These conclusions must be tempered by several limitations with the above data. First, suicides and even suicidal ideation are rare events (e.g. the incidence of completed suicide in a lifetime is about 1/10,000 and for suicide ideation is about 1/1000) (Sudak 1999); thus, to detect the onset of suicide in a study lasting only a year probably requires sample sizes of over 10,000. Second, most studies of suicide and smoking have been case-control studies and this type of study almost always suffers from some self-selection or indication bias (Rothman & Greenland 1998; Shiffman et al. 2005; Klungel et al. 2004). None of the case-control studies adequately measured all the important possible confounders. Third, among the few prospective studies; most could have produced false negative reports because those with suicidal ideation may not report it, or may drop out of the study, or may be lost-to-follow-up. Since participants in smoking studies are usually reporting outcomes to non-counselors, they may be especially reluctant to reveal suicidal ideation. In addition, most prospective studies did not mention suicide; thus, it is unclear if they
