Ben Goldacre posted an interesting illustration of how data can mislead and needs to be analysed before posting. His example shows a correlation between lung cancer and drinking, that is, if you drink you are more likely to get lung cancer than if you don't drink. Does that really mean that drinking causes lung cancer or is there another factor in play?
When the figures are split into smokers and non-smokers, the smoker who drink and the smokers who don't drink have the same occurrence of lung cancer. Similarly, for non-smokers, drinker and non-drinkers have the same occurrence of lung cancer. This shows that the occurrence of lung cancer is explained by smoking versus not smoking rather than drinking versus not drinking.
No comments:
Post a Comment