would expect one in 25 tests to be this significant, purely because of random variation. And that's with a one-tailed test - given the vagueness of your initial hypothsis, I think a two-tailed test would be preferable.
I think your best course of action from here would be to look into either some classes in probability and statistics, or at least get hold of a textbook and work through it. Wikipedia's statistics pages are pretty good (by Wikipedia standards), but they're no substitute for a systematic education in the subject.