View Single Post
  #167 (permalink)  
Old 14-May-2008, 06:26 PM
Disinfo Agent Disinfo Agent is offline
Senior Member
 
Join Date: Apr 2004
Posts: 6,342
Default

Quote:
Originally Posted by mike alexander View Post
First sentence I agree with completely. Second I would qualify by saying that if the correlation increases as the sampled population increases you might be looking at a real causal link.
I'm not sure if that's what you actually meant to write, but note that I did not speak of the correlation increasing as the sample size increases...

However, if you dump enough raw data into a sample, any correlation, no matter how tiny (and irrelevant in practice) will eventually become statistically significant.

Quote:
Originally Posted by mike alexander View Post
Part of the problem is that in my line of work an r of 0.5 would get me laughed out of the room.
Yes, statistical significance must always be weighted against other considerations, one of which is the practical context.
__________________
"All your bias are belong to us." Ara Pacis
"A witty saying proves nothing." Voltaire
Reply With Quote