On 01-04-98 William Elliot wrote to Dennis Menard... 
 
        Hello William, 
  
        [snip] 
  
 WE> WE> Now, if you considered the number of arrests to the number of 
 WE> WE> convictions in each group and compared those ratios, a bias would 
 WE> WE> become more apparent as the ratios should be nearly equal.  Yes, 
 WE> WE> this 
 WE> WE> is testing for a different sort of bias, but it has an advantage 
 WE> WE> of being more objective. 
   
 WE> DM> Yes.  But you need lots of raw data for one to be able to do this 
 WE> DM> oneself; and where does one find this kind of analysis performed 
 WE> DM> and the results made available to the general public? 
 
  As I pointed before, the words above "a bias would become more apparent 
as the ratios should be nearly equal" are somewhat misleading particularly 
with their racial overtones and not quite accurate from a statistical 
standpoint.  
  From "A Study Guide to Epidemiology and Biostatics" page 97, 
bias is: 
 
        Bias is systemic error, resulting in over or underestimation 
        of the strength of the association." 
 
  Please notice, -caustion- may give a strong association without 
BIAS. 
  Substituting in the above quoted sentence, "systemic error, 
resulting in over or underestimation of the strength of the 
association would become more apparent as the ratios should 
be nearly equal." 
  As I pointed out before, WITHOUT BIAS, (systemic error, resulting 
in over or underestimation of the strength of the association), 
one -MAY HAVE DIFFERENCES- in ratios from causation. 
  The point "unequal ratios" don't imply bias.  
 
Take care, 
John 
 
___ 
 * OFFLINE 1.54 
--- Maximus 3.01
---------------