A mega-analysis of personality prediction: Robustness and boundary conditions

Abstract

Decades of studies identify personality traits as prospectively associated with life outcomes. However, previous investigations of personality characteristic-outcome associations have not taken a principled approach to covariate use or other sampling strategies to ensure the robustness of personality-outcome associations. The result is that it is unclear (a) whether personality characteristics are associated with important outcomes after accounting for a range of background variables, (b) for whom and when personality-outcome associations hold, and (c) that background variables are most important to account for. The present study examines the robustness and boundary conditions of personality-outcome associations using prospective Big Five associations with 14 health, social, education/work, and societal outcomes across eight different person- and study-level moderators using individual participant data from 171,395 individuals across 10 longitudinal panel studies in a mega-analytic framework. Robustness and boundary conditions were systematically tested using two approaches: propensity score matching and specification curve analysis. Three findings emerged: First, personality characteristics remain robustly associated with later life outcomes. Second, the effects generalize, as there are few moderators of personality-outcome associations. Third, robustness was differential across covariate choice in nearly half of the tested models, with the inclusion or exclusion of some of these flipping the direction of association. In summary, personality characteristics are robustly associated with later life outcomes with few moderated associations. However, researchers still need to be careful in their choices of covariates. We discuss how these findings can inform studies of personality-outcome associations, as well as recommendations for covariate inclusion. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Figure 1.

Odd ratios fixed $({\mathrm{\gamma }}_{10}$) effects of personality characteristics on outcomes (dots) following propensity score matching. Blue squares represent effects whose 89% Bayesian Credible intervals of log odds did not overlap with 0, indicating reliable effects. Rows represent different outcomes, while columns represent different personality characteristics. Lighter, gray error bars represent 95% Credible Intervals, while black error bars represent 89% Credible Intervals.

Figure 2.

Odds ratios of fixed $({\mathrm{\gamma }}_{10}$) and study-specific (${\mathrm{\beta }}_{1j}$) effects of personality characteristics on mortality (dots) following propensity score matching. Lighter, gray error bars represent 95% Credible Intervals, while black error bars represent 89% Credible Intervals around each effect in the pooled models. Error bars that do not overlap with 1 are considered reliable associations of personality characteristics with mortality.

Figure 3.

Simple effects of all personality-outcome associations that were reliably moderated by parental education (0 = high school or less, 1 = college, 2 = beyond college). Each panel indicates a separate personality-outcome association in terms of the odds ratio associated with a particular personality score for different levels of parental education. groups. Dotted lines represent those who parents had less than a college degree, while solid and dashed lines represent individuals in each study whose parents had a college degree or beyond a college degree, respectively. Shaded regions represent the 89% Credible Interval around the fitted estimates. Personality characteristics are presented as Percentage of Maximum Possible (POMP), which ranges from 0–10 for each study, such that 10 represents the maximum possible score, while 0 represents the minimum possible score. The scale on each panel have been constrained to be the same. As a result, unreliable effects, like the Optimism-volunteering effect for non-Black people of color, appear exaggerated.

Figure 4.

Specification Curve of mega-analytic estimates of prospective Openness to Experience-mental health event associations across studies and all possible combinations of covariates. The top panel displays the odds ratio associated with a one-unit change in personality characteristic level (operationalized as POMP, 0–10) from each specification arranged by the size of the odds ratio from low to high. The bottom panel displays the details of the specifications by indexing which covariates were included in each specification and the significance of the relationship between Openness and mental health events. Specifications with significant Openness-mental health event effects (p < .05, using cluster-corrected robust standard errors) are in red and indicated by longer lines.

Figure 5.

Specification Curve of mega-analytic estimates of prospective Conscientiousness-major health event associations across studies and all possible combinations of covariates. The top panel displays the odds ratio associated with a one-unit change in personality level (operationalized as POMP, 0–10) from each specification arranged by the size of the odds ratio from low to high. The bottom panel displays the details of the specifications by indexing which covariates were included in each specification and the significance of the relationship between Conscientiousness and major health events. Specifications with significant Conscientiousness-major health event effects (p < .05, using cluster-corrected robust standard errors) are in red and indicated by longer lines.

Figure 6.

Specification Curve of mega-analytic estimates of prospective Extraversion-Divorce associations across studies and all possible combinations of covariates. The top panel displays the odds ratio associated with a one-unit change in personality characteristic level (operationalized as POMP, 0–10) from each specification arranged by the size of the odds ratio from low to high. The bottom panel displays the details of the specifications by indexing which covariates were included in each specification and the significance of the relationship between Extraversion and divorce. Specifications with significant Extraversion-divorce effects (p < .05, using cluster-corrected robust standard errors) are in red and indicated by longer lines.

Figure 7.

Specification Curve of mega-analytic estimates of prospective Neuroticism-mortality associations across studies and all possible combinations of covariates. The top panel displays the odds ratio associated with a one-unit change in personality characteristic level (operationalized as POMP, 0–10) from each specification arranged by the size of the odds ratio from low to high. The bottom panel displays the details of the specifications by indexing which covariates were included in each specification and the significance of the relationship between Neuroticism and mortality. Specifications with significant Neuroticism- mortality effects (p < .05, using cluster-corrected robust standard errors) are in red and indicated by longer lines.