>>Using proper statistical methods, a person who gets lung cancer having been exposed to only a trivial amount of secondhand smoke should be regarded as a data point against the proposition that secondhand smoke is significantly harmful, in that it shows that not all cases of lung cancer are attributable to SHS. To a statistical homeopathist, however, any case of cancer by anyone with any exposure whatsoever to SHS proves that SHS is dangerous.<<
Both of the approaches are incorrect.
I'm not suggesting that a statististician should say "Hmm... this person was only exposed to second hand smoke and got cancer, ergo smoking is not harmful." Rather, I'm saying that if a statistician were to separately consider two data sets, identical except that the former data set included a person who got lung cancer after having once walked within 500 feet of a bar where people were smoking but had no other exposure to tobacco smoke, a proper statistician should likely regard the correlation between smoke exposure and cancer to be slightly weaker in the former data set (with that data point) than in the latter. If the data sets are of reasonable size, one data point shouldn't affect the conclusion much, but its effect should be in the direction of reducing the correlation.
To put it simply, there is a certain "background level" of cancer which will occur independent of any exposure to tobacco smoke. Anti-smokers regard every single instance of cancer among those who have any exposure to tobacco as being a "tobacco-related" cancer; further, they aggressively classify people who get cancer as having been "exposed" to tobacco. By contrast, people who don't get cancer aren't considered to have been exposed to tobacco unless such exposure is undeniable.