Measuring well-being: A comparison of subjective well-being and PERMA

Abstract

We compared Seligman’s PERMA model of well-being with Diener’s model of subjective well-being (SWB) to determine if the newer PERMA captured a type of well-being unique from the older SWB. Participants were 517 adults who completed self-report measures of SWB, PERMA, and VIA character strengths. Results from four analytic techniques suggest the factor underlying PERMA is capturing the same type of well-being as SWB. Confirmatory factor analysis yielded a latent correlation of r = 0.98 between SWB and PERMA. Exploratory structural equation modeling found two highly related factors (r = 0.85) that did not map onto PERMA and SWB. SWB and PERMA factors showed similar relationships with 24 character strengths (average correlation difference = 0.02). Latent profile analyses yielded subgroups of people who merely scored high, low, or mid-range on well-being indicators. Our findings suggest that while lower-order indicators SWB and PERMA have unique features, they converge onto a single well-being factor.

Keywords:

• Well-being
• happiness
• flourish
• character strengths
• positive psychology

Notes

1. Ryff’s (1989) Scales of Psychological Well-being is most often used to operationalize eudaimonic well-being, but the theoretical and conceptual ideas stem from Jahoda (1958). Jahoda proposed the inclusion of positive states in defining and measuring mental health. Her model of ‘Ideal mental health’ includes six components (that have a great degree of overlap with Ryff’s PWB model): efficient self-perception, realistic self-esteem and acceptance, voluntary control of behavior, true perception of the world, sustaining relationships and giving affection, and self-direction and productivity.

2. As Long et al. (2016) described in their paper, they originally found a latent correlation of 0.76. While at first glance this correlation suggests some (albeit weak) discrimination, Longo et al. (2016) found that the correlation was attenuated by method variance: most of the positive feelings items were normally coded and most of the positive functioning items were reverse coded. Negatively worded items have been shown to load onto their own factors for purely methodological reasons (Marsh, 1996; Woods, 2006). After correcting for the method variance by replicating the study with all normal coded items, the latent correlation was 0.97, suggesting negligible discriminant validity between the two constructs of well-being.

3. Finite mixture models such as LPA are prone to converging on local maxima of the log likelihood function rather than the global maxima. To increase confidence that the expectation-maximization algorithm has converged onto the global maxima, it is recommended that researchers do the analysis 50 times or more with unique starting values and determine if the algorithm converged on the same log likelihood function. If yes, then there is some support (although not definite proof) that the algorithm has found the global maxima. If not, the results are suspect and should be interpreted with caution.

4. When conducting the factor analyses at the item level, results were very similar. We used polychoric correlations with the weighted least squares estimator with a scale-shifted chi-square value (WLSMV in Mplus). We created a bifactor measurement model for SWB and bifactor measurement model for PERMA. The two general factors had a latent correlation of 0.98 (CFI = 0.98, TLI = 0.97, RMSEA = 0.08).

5. Inclusion of an error between happiness and positive emotions (r = 0.20) barely changed the results. The factor loadings for happiness and positive emotions were slightly less and the latent correlation was 0.95.

6. Results were nearly identical when the alternative two-factor CFA model was used.

7. Results for the two and three-profile models were nearly identical when happiness was included. Four and five-profile models did not converge.