Basic caution about cancer clusters

[Vincent.King@DOEGJPO.COM (Vincent King)]

     Norm,
     
     A word of caution on reading too much into cancer 'clusters'.  This 
     is an area that is routinely abused to 'prove' whatever effect one 
     is looking for, and it always works.  Here's why:
     
     Let's use an example you mentioned, breast cancer rates around 
     nuclear power plants. Assume for a moment that there is absolutely 
     no connection between the two (I know RADSAFERs are notoriously 
     shy, but you've probably detected by now that this opinion is held 
     by many on the list).  
     
     Find the average breast cancer rate; now, how many of ALL the NPP 
     sites are above the average?  By definition, 50%.  (As obvious as 
     this is, it would not surprise me at all to see a reporter flash 
     this as an 'alarming' headline, like the old joke about how 50% of 
     kids read at a below-average level.)
     
     And when you look at JUST the top 5%, or 1%, then you will almost 
     certainly find what amounts to a statistically significant 
     correlation.  But ONLY if you ignore the fact that the bottom 5% 
     looks the same, except in the opposite direction.  Any study where 
     the population is hand-picked to include only the top few sites can 
     do this.
     
     More importantly, if you want to find an effect at one particular 
     site, what do you do if it falls in the 'below average' category?  
     Simple: you just look for a different 'effect'.  After all, if 
     there is NO effect on anything, you will still find that around 
     half of all cancer types at your site will be above the national 
     average, and half below average. If breast cancer rates don't give 
     the answer you want, dump it as an 'effect' and go with a different 
     effect that does work.  (This is the more subtle abuse of 
     statistics, in my opinion.)
     
     These abuses of statistics are so easy to make that they sometimes 
     occur as honest mistakes made by people who should know better.  
     But they also regularly show up as intentional distortions.  That's 
     why you see such dogged insistence on proper controls for 
     statistical studies from many of us.
     
     By the way, I'll bet Karen wishes she could use the computer once 
     in a while ;)
     
     Have a good day,
     
     Vincent King
     vincent.king@doegjpo.com 


************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html