This is part 4 in a mini-series of blogs that illustrate the usefulness of structural equation modeling to test causal models of well-being. The first causal model of well-being was introduced in 1980 by Costa and McCrae. Although hundreds of studies have examined correlates of well-being since then, hardly any progress has been made in theory development. In 1984, Diener (1984) distinguished between top-down and bottom-up theories of well-being, but empirical tests of the different models have not settled this issue. The monster model is a first attempt to develop a causal model of well-being that corrects for measurement error and fits empirical data.
The first part (Part1) introduced the measurement of well-being and the relationship between affect and well-being. The second part added measures of satisfaction with life-domains (Part 2). Part 2 ended with the finding that most of the variance in global life-satisfaction judgments is based on evaluations of important life domains. Satisfaction in important life domains also influences the amount of happiness and sadness individuals experience, whereas positive affect had no direct effect on life-evaluations. In contrast, sadness had a unique negative effect on life-evaluations that was not mediated by life domains.
Part 3 added extraversion to the model. This was a first step towards a test of Costa and McCrae’s assumption that extraversion has a direct effect on positive affect (happiness) and no effect on negative affect (sadness). Without life domains in the model, the results replicated Costa and McCrae’s (1980) results. Yes, personality psychology has replicable findings. However, when domain satisfactions were added to the model, the story changed. Costa and McCrae (1980) assumed that extraversion increases well-being because it has a direct effect on cheerfulness (positive affect) that adds to well-being. However, in the new model, the effect of extraversion on life-satisfaction was mediated by life domains rather than positive affect. The strongest mediation was found for romantic satisfaction. Extraverts tended to have higher romantic satisfaction and romantic satisfaction contributed significantly to overall life-satisfaction. Other domains like recreation and work are also possible mediators, but the sample size was too small to produce more conclusive evidence.
Part 4 is a simple extension of the model in part 3 by adding the other personality dimensions to the model. I start with neuroticism because it is by far the most consistent and strongest predictor of well-being. Costa and McCrae (1980) assumed that neuroticism is a general disposition to experience more negative affect without any relation to positive affect. However, most studies show that neuroticism has a negative relationship with positive aspect as well, although it is not as strong as the relationship with negative affect. Moreover, neuroticism is also related to lower satisfaction in many life domains. Thus, the model simply allowed for neuroticism to be a predictor of both affects and all domain satisfaction. The only assumption made by this model is that the negative effect of neuroticism on life-satisfaction is fully mediated by domain satisfaction and affect.
Figure 1 shows the model and the path coefficients for neuroticism. The first important finding is that neuroticism has a strong direct effect on sadness that is independent of satisfaction with various life domains. This finding suggests that neuroticism may have a direct effect on individuals’ mood rather than interacting with situational factors that are unique to individual life domains. The second finding is that neuroticism has sizeable effects on all life domains ranging from b = -.19 for satisfaction with housing to -31 for satisfaction with friendships.
Following the various paths from neuroticism to life-satisfaction produces a total effect of b = -.38, which confirms the strong negative effect of neuroticism on well-being. About a quarter of this effect is directly mediated by negative affect (sadness), b = -.09. The rest is mediated by the top-down effect of neuroticism on satisfaction with life domains and the bottom-up effect of life domains on global life-evaluations.
McCrae and Costa (1991) expanded their model to include the other Big Five factors. They proposed that agreeableness has a positive influence on well-being that is mediated by romantic satisfaction (adding Liebe) and that conscientiousness has a positive influence on well-being that is mediated by work satisfaction (adding Arbeit). Although this proposal was made three decades ago, it has never been seriously tested because few studies measure domain satisfaction (but see Heller et al., 2004).
To test these hypotheses, I added conscientiousness and agreeableness to the model. Adding both together was necessary because agreeableness and conscientiousness were correlated as reflected in a large modification index when the two factors were assumed to be independent. This does not mean that agreeableness and conscientiousness are correlated factors, an issue that is debated among personality psychologists (Anusic et al., 2009; Biesanz & West, 2004; DeYoung, 2006). One problem is that secondary loadings can produce spurious correlations among scale scores that were used for this model. This could be examined by using a more complex item-level model in the future. For now, agreeableness and conscientiousness were allowed to correlate. The results showed no direct effects of conscientiousness on PA, NA, and LS. In contrast, agreeableness was a positive predictor of PA and a negative predictor of NA. Most important are the relationships with domain satisfactions.
Confirming McCrae and Costa’s (1991) prediction, work satisfaction was predicted by conscientiousness, b = .21, z = 3.4. Also confirming McCrae and Costa, romantic satisfaction was predicted by agreeableness, although the effect size was small, b = .13, z = 2.9. Moreover, conscientiousness was an even stronger predictor, b =.28, z = 6.0. This confirms the old saying “marriage is work.” Also not predicted by McCrae and Costa was that conscientiousness is related to higher housing satisfaction, b = .20, z = 3.7, presumably because conscientious individuals take better care of their houses. The other domains were not significantly related to conscientiousness, |b| < .1.
Also not predicted by McCrae and Costa are additional relationships of agreeableness with other domains such as health, b = .18, z = 3.7, housing, a = .17, z = 2.9, recreation, b = .25, z = 4.0, and friendships, b = .35, z = 5.9. The only domains that were not predicted by agreeableness were financial satisfaction, b = .05, z = 0.8, and work satisfaction, b = .07, z = 1.3. Some of these relationships could reflects benefits for social relationships aside from romantic relationships. Thus, the results are broadly consistent with McCrae and Costa’s assumption that agreeableness is beneficial for well-being.
The total effect of agreeableness in this dataset was b = .21, z = 4.34. All of this effect was mediated by indirect paths, but only the path through romantic satisfaction achieved statistical significance due to a lack of power, b = .03, z = 2.6.
The total effect of conscientiousness was b = .18, z = 4.14. Three indirect paths were significant, namely work, b = .06, z = 3.3. romantic satisfaction, b = .06, z = 4.2, and housing satisfaction, b = .04, z = 2.51.
Overall, these results confirm previous findings that agreeableness and conscientiousness are also positive predictors of well-being and shed first evidence on potential mediators of these relationships. These results need to be replicated in datasets from other populations.
When openness was added to the model, a modification index suggested a correlation between extraversion and openness, which has been found in several multi-method studies (Anusic et al., 2009; DeYoung, 2006). Thus, the two factors were allowed to correlate. Openness had no direct effects on positive affect, negative affect, or life-satisfaction. Moreover, there were only two, weak, just significant relationships with domain satisfaction for work, b = .12, z = 2.0, and health, b = .12, z = 2.2. Consistent with meta-analysis, the total effect is negligible, b = .06, z = 1.3. In short, the results are consistent with previous studies and show that openness is not a predictor of higher or lower well-being. To keep the model simple, it is therefore possible to omit openness from the monster model.
At this point, we have built a complex, but plausible model that links personality traits to subjective well-being by means of domain satisfaction and affect. However, just because this model is plausible and fits the data, does not ensure that it is the right model. An important step in causal modeling is to consider alternative models and to do model comparisons. Overall fit is less important than relatively better fit among alternative models.
The previous model assumed that domain satisfaction causes higher levels of PA and lower levels of NA. Accordingly, affect is a summary of the affect generated in different life domains. This assumption is consistent with bottom-up models of well-being. However, a plausible alternative model assumes that affect is largely influenced by internal dispositions which in turn color our experiences of different life domains. Accordingly neuroticism may simply be a disposition to be more often in a negative mood and this negative mood colors perception of marital satisfaction, job satisfaction, and so on. Costa and McCrae (1980) proposed that neuroticism and extraversion are global affective dispositions. So, it makes sense to postulate that their influence on domain satisfaction and life satisfaction is mediated by affect. McCrae and Costa (1991) postulated that agreeableness and conscientiousness are not affective dispositions, but rather only instrumental for higher satisfaction in some life domains. Thus, their effects should not be mediated by affect. Consistent with this assumption, conscientiousness showed only significant relationships with some domains, including work satisfaction. However, agreeableness was a positive predictor of all life domains, suggesting that it is also a broad affective disposition. I thus modeled agreeableness as a third global affective disposition (see Figure 2).
The effect sizes for affect on domain satisfaction are shown in Table 1.
A comparison of the fit indices for the top-down and bottom-up models shows that both models meet standard criteria for global model fit (CFI > .95; RMSEA < .06). In addition, the results show no clear superiority of one model over the other. CFI and RMSEA show slightly better fit for the bottom-up model, but the Bayesian Information Criterion favors the more parsimonious top-down model. Thus, the data are unable to distinguish between the two models.
Both model assume that conscientiousness is instrumental for higher well-being in only some domains. The key difference between the models is the assumption of the top-down model that changes in domain satisfaction have no influence on affective experiences. That is, an increase in relationship satisfaction does not produce higher levels of PA or a decrease in job satisfaction does not produce a change in NA. These competing predictions can be tested in longitudinal studies.
To conclude part 4 of the monster model series. As surprising as it may sound, the present results provide one of the first tests of McCrae and Costa’s causal theory of well-being (Costa & McCrae, 1980, McCrae & Costa, 1991). Although the present results are consistent with their proposal that agreeableness and conscientiousness are instrumental for higher well-being because they foster higher romantic and job satisfaction, respectively, the present results also show that this model is too simplistic. For example, conscientiousness may also increase well-being because it contributes to higher romantic satisfaction (marriage is work).
One limitation of the present model is the focus on the Big Five as a measure of personality traits. The Big Five are higher-order personality traits of more specific personality traits that are often called facets. Facet level traits may predict additional variance in well-being that is not captured by the Big Five (Schimmack ,Oishi, Furr, & Funder, 2004). Part 5 will add the strongest facet predictors to the model, namely the Depressiveness facet of Neuroticism and the Cheerfulness facet of Extraversion (see also Payne & Schimmack, 2020).