Multiple testing with questionnaire data

It is wise to look for supporting evidence before interpreting an association as evidence of a cause and effect relationship. In particular, you should be looking for a credible mechanism for any such relationship. Where there is a reasonable biolo-gical/chemical/physical explanation for how the alleged causal factor would bring about the effect, you will be on much safer ground. 

The classic example is smoking and lung cancer. The initial evidence was necessarily epidemiological and the tobacco barons were long able to muddy the waters with disingenuous arguments that the association might be confounded. 

Now that we can identify the relevant chemicals in cigarette smoke and have elucidated the details of how they damage DNA, the case for a causal relationship is clear to all but a few diehard smokers (there are none so blind as those who do not wish to see). 

A commonly used way of side-stepping the issue is to talk in terms of association rather than causality. In the case of the opinions on contraceptive supply, we might have collected data on health status, but not on age. In that case, we would simply be faced with the rather odd fact that those with poorer health generally held more conservative opinions. Recognizing that a cause and effect relationship was unlikely, we could limit ourselves to saying that Poor health is associated with a more conservative opinion on changes in the legislation . 

The great example of confounding (OK, it is probably mythical) was the survey that found that people who carried matches were more likely to contract lung cancer - a finding that might be perfectly credible, but needs to be interpreted with caution. 

---

Emphasize the potential hazard of multiple testing with questionnaire data... 

---