The Structure of Affective Dispositions

Extraversion and Neuroticism are some of the oldest constructs in personality psychology. They were the key dimensions in Eysenck’s theory of personality that was prominent in the 1970s. Although Eysenck’s theory of Neuroticism and Extraversion failed to be supported, extraversion and neuroticism remained prominent dimensions in the Big Five model of personality that emerged in the 1980s.

In an influential article, Costa and McCrae (1980) reconceptualized Extraversion and Neuroticism as two broad affective disposition that influences positive and negative affective experiences. They found that extraversion predicted positive affect and neuroticism predicted negative affect on Bradburn’s affect measure.

A key assumption of this model is that the disposition to experience negative affects and the disposition to experience positive affects are largely independent traits. This model underlies the development of the popular Positive Affect and Negative Affect Schedule (PANAS) that is widely used to measure affective experiences over longer time periods (Watson, Tellegen, & Clark, 1988).

The independence model of PA and NA has created a heated controversy in the emotion literature (see JPSP special issue 1999). Most of the debate focussed on the structure of momentary affective experiences. However, some articles also questioned the independence of positive and negative affective dispositions. Specifically, Diener, Smith, & Fujita (1995) used a mutli-method approach to measure a variety of positive and negative affective traits. They did find separate factors for PA and NA, but these factors were strongly negatively correlated (see Zou, Schimmack, & Gere, 2013, for a conceptual replication). Findings like these suggest that the independence model is too simplistic.

There are several ways to reconcile the negative relationship between positive and negative affects with the independence model. First, it is possible that the relationship between PA and NA depends on the selection of specific affects. Whereas happiness and sadness/depression may be negatively correlated, excitement and anxiety may be independent. If the correlation varies for specific affects, it is necessary to use proper statistical methods like CFA to examine the relationship between PA and NA without the influence of variance due to specific emotions. An alternative approach would be to directly measure the valence of emotions. Few studies have used this approach to remove emotion-specific variance from studies of PA and NA.

Another possibility is that PA and NA are not as strongly aligned with Extraversion and Neuroticism as Costa and McCrae’s (1980) model suggest. In fact, Costa and McCrae also developed a model in which positive affect was merely one of several traits called facets that are related to extraversion. According to the facet model, extraversion is a broader trait that encompasses affective and non-affective dispositions. For example, extraversion is also related to behaviours in social situations (sociability, assertiveness) and situations with uncertainty (risk taking). One implication of this model is that Extraversion and Neuroticism could be independent, while PA and NA can be negatively correlated.

The relationship between Extraversion and Neuroticism has been examined in hundreds of studies that measured the Big Five. Simple correlations between Extraversion and Neuroticism scales typically show small to moderate negative correlations. This finding contradicts the assumption that E and N are independent, but this finding has often been ignored. For example, structural models that allow for correlations between the Big Five maintain that E and N are independent (DeYoung, 2015).

One explanation for the negative correlation between E and N are response styles. Extraversion items tend to be desirable, whereas Neuroticism items tend to be undesirable. Thus, socially desirable responding can produce spurious correlations between E and N measures. In support of this hypothesis, the correlation weakens and sometimes disappears in multi-rater studies (Anusic et al., 2009; Biesanz & West, 2004; DeYoung, 2006). However, the correlation between E and N also depends on the item content. Scales that focus on sociability and anxiety tend to find weaker correlations than scales that measure E and N with a broader range of facets like the NEO-PI. Once more, this means that scale content moderates the results and that proper analyses of the relationship between the higher-order factors E and N requires a hierarchical CFA model to remove facet-specific variance from the correlation.

The aim of this blog post is to examine the structure of Extraversion and Neuroticism facets with a hierarchical CFA. A CFA model can reveal aspects of the data that a traditional EFA cannot reveal. Most importantly, it can reveal relationships between facets that are independent of the higher-order factors E and N. These residual correlations are important aspects of the relationship between traits that have been neglected in theoretical models based on EFA because EFA does not allow for these relationships.

Data

Over the past decade, Condon and Revelle have assembled an impressive data set from over 50,000 participants who provided self-ratings of their personality for subsets of over 600 personality items that cover a broad range of personality traits at the facet level. Modern statistical methods make it possible to analyze these data with random missing data to examine the structure of all 600 personality items. The authors generously made their data openly available. I used the datasets that represent data collected between 2013 and 2014 and 2014 to 2015 (https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/SD7SVE). I did not use all of the data to allow cross-validation of the results with a new sample.

Even modern computers would take too long to analyze the structure of over 600 items. For the present purpose, I focussed on items that have been shown to be valid indicators of extraversion and neuroticism facets (Schimmack, 2020a, 2020b). The actual items and their primary factor loadings are shown below in Tables 1-3.

Results

Preliminary analyses showed problems with model identification because some Neuroticism and Extraversion scales were strongly negatively related. Specifically, E-Boldness was strongly negatively related to N-Self-Consciousness, E-Happiness was strongly negatively correlated with N-Depression, and E-Excitement Seeking was strongly negatively related to N-Fear. These findings already show that E and N are not independent domains that fit a simple structure; that is, E facets are not related to N-facets. To accommodate these preliminary findings, I created three bipolar facet factors. I then fitted a measurement model for 4 N-facets, 6 E-facets, and the three bipolar facets. The measurement model allowed for secondary loadings and correlated item residuals based on modification indices. All primary factors were allowed to correlate freely. In addition, the model included a method factor for acquiescence bias with fixed loadings depending on the scoring of items. As this model was data-driven, the results are exploratory and require cross-validation in a new sample.

Model fit of the final model met standard fit indices for overall model fit (CFI > .95, RMSEA < .06), CFI = .951, RMSEA = .006. However, standard fit indices have to be interpreted with caution for models with many missing values (Zhang & Savaley, 2019). More important, modification indices suggested no major further improvements to the measurement model by allowing additional secondary loadings. It is also known that minor misspecifications of the measurement model have relatively little influence on the theoretically important correlations among the primary factors. Thus, results are likely to be robust across different specifications of the measurement model. Table 1 shows the items and their primary factor loadings for the Neuroticism facets.

Table 2 shows the items and their primary loadings for the Extraversion facets.

Table 3 shows the items for the three bipolar Extraversion-Neuroticism factors.

Table 4 shows the correlations among the 13 Extraversion and Neuroticism facets.

All correlations among the N-facets are positive and above r = .3. All of the correlations among the E-facets are positive and only two are below .30. Two of the bipolar facets, namely Happiness-Depression and Boldness-Self-Consciousness are negatively correlated with neuroticism facets (all r < -.3). Most of the correlations with the extraversion facets are positive and above .3, but some are smaller and one is practically zero (r = -.03). Surprisingly, the Excitement Seeking versus Fear facet has very few notable relationships with neuroticism and extraversion facets. This suggests that this dimension is not central to either domains.

The correlations between extraversion facets and neuroticism facets tend to be mostly negative, but most of them are fairly small. This suggests that Extraversion and Neuroticism are largely independent and relatively weakly correlated factors. The only notable exceptions were negative correlations of anxiety with novelty seeking and liveliness.

The aim of any structural theory of personality is to explain this complex pattern of relationships in Table 4. Although a model with two factors is viable, a model with a simple structure that assigns each facet to only one factor does not fit these data. To account for the actual structure it is necessary to allow for some facets to be related to extraversion and neuroticism and to allow for additional relationship between some facets. Allowing for these additional relationships produced a model that fit the data nearly as well as the measurement model: CFI = .938 vs. .951, RMSEA = .007 vs. .006. The results are depicted in Figure 1.

Figure 1 does not show the results for the excitment-seeking vs. fear factor which was only weakly related to E and N, but strongly related to Novelty-Seeking and Boldness. To accommodate this factor, the model included direct paths from Anxiety (-.55), Anger (.33), Happiness-Depression (-.58), Novelty Seeking (.45), and liveliness (.24). The strong positive relationship with Happiness-Depression is particularly noteworthy. It could mean that depression or related negative affects like boredom motivate people to seek excitement. However, these results are preliminary and require further investigation.

The key finding is that Extraversion and Neuroticism emerge as slightly negatively correlated factors. The negative correlation in this study could be partially due to evaluative biases in self-ratings. Thus, the results are consistent with the conceptualization of Extraversion and Neuroticism as largely independent higher-order factors of personality. However, this does not mean that affective dispositions are largely independent.

The happiness factor lacked discriminant validity from the depression factor, showing a strong negative relationship between these two affective traits. Moreover, the happiness-depression factor was related to anger and anxiety because it was related to neuroticism. Thus, high levels of neuroticism not only increase NA, they also lower happiness.

The results also explain the independence of PA and NA in the PANAS scales. The PANAS scales were not developed to measure basic affects like happiness, sadness, fear and anxiety. Instead, they were created to measure affect with two independent traits. While the NA dimension closely corresponds to neuroticism, the PA dimension corresponds more closely to positive activation or positive energy than to happiness. The PANAS-PA construct of Positive Activation is more closely aligned with the liveliness factor. As shown in Figure 1, liveliness loads on Extraversion and is fairly independent of negative affects. It is only related to anxiety and anger through the small correlation between E and N. For depression, it has an additional relationship because liveliness and depression load on Extraversion. It is therefore important to make a clear conceptual distinction between Positive Affect (Happiness) and Positive Activation (Liveliness).

Figure 1 also shows a number of correlated residuals that were needed to achieve model fit. These correlated residuals are by no means arbitrary. Activity is related to being lively, presumably because energy is required to be active. Amusement is related to sociability, presumably because humor helps to establish and maintain positive relationships. Boldness is related to assertiveness because both traits require a dominant role in social relationships and groups. Anxiety is negatively related to boldness because bold behaviours are risky. Moody is related to anger and depression, presumably because mood swings can be produced by either anger or depressive episodes. Although these relationships are meaningful they are often ignored because EFA fails to show these relationships and fails to show that models without these relationships do not fit the data. The present results show that theoretical progress requires developing models that explain these relationships. In this regard, the present results merely show that these relationships exist without explaining them.

It is also noteworthy that the correlated residuals do not show a simple pattern that is postulated by some theories. Most notably, DeYoung (2015) proposed that facets are linked to Big Five factors by means of aspects. Aspects are supposed to represent shared variance among some facets that is not explained by the Big Five traits. A simple way to examine the presence of aspects is to find groups of facets that share correlated residuals. Contrary to the aspect model, most facets have either one or no correlated residuals. Sociability has two correlated residuals. It is related to amusement and dependence, but amusement and dependence are not related. Thus, there is no aspect linking these three facets. Moody is related to anger and depression, but anger and depression are unrelated. Again, this implies that there is no aspect linking these three facets. Boldness is linked to three facets. It is positively related to assertiveness, but negatively related to anxiety and dependence, but anxiety and dependence are unrelated, and assertiveness is only related to boldness. This means that there is no evidence for DeYoung’s Extraversion and Neuroticism aspects. These results are by no means inconsistent with previous findings. The aspect model was developed with EFA and EFA may separate a facet from other facets and create the illusion of aspects. This is the first test of the aspects model with CFA and it shows no support for the model.

Conclusion

In conclusion, the present study examined the structure of affective traits using hierarchical CFA. The results broadly confirm the Big Five model of personality. Neuroticism represents shared variance among several negative affective traits like anxiety, anger, and depression, and self-conscious emotions. Extraversion is a broader trait that includes affective and non-affective traits. The core affective traits are happiness and positive energy (liveliness). Extraversion and Neuroticism are only slightly negatively correlated and this correlation could be inflated by rating biases. Thus, it is reasonable to conceptualize them as largely independent higher-order traits. However, at the facet level, the structure is more complex and does not fit a simple structure. Some E-facets and N-facets are highly negatively correlated and could be conceptualized as opposite ends of a single trait, namely Happiness-Depression, Boldness-Self-Consciousness, and Excitement-Seeking vs. Fear. It is therefore questionable to classify Happiness, Boldness, and Excitement-Seeking under Extraversion and Depression, Self-Consciousness and Fear under Neuroticism. These traits are related to Extraversion and Neuroticism. The present results do not provide explanations for the structure of affective trait. The main contribution is to provide a description of the structure that actually represents the structure in the data. In contrast, many prominent models are overly simplistic, focus on subsets of facets, and do not fit the data. The present results integrate these models into one general model that can stimulate future research.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s