Health benefits: Meta-analytically determining the impact of well-being on objective health outcomes

Abstract

This research synthesis integrates findings from 150 experimental, ambulatory and longitudinal studies that tested the impact of well-being on objective health outcomes. Results demonstrated that well-being positively impacts health outcomes (r=0.14). Well-being was found to be positively related to short-term health outcomes (r=0.15), long-term health outcomes (r=0.11), and disease or symptom control (r=0.13). Results from the experimental studies demonstrated that inductions of well-being lead to healthy functioning, and inductions of ill-being lead to compromised health at similar magnitudes. Thus, the effect of subjective well-being on health is not solely due to ill-being having a detrimental impact on health, but also to well-being having a salutary impact on health. Additionally, the impact of well-being on improving health was stronger for immune system response and pain tolerance, whereas well-being was not significantly related to increases in cardiovascular and physiological reactivity. These findings point to potential biological pathways, such that well-being can directly bolster immune functioning and buffer the impact of stress.

Keywords:

• Objective health outcomes
• physical functioning
• subjective well-being
• positive affect
• health processes
• meta-analysis

Acknowledgements

We are appreciative of the numerous comments and edits on early drafts of this paper by Colleen J. Howell, Ph.D. We are also grateful to Danielle O'Brien, Yazmin Perez, and Katrina Rodzon for assistance with preparing the manuscript.

Notes

1. For the studies included in the meta-analysis, experimental investigations typically followed a similar paradigm. Well-being and physiological variables were measured at baseline, mood or emotion was manipulated, and the physiological variables were measured one or more times; mood/emotion was again assessed immediately following a manipulation check, and the physiological variables were again assessed. In some experiments, subjects acted as their own control, experiencing each mood condition; their reactivity in each condition was compared across conditions and to their baseline level, using repeated measures analysis of variance or similar methods (e.g., Brosschot & Thaler, 2003; Clark, Iverson, & Goodwin, 2001; Codispoti et al., 2003).

Other studies randomly assigned participants to a single mood/emotion condition and compared between subjects, either controlling for baseline levels or using change scores (e.g., Gendolla & Krüsken, 2001a, b). Some of the more recent studies have incorporated multi-level modeling methods to analyze within and between person changes (e.g., Polk, Cohen, Doyle, Skoner, & Kirschbaum, 2005). We do note that there is a lot of variation by study, depending on the outcome of interest, the size of the sample, and the methods used. For example, in one study, pictures were used to induce positive, negative, or neutral moods (Codispoti et al., 2003). Ten participants experienced each condition, 1 week apart, in counterbalanced order. Blood was drawn at baseline, 30 min after baseline, and after the manipulation, and one-way repeated multivariate analysis of variance was used to analyze the effect of picture valance on several neuroendocrine markers. In another study, 54 students were randomly assigned to a negative or positive mood, and to an easy or difficult task (Gendolla & Krüsken, 2001a). Heart rate, blood pressure, and skin conductance were continually monitored. Change scores between the average baseline function, during manipulation, and post-manipulation were used to assess the effect of mood on physiological response.

2. The most straightforward way to interpret the meta-regression statistics from Table VII is to write out the regression equations from these tables. The basic equation would be as follows:

where y is the predicted effect size; β ₀ is the intercept (when β ₁=0); and β ₁ is the slope (the change in the predicted effect size with a unit change in the predictor). For example, if we consider the first significant slope from Table VII (percentage of male respondents), we can use the meta-regression coefficients to predict the effect size between well-being and improved immune functioning for samples differing in gender composition. In this case, the predictor (gender composition) runs from 0.00 (a completely female sample) to 1.00 (a completely male sample). Thus, the regression equation for this example (see Table VII) would be

where X ₁ is the proportion of the sample that is male. For example, if the proportion is 0.00 (completely female sample), the predicted effect size between well-being and improved immune functioning is 0.392. If the proportion is 0.50 (half female/half male), the predicted effect size is 0.258. If the proportion is 1.00 (completely male sample), the predicted effect size is 0.125. We observe that as the sample composition becomes more dominated by males, the strength of the well-being–improved health effect size decreases. These data suggest that well-being may be more strongly related to improved immune functioning for females.