One method to test this “smoking as a marker” explanation is to determine if the association of current smoking and suicide is eliminated when such “confounders” are entered as covariates in an analysis (Smith & Phillips 1992). Some of the epidemiological studies described below found the association persisted in spite of entering confounders (Iwasaki et al. 2005; Miller et al. 2000a; Miller et al. 2000b; Tverdal et al. 1993); however, these analyses may be inaccurate because they did not include the most important confounders. Other studies found the association did not persist when “confounders” were entered (Hemmingsson & Kriebel 2003; Kessler et al. 2007; Leistikow et al. 2000; Boden et al. 2007); however, these may be inaccurate because what they classified as confounders are actually probably mediators (Breslau et al. 2005). For example, if the association of smoking and suicide disappears when depression is added as a “confounder” (Breslau et al. 2005), this may not rule out a causal effect of smoking but rather may indicate there is a causal effect of smoking, but it is mediated by increased risk
