The Perils of Partialing Optimism: Relationships with Personality and Health

The question of whether optimistic or pessimistic items in the Life Orientation Test-Revised better predict psychological and physical outcomes is often addressed by entering both subscales in the same model. This approach risks the “perils of partialing” in which partial constructs have lower reliability and different nomological nets than their unpartialed counterparts. The present studies tested the effects of partialing optimism and pessimism using structural equation modeling to test differences in external validity and intraclass correlations (ICC) to test similarity of profiles. In Study 1 (N = 2,281 older adults), the optimism-pessimism correlation was modest (r = -0.32), but nomological nets for unpartialed and partialed constructs with personality and health were significantly different (all p < .00001). In Study 2 (N = 739 younger and midlife adults), the optimism-pessimism correlation was higher (r = -0.61). Nomological nets were significantly different (all p < .00001). Examination of item correlations indicated that partialed constructs may have lost convergent validity with key theoretical elements of trait optimism. Differences between optimism and pessimism were fewer than those between each and its partialed counterpart, and results with partialed optimism and pessimism constructs should be viewed with caution.