Aspinwall and Tedeschi discuss the association between optimism and health in two noncontiguous paragraphs. In the first paragraph, they cite a study of optimism and survival in head and neck cancer in support of an association between optimism and survival [30
]. Yet, in the second paragraph, they note that associations between optimism and cancer mortality are more generally weak or nonsignificant. Examining the study of head and neck cancer patients, we note that it was a small, underpowered investigation (45 deaths being explained), the lower limit of the confidence interval barely excluded 1.0 (confidence interval
1.24), and even this effect depended upon multivariate analyses inappropriate for such a small number of deaths being explained.
We can underscore that the association between positive psychological traits and states and cancer incidence or mortality is small or nonexistent with references to recent studies. A well-controlled study of the survival of head and neck patients with an ample number of deaths being explained, 646, found no evidence of an association between emotional well-being and survival in simple or multivariate analyses, despite an exhaustive search for main or interaction effects [31
]. Tindle et al. [32
] examined mortality in a Women’s Health Initiative cohort of 97,253 women, 7,994 of them African-American. Optimism did not predict cancer-related mortality in the full sample or among white women, but it did predict cancer-related mortality among African-American women. We might be tempted to single out the apparent positive finding for African-American women, but then, we must ask what a priori explanation is there for the association not holding for white women or the full sample, but only for African-Americans? Moreover, mortality was more related to a full range of background factors and strongly related to depressive symptoms. Risk factors other than optimism were entered as statistical controls for the prediction of cancer-related mortality by optimism, but the categorization of these variables for entry into the multivariate proportional hazards was crude, and still allowed for residual confounding (for a discussion of this issue, see our earlier paper [10
] as well as the humorous, but insightful article by Davey Smith and Ebrahim [33
]). Finally, given the strong negative association between optimism and depressive symptoms, it is not clear whether the prediction of mortality from depressive symptoms controlling for optimism would not have been stronger than the investigators’ preferred prediction of mortality from optimism controlling for depressive symptoms.
Aspinwall and Tedeschi dismiss the possibility that the relation between positive states and health outcomes can be explained by the “detrimental effects of either pessimistic expectations or state or trait forms of negative affect, such as distress, depression, or anger” (in this issue). They cite a recent systematic review and meta-analysis of the association between optimism and physical health [34
]. Yet, in this review, Rasmussen et al. actually found that pessimism predicted health outcomes as well as or better than optimism, raising questions as to why optimism should be favored.
Aspinwall and Tedeschi indicate that even if an association between positive thoughts and feelings and mortality cannot be demonstrated for cancer, it has been established for cardiovascular disease and cite a recent comprehensive review and meta-analysis by Chida and Steptoe [35
]. At first glance, the review does indeed give that impression, and so deserves a closer look. For cardiac mortality occurring in nonpatient samples, Chida and Steptoe included six effect sizes (see their Table 1), but two came from the same study, violating basic assumptions of meta-analysis. Chida and Steptoe also found overall highly significant statistical heterogeneity in the effect sizes of the studies included in their analyses of nonpatient samples and significant publication bias. Examining their Fig. 2, it is clear that the statistical heterogeneity comes largely from a strong publication bias, whereby published small, underpowered studies nonetheless yield larger effect sizes than larger studies. Indeed, none of the larger studies produced a significant association between positive psychological variables and cardiac mortality. For persons already diagnosed with cardiac disease, the picture is confusing (their Table 2): there are five effect sizes entered into the meta-analysis, two from the same study, and of the five, two are not significant, one is significant in the right direction, but two are significant in the wrong direction, with higher scores on positive well-being associated with greater mortality.
No doubt that associations between positive psychology variables and reduced cardiac mortality can sometimes be found in some studies, as in the Women’s Health Initiative for white women, but not in the sample of 7,994 African-American women [32
]. Note the contradiction of these findings with the ones obtained for cancer in the same sample. Yet, overall, the association has a will-o’-the-wisp quality, leaving “serious conceptual and methodological reservations” [13
, p. 960] about any substantive interpretations, and little obvious public health or clinical implications. Moreover, as Chida and Steptoe [35
] indicate, there is a significant publication bias in favor of positive findings, and a perusal of the positive psychology and health literature indicates a persistent amplification of any signals of an association. Why has so much importance been attached to demonstrating that optimism predicts health and mortality? Such claims have taken on an ideological importance and are resistant to null and inconsistent findings. The notion that being optimistic improves health is invoked in promoting positive psychology intervention research and in the marketing of positive psychology as a commercial enterprise, even if, ironically, optimism is theoretically a relatively immutable trait.