Here's Frances Woolley's review of Phishing for Phools. She there takes on their argument that alcohol is more harmful than tobacco:
I'd point to this post on alcohol and co-morbidity.So it's not that hard to show alcohol is more harmful than tobacco. The trick is to focus on total harm, as opposed to harm per user, and on measures of morbidity, rather than mortality.The belch in Akerlof and Shiller's argument is that they mostly ignore this evidence on disability, focussing instead on a radically different type of harm: "loss of affect" or "loss of capacity for intimacy". Alcohol, they argue, causes "subjective, hard-to-observe changes in personality", and leads to divorce and the break up of relationships. The major piece of evidence backing up this position is the Harvard Grant Study, which tracked Harvard students from the classes of 1939 to 1944, and found that those abused alcohol at some point in their lives (23%) died young. Moreover, "alcohol wrecked their ability to relate to others."Making any kind of statement about the psychological impacts of alcohol is hard, because so many people use alcohol to self-medicate. As Eric Crampton has argued here, alcohol abuse can be symptom of depression, anxiety, or other personality disorders, rather than a cause. Akerlof and Shiller argue that the alcoholics in the Harvard Grant study were, at the start of the study, "no different from their more sober peers" in terms of their personalities and family backgrounds. Instead, "alcoholism had changed their personalities." This reading of the evidence seems to seriously underrate the ability of present or future alcoholics, and their families, to lie and dissemble, and seriously overrate the assessment abilities of late-1930s social scientists. I can believe that alcohol use impacts people's personalities. But I find it hard to believe that, absent alcohol use, alcohol abusers would be just like anyone else.
Co-morbidity is complicated and hard. Just comparing the life satisfaction survey figures, or other outcomes, between users and non-users is hardly sufficient: where some substance use is part of self-medication for prior illnesses, attributing the total difference between users and non-users to use is simply wrong. Could be that outcomes are better than you might have expected, if the self-medication works; could be that outcomes are worse. Either way you need to be looking at the marginal increment against the proper baseline rather than just comparing outcomes across users and non-users.
You might think this is just a problem for crappy cost-of-illness studies that nobody should ever take seriously anyway. But look at the psychiatric hospitals that went and banned smoking. Turns out now that smoking behaviours among schizophrenics is very plausibly self-medication. See also here.
People consume psychoactive substances for reasons. Assuming away those reasons and concluding people are irrational or phools...