In 1980, Costa and McCrae proposed an influential model of well-being. Their seminal article has 1,400 citations in WebofScience so far.

The model assumed that personality traits are stable (during adulthood) and that stable internal dispositions are the major determinant of wellbeing.

The model has led to the development of a two-dimensional model of affect that assumes positive affect and negative affect are independent dimensions (Watson, Tellegen, & Clark, 1980). Accordingly, wellbeing is characterized by high PA and low NA. The model also implies that happy people are extroverted and emotionally stable (low neuroticism).

This model had a strong influence on wellbeing research in psychology for two decades (see Diener et al., 1999, Psych Bull, for a review).

The model implies that (a) life-satisfaction is stable over time and (b) personality traits should account for most of the stable variance in life-satisfaction. This follows directly from Costa and McCrae’s (1980) assumption that personality effects are stable, while environmental factors change over time.

“Few would argue against the position that, for normal people, the major determinant of
momentary happiness is the specific situation in which the individual finds himself or herself. Social slights hurt our feelings, toothaches make us miserable, compliments raise
our spirits, eating a good meal leaves us satisfied. The contribution of personality to any
one of these feelings is doubtless small. Yet over time, the small but persistent effects of
traits emerge as a systematic source of variation in happiness, whereas situational determinants that vary more or less randomly tend to cancel each other out (cf. Epstein, 1977).” (Costa and McCrae, 1980, p. 676).

Since 2000, well-being researchers have revised their views about the major determinants of well-being. Thanks to large panel studies (longitudinal studies with repeated measurements) like the SOEP it became clear that environmental influences are more important than Costa and McCrae (1980) suggested. After separating variance in life-satisfaction into a stable component and a changing component, changing factors still had an annual stability around .9 (Lucas & Donnellan, 2007; Schimmack, Schupp & Wagner, 2008; Schimmack & Lucas, 2010). For example, unemployment has been shown to influence well-being and individuals who are unemployed are more likely to remain unemployed in the next year. Similarly, income is highly stable from year to year.

To complicate maters further, there is some controversy about the stability of personality traits and it has been suggested that changes in well-being may produce personality changes (Scollon & Diener, 2006). Studies of the stability of personality and life-satisfaction partially resolve this question (Conley, 1984; Anusic & Schimmack, 2016). These studies show that personality is more stable than life-satisfaction judgments. This finding suggests that personality influences wellbeing. The reason is that a causal influence implies that changes in one variable produce changes in another variable. Thus, if life-satisfaction changes and personality remains stable, life-satisfaction cannot be a cause of personality. However, this line of reasoning is indirect. A better test of the relationship between personality and life-satisfaction would require longitudinal studies that measure personality and life-satisfaction over time. Until recently, these kind of studies were largely absent. Here I use the SOEP and MIDUS data to examine concurrent changes in personality and life-satisfaction.

SOEP

Data

The SOEP measured personality in 2005, 2009, and 2013 with the 15-item BFI-S. I developed a measurement model for these items (Schimmack, 2019). Here, the same measurement model was fitted the the data from all three waves, while imposing metric invariance. In addition, the 11-point life-satisfaction item was included as an indicator of well-being.

Model

Longitudinal stability and change was modeled with Heise’s (1969) autoregressive model that separates variance into an occassion-specific error component and a state component. The state component changes gradually over time. This model does not include a trait component because four measurements are needed to identify the influence of stable factors. However, over shorter time intervals of a decade, an autoregressive model can approximate a trait model.

The personality model has seven factors. Five factors represent the Big Five personality traits. The other two traits reflect acquiescence bias and halo bias. Unfortunately, there are model identification problems when life-satisfaction judgments are regressed on all seven factors. Previous studies have shown that openness is a very weak predictor of life-satisfaction, while halo bias predicts life-satisfaction ratings (Kim, Schimmack, & Oishi, 2012). Thus, the relationship between openness and life-satisfaction was fixed to zero, while the relationship for halo was estimated freely.

Life-satisfaction at all time-points was regressed on neuroticism (N), extraversion (E), agreeableness (A), conscientiousnes (C), and halo (H). Costa and McCrae’s (1980) model predicts that extraversion and neuroticism account for the lion share of the correlation between personality and life-satisfaction concurrently and over time. McCrae and Costa (1991) also found that agreeableness and conscientiousness added to the prediction of life-satisfaction. Thus, these personality traits should also predict life-satisfaction concurrently and over time.

The novel question is whether changes in personality also predict changes in life-satisfaction. To examine this question, the residual variances at time 2 and time 3 were allowed to predict life-satisfaction at time 2 and time 3, respectively.

The second novel question was whether personality accounts for all of the stability in life-satisfaction. If this were the case, the residuals of life-satisfaction should be fairly independent over time. That is, the variance in life-satisfaction that is not explained by personality at time 1 should not predict variance in life-satisfaction at time 2.

The syntax for this (somewhat complex) model and the complete results can be found on OSF ( https://osf.io/vpcfd/). The model fit was acceptable, CFI = .964, RMSEA = .022, SRMR = .031.

Results

Table 1 shows the reliability and stability estimates.

	Reliability	Stability	1Y-Stability
Life Satisfaction	0.61	0.80	0.95
Neuroticism	0.85	0.92	0.98
Extraversion	0.86	0.93	0.98
Openness	0.88	0.92	0.98
Agreeableness	0.77	0.93	0.98
Conscientiousness	0.88	0.77	0.94
Halo	0.83	0.85	0.96
Acquiescence	0.45	0.85	0.96

The reliability estimate for the life-satisfaction ratings is consistent with estimates based on extensive multi-wave models (Lucas & Donnellan, 2007; Schimmack et al., 2008). The reliabilty estimates for personality are higher because they are based on three item measures and a latent-variable model that reduces measurement error. The results also show that stability of life-satisfaction is lower than stability of personality traits, although the difference is small and conscientiousness is a notable exception.

Table 2 shows the effect sizes for the relationships of personality at time 1 with life-satisfaction at times 1 to 3. The pattern of relationships is consistent with previous studies. Neuroticism is the strongest Big Five predictor of life-satisfaction, with about 10% explained variance. Extraversion is a significant predictor, but only explains about 1% of the variance in life-satisfaction. Effects for agreeableness and conscientiousness are weak. Halo is a notable predictor that explains another 10% of the variance.

The second important finding is that personality traits measured at time 1 predict life-satisfaction with equal strength across the three time points covering eight years. This confirms the hypothesis that personality accounts for stability in life-satisfaction. However, all personality measures combined account for less than a quarter of the variance in life-satisfaction. Given the much higher stability of life-satisfaction (see Table 1), the Big Five are not sufficient to explain stability in life-satisfaction.

	LS-T1	LS-T2	LS-T3
Neuroticism	-0.33	-0.32	-0.35
Extraversion	0.11	0.13	0.11
Openness	–	–	–
Agreeableness	0.08	0.13	0.09
Conscientiousness	0.04	0.03	0.05
Halo	0.29	0.27	0.30
Acquiescence	–	–	–

Table 3 shows the relationship between residual variances in personality and life-satisfaction at times 2 and 3. These results show whether changes in personality predict changes in well-being. The coefficients in Table 3 cannot be directly compared to those in Table 2 because they are standardized coefficients and the residual variance in personality is much smaller than the stable variances. However, the results do provide seminal information whether changes in personality can predict changes in life-satisfaction.

	LS-T2	LS-T3
Neuroticism	-0.09	-0.20
Extraversion	0.02	0.13
Openness	–	–
Agreeableness	-0.17	-0.05
Conscientiousness	0.03	0.05
Halo	0.27	0.30
Acquiescence	–	–

Results for neuroticism and extraversion trend in the right direction, but weak at time 2. Results for agreeableness would point towards a negative effect, which is difficult to interpret, and effects for conscientiousness are weak at time 2 and 3. The strongest effect was found for halo. When halo bias changes, life-satisfaction ratings change in the same direction. This is consistent with an evaluative bias effect. Overall, these results are consistent with the view that personality is stable and mostly accounts for stable variance in life-satisfaction judgments.

Conclusion

Forty years ago, Costa and McCrae (1980) proposed that well-being is influenced by stable personality dispositions and that these stable dispositions account for stability in well-being. Fourty years later, we can evaluate their theory with new and better data. The results confirm the prediction that neuroticism is a stable personality disposition that produces lasting individual differences in well-being. The results for extraversion are also consistent with the theory, although the effect size is weaker and extraversion accounts for only a small portion of variance in life-satisfaction. The addition of agreeableness and conscientiousness as predictors of well-being is less supported by the data.

The most important limitation of Costa and McCrae’s model is that neuroticism and extraversion explain only a portion of stability in life-satisfaction. Life-satisfaction shows additional stability that is not explained by personality, or at least the Big Five. An important avenue for future research is to find additional predictors of stable variance in life-satisfaction. Costa and McCrae and subsequent researchers may have understimated the stabiltity and importance of enviromental influences on well-being (Schimmack & Lucas, 2010).