The construct of neuroticism is older than psychological science. It has its roots in Freud’s theories of mental illnesses. Thanks to to influence of psychoanalysis on the thinking of psychologists, the first personality questionnaires included measures of neuroticism or anxiety, which were considered to be highly related or even identical constructs.

Eysenck’s research on personality first focussed on Neuroticism and Extraversion as the key dimensions of personality traits. He then added psychoticism as a third dimension.

In the 1980s, personality psychologists agreed on a model with five major dimensions that included neuroticism and extraversion as prominent dimensions. Psychoticism was divided into agreeableness and conscientiousness and a fifth dimension openness was added to the model.

Today, the Big Five model dominates personality psychology and many personality questionnaires focus on the measurement of the Big Five.

Despite the long history of research on neuroticism, the actual meaning of the term and the construct that is being measured by neuroticism scales is still unclear. Some researchers see neuroticism as a general disposition to experience a broad range of negative emotions. In the emotion literature, anxiety, anger, and sadness are often considered to be basic negative emotions, and the prominent NEO-PI questionnaires considers neuroticism to be a general disposition to experience these three basic emotions more intensely and frequently.

Neuroticism has also been linked to more variability in mood states, higher levels of self-consciousness and lower self-esteem.

According to this view of neuroticism, it is important to distinguish between neuroticism as a more general disposition to experience negative feelings and anxiety, which is only one of several negative feelings.

A simple model of neuroticism would assume that a general disposition to respond more strongly to negative emotions produces correlations among more specific dispositions to experience more anxiety, anger, sadness, and self-conscious emotions like embarrassment. This model implies a hierarchical structure with neuroticism as a higher-order factor of more specific negative dispositions.

In the early 2000s, Ashton and Lee published an alternative model of personality with six factors called the HEXACO model. The key difference between the Big Five model and the HEXACO model is the conceptualization of pro- and anti-social traits. While these traits are considered to be related to a single higher-order factor of agreeableness in the Big Five model, the HEXACO model distinguishes between agreeableness and honesty-humility as two distinct traits. However, this is not the only difference between the two models. Another important difference is the conceptualization of affective dispositions. The HEXACO model does not have a factor corresponding to neuroticism. Instead it has an emotionality factor. The only common trait to neuroticism and emotionality is anxiety, which is measured with similar items in Big Five questionnaires and in HEXACO questionnaires. The other three traits linked to emotionality are unique to the HEXACO model.

The four primary factors, also called facets) that are used to identify and measure emotionality are anxiety, fear, dependence, and sentimentality. Fear is distinguished from anxiety by a focus on immediate and often physical danger. In contrast, anxiety and worry tend to be elicited by thoughts about uncertain events in the future. Dependence is defined by a need for social comfort in difficult times. Sentimentality is a disposition to respond strongly to negative events that happen to other people, including fictional characters.

In a recent target article, Ashton and Lee argued that it is time to replace the Big Five model with the superior HEXACO model. A change from neuroticism to emotionality would be a dramatic shift given the prominence of neuroticism in the history of personality psychology. Here, I examine empirically how Emotionality is related to Neuroticism and whether personality psychologists should adapt the HEXACO framework to understand individual differences in affective dispositions.

Data

A key problem in research on the structure of personality is that researchers often rely on questionnaires that were developed with a specific structure in mind. As a result, the structure is pre-determined by the selection of items and constructs. To overcome this problem, it is necessary to sample a broad and ideally representative sample of primary traits. The next problem is that motivation and attention-span of participants limits the number of items that a personality questionnaire can include. These problems have been resolved by Revelle and colleagues survey that asks participants to complete only a subset of over 600 items. Modern statistical methods can analyze datasets with planned missing data. Thus, it is possible to examine the structure of hundreds of personality items. Condon and Revelle (2018) also made these data openly available (https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/SD7SVE). I am very grateful for their initiative that provides an unprecedented opportunity to examine the structure of personality.

The items were picked to represent primary factors (facets) of the HEXACO questionnaire and the NEO-PI questionnaire. In addition, the questionnaire covered neuroticism items from the EPQ and other questionnaires. The items are based on the IPIP item pool. Each primary factor is represented by 10 items. I picked the items that represent the four HEXACO Emotionality factors, anxiety, fear, dependency, sentimentality, and four of the NEO-PI Neuroticism factors, anxiety, anger, depression, and self-consciousness. The anxiety factor overlaps and is represented by mostly overlapping items. Thus, this item selection resulted in 70 items that were intended to measure 7 primary factors. I added four additional items that represented variable moods (moodiness) that were included in the BFAS and EPQ, which might form an independent factor.

Analysis

The data were analyzed with confirmatory factor analysis (CFA), using the MPLUS software. CFA has several advantages over traditional factor analytic methods that have been employed by proponents of the HEXACO and the Big Five models. The main advantages are that it is possible to model hierarchical structures that represent the Big Five or HEXACO factors as higher-order factors of primary factors. A second advantage is that CFA provides information about model fit whereas traditional EFA produces solutions without evaluating model fit.

Measurement Model

A first step in establishing a measurement model was to select items with high primary loadings, low secondary loadings, and low correlated residuals. The aim was to represent each primary factor with the best four items. While four items may not be enough to create a good scale, four items are sufficient to establish a measurement model of primary factors. Limiting the number of items to four items is also advantages because computing time increases with additional items and models with missing data can take a long time to converge.

Aside from primary loadings, the model included an acquiescence factor based on the coding of items. Directed coded items had unstandardized loadings of 1 and reverse coded items had an unstandardized loading of -1. There were no secondary loadings or correlated residuals.

The model met standard criteria of model fit such as a CFI > .95 and RMSEA < .05, CFI = .954, RMSEA = .007. However, models with missing data should not be evaluated based on these fit indices because fit is determined by a different formula ( (Zhang & Savaley, 2019).  More importantly, modification indices showed no notable changes in model fit if fixed parameters were freed. Table 1 shows the items and their primary factor loadings.

Table 2 shows the correlations among the primary factors.

The first four factors are assumed to belong to the HEXACO-Emotionality factor. As expected, fear, anxiety, and dependence are moderately to highly positively correlated. Contrary to expectations, sentimentality showed low correlations especially with fear.

Factors 4 to 8 are assumed to be related to Big Five neuroticism. As expected, all of these factors are moderately to highly correlated.

In addition, the dependence factor from the HEXACO model also shows moderate to high correlations with all Big Five neuroticism factors. The fear factor also shows positive relations with the neuroticism factors, especially for self-consciousness.

With the exception of Sentimentality, all of the factors tend to be positively correlated, suggesting that they are related to a common higher-order factor.

Overall, this pattern of results provides little support for the notion that HEXACO-Emotionality is a distinct higher-order factor from Big-Five neuroticism.

MODEL 1

The first model assumed that all factors are related to each other by means of a single higher-order factor. In addition, the model allowed for correlated residuals among the four HEXACO factors. This makes it possible to examine whether these four factors share additional variance with each other that is not explained by a general Negative Emotionality factor.

Model fit decreased compared to the measurement model which serves as a comparison standard for theoretical models, CFI: .916 vs. 954, RMSEA = .009 vs. .007.

All primary factors except sentimentality had substantial loadings on the Negative Emotionality factor. Table 3 shows the residual correlations for the four HEXACO factors.

All correlations are positive suggesting that the HEXACO Emotionality factor captures some shared variance among these four factors that is not explained by the Negative Emotionality factor. However, two of the correlations are very low indicating that there is little shared variance between sentimentality and fear or dependence and anxiety.

MODEL 2

The second model, modeled the relationship among the HEXACO factors with a factor. Model fit decreased, CFI = .914 vs. .916, RMSEA = .010 vs. 009. Loadings on the Emotionality factor ranged from .27 to .46. Fear, anxiety, and dependence had higher loadings on the Negative Emotionality factor than on the Emotionality factor.

The main conclusion from these results is that it would be problematic to replace the Big Five model with the HEXACO model because the Emotionality factor in the HEXACO model fails to capture the nature of the broader Neuroticism factor in the Big Five model. In fact, there is little evidence for a specific Emotionality factor in this dataset.

MODEL 3

The discrepancy between the measurement model and Model 1 suggests that there are additional relationships between some primary factors that are not explained by the general Negative Emotionality factor. Examining modification indices suggested several changes to the model. Model 3 shows the final results. This model fit the data nearly as well as the measurement model, CFI = .949 vs. 954, RMSEA = .007 vs. .007. Inspection of the Modification Indices showed no further ways to improve the model by freeing correlated residuals among primary factors. In one case, three correlated residuals were consistent and were modeled as a factor. Figure 1 shows the results.

First, the model shows notable and statistically significant effects of neuroticism on all primary factors except sentimentality. Second the correlated residuals show an interesting patterns where primary factors can be arranged in a chain. that is, depression is related to moody, moody is related to anger, anger is related to anxiety, anxiety is related to fear, fear is related to self-consciousness and dependence, self-consciousness is related to dependence and finally, dependence is related to sentimentality. This suggests the possibility that a second broader dimension might be underlying the structure of negative emotionality. Research on emotions suggests that this dimension could be activation (fear is high, depression is low) or potency (anger is high, dependence is low).This is an important avenue for future research. The key finding in Figure 1 is that the traditional Neuroticism dimension is an important broad higher-order factor that accounts for the correlations among 7 of the 8 primary factors. These results favor the Big Five model over the HEXACO model.