Abstract
Despite decades of personality state research at the Big Five domain level, it is unclear if personality states have valid structure below the Big Five because there are no validated scales that measure traits and states at lower levels. We therefore developed the Big Five Aspect Adjectives (BFAA), an adjective-based inventory that measures personality aspects in both trait and state forms. Study 1 (n = 352, US sample) found support for a 60-item inventory. The BFAA showed excellent psychometric properties, replicated complex structures found in the BFAS, and showed similar correlation patterns with criterion variables as the BFAS. Study 2 (n = 144, Australian sample) validated the BFAA against the BFAS and showed good 4-week test-retest reliability. Study 3 (n = 549, US sample) used multivariate modeling to establish cross-instrument correspondence between BFAS traits and BFAA states and showed that BFAA states predict momentary positive and negative affect consistent with theory. Study 4 (n = 265, UK sample) used multilevel multivariate modeling to replicate Study 3 in an experience sampling design across 24 surveys and 3,999 observations, providing strong evidence for BFAA trait-state correspondence. This research provides the first systematic empirical evidence that personality states are valid constructs at the aspect level.
Introduction
A fundamental problem in personality science is whether the structures and relationships observed at the stable personality trait level occur at the state, or momentary experience, level. A leading contemporary personality framework that attempts to explain the relationships between traits and states is Whole Trait Theory (Fleeson, 2001; Fleeson & Gallagher, 2009; Jayawickreme et al., 2019). Whole Trait Theory proposes that traits represent density distributions of states such that individuals higher on a trait spend more time enacting corresponding states but that states are still meaningful units of analysis beyond traits. This framework has generated substantial research on trait–state correspondence at the Big Five domain level – the five domains being agreeableness (A), conscientiousness (C), extraversion (E), neuroticism (N), openness/intellect (O) – using experience sampling methodology (ESM) to demonstrate that domains function as both stable individual differences and dynamic within-person processes through personality states. At the same time, personality models increasingly incorporate personality constructs below the Big Five domains. One such model, which is central to the current research, is the Big Five Aspect Scales (BFAS; DeYoung et al., 2007). The BFAS divides each of the five domains into two aspects, each of which was discovered through research on the genetic covariance between personality traits at the facet level. Whether these aspect-level distinctions can be meaningfully measured as states remains untested. Although preliminary evidence suggests that state structures below the Big Five domain level may be detectable in momentary assessments (see Rauthmann et al., 2019), formal validation of sub-domain states has not been conducted.
The present research responds to this gap by developing and validating the Big Five Aspect Adjectives (BFAA), an adjective-based inventory designed to measure personality at both the aspect and Big Five domain levels in both trait and state forms. By using brief adjective markers rather than sentences, the BFAA is purpose-built for efficient repeated measurement in ESM designs while retaining the psychometric rigor required for trait assessment. Across four studies, we evaluate the psychometric properties of the BFAA, establish its convergence with the BFAS at the trait level, and validate its use as a state measure. We demonstrate that the BFAA recovers the hierarchical structure of the Big Five domains and aspects, exhibits robust trait–state correspondence at both levels, and yields theoretically coherent associations with momentary outcomes such as affect. Importantly, these results provide direct empirical evidence that Big Five aspects can be meaningfully assessed as personality states, establishing their validity in capturing within-person dynamics as well as stable individual differences. The BFAA therefore does not simply enable future tests of aspect-level personality processes, but directly demonstrates their theoretical potential while providing a validated, general-purpose measure that can be readily adopted by researchers investigating personality dynamics across the diverse domains where trait-state interplay matters.
The Big Five Aspect Scales and its Psychometric Derivation
The Big Five personality model is hypothesized to fit within a hierarchical structure of personality with potentially six levels. From the highest to lowest levels, there is the general factor of personality (see Musek, 2007), then meta traits (plasticity and stability; see DeYoung, 2006; Digman, 1997), then the Big Five (e.g., Costa & McCrae, 1992; Digman, 1990), then aspects (see DeYoung et al., 2007; Weisberg et al., 2011), then facets (e.g., Goldberg, 1999), and finally nuances (see Mõttus et al., 2017). We note that there is no consensus on the exact structure of personality, however consistent support has been found for the meta-trait, the Big Five domain, and the aspect levels. The aspect level therefore represents a sensible focal point for developing a state-based measure of personality because there are no competing constructs at this level of analysis, and it is more manageable for state researchers than measuring 30+ states at the facet level.
The BFAS measures the Big Five traits and further splits these five traits into two aspects each. A splits into compassion (Ac) and politeness (Ap) where Ac captures the emotional elements of empathizing and feeling altruistic towards others, whereas Ap is focused on the avoidance of being rude and is putatively a more cognitively-oriented form of A. Ac and Ap are correlated at approximately .45. C splits into industriousness (Ci) and orderliness (Co) where Ci refers to a proclivity to be productive and to work hard, whereas Co refers to the degree to which one likes order, cleanliness, and schedules in their environment. Ci and Co are correlated at approximately .40. E splits into assertiveness (Ea) and enthusiasm (Ee) where Ee is focused on the emotional aspects of E, such as having fun and experiencing high-activation positive emotion, and Ea is focused on the goal-oriented pursuits of social dominance, leadership, and status. Ea and Ee are correlated at approximately .50. N splits into volatility (Nv) and withdrawal (Nw) where Nv relates to an outward manifestation of negative affect including irritability and anger, whereas Nw is related to an internalized negative affect experience capturing feelings related to anxiety, depression, and stress and is thought to be related to the behavioral inhibition system (DeYoung, 2015). Nv and Nw are correlated at approximately .60. O splits into intellect (Oi) and openness (Oo) where Oi is related to one’s view of their ability to think and learn and is a measure of intellectual confidence, and Oo captures a proclivity to be creative, to enjoy artistic and aesthetic pursuits, and experience divergent thinking. Oi and Oo are correlated at approximately .35.
The BFAS was derived from an observation by Jang et al. (2002) that the shared variance of the facets within each of the Big Five, as measured by the NEO-PI-R (Costa & McCrae, 1992), are underpinned by two genetic factors per domain. DeYoung et al. (2007) reasoned that such factors ought to be measurable with items from existing inventories and hence set about testing this hypothesis across three studies. In Study 1, using data from 481 Eugene-Springfield community sample participants, six facets for each Big Five domain from the NEO-PI-R and nine facets for each Big Five domain from the AB5C-IPIP (Goldberg, 1999) were factor analyzed to extract the Big Five. Then, within each domain, Velicer’s minimum average partial test was conducted to assess whether two factors might best represent the systematic variance of the facets within each domain, which indeed was the case. Finally, a two-factor principal-axis factoring with direct oblimin rotation (Δ = 0) analysis was used on the facets within each domain, which showed a clear pattern of two complementary aspects per domain.
In Study 2, factor scores were calculated for each aspect and correlated with every one of the 2,000+ IPIP items in the Eugene-Springfield community sample. For each aspect, 15 items were selected for further factor analysis in a student sample (n = 480) based on certain criteria (e.g., no item could correlate with another aspect within .10, redundancy was minimized, a balance between positively- and negatively-worded items was maintained). The 15 items per aspect were then factor analyzed (principal-axis factoring with direct oblimin rotation, Δ = 0) to arrive at the final inventory of 100 items with 10 items per aspect. Care was taken to ensure that no item cross-loaded on the complementary aspect within .10, although this criterion was relaxed for negatively valenced items to ensure that a ratio of 6/4 positively- to negatively-valenced (or vice versa) items were included in each aspect. In Study 3, the residualized aspects were correlated with the genetic factors reported by Jang et al. (2002). Support was broadly found for a concordance between the phenotypic factors (traits as Big Five aspects) and the genetic factors (the genes), suggesting that aspects represent phenotypic expressions of underlying genetic factors.
The psychometric derivation of the BFAS provides both the theoretical foundation and methodological scaffolding for the development of the BFAA. In developing an adjective-based version of the BFAS, our approach replicates Study 1 and Study 2 of DeYoung et al.’s (2007) methodology as much as practicable, most notably in our Study 1.
Using Adjectives to Address the Problem of Measuring Aspects at the State Level
Most trait inventories are unsuitable for state measurement because they use longer-form statements. The BFAS is no exception in this respect (e.g., “[I a]m a person whose moods go up and down easily”; DeYoung et al., 2007), where a total of 463 words (excluding instructions) are used to measure 100 items. Even the recently developed BFAS-40 (Gallagher et al., 2022), which is a 40-item version of the BFAS, still uses longer-form statements, contains 183 words (excluding instructions), and was not designed for state measurement.
Researchers have measured personality states at the Big Five domain level (e.g., Fleeson, 2001; Jacques-Hamilton et al., 2019; McNiel & Fleeson, 2006; Margolis & Lyubomirsky, 2020; Spark & O'Connor, 2021; Zelenski et al., 2013), relying on the use of adjective-based inventories derived from the likes of Goldberg (1992), Saucier (1994), and Thompson (2008), because they can be easily converted to measure how personality manifests in shorter timeframes and are quicker to administer with only one word per item. Encouragingly, adjectives are not just time-efficient and flexible, they are also excellent markers of quantifiable behavior. For example, Wiedenroth and Leising (2020), found that self-descriptions using adjectives like ‘clumsy’ strongly predicted quantifiable behaviors, such as the frequency of dropping objects, or unintentionally hurting oneself. Adjectives may therefore be higher in ‘facticity’ (a property of an item that describes how much it refers to objective facts as opposed to subjective reality; Wiedenroth & Leising, 2020), be more meaningful to the participant in terms of practical behavior, and subsequently be better suited to measuring specific acts of behavior compared to statement-based items that are often developed idiosyncratically by the researcher. Given these advantages, an adjective-based version of the BFAS should enable efficient state measurement at the aspect level while preserving the theoretical and psychometric rigor of the original trait measure.
Validating the BFAA: Replicating Subtle Relationships Within the Big Five Aspect Scales
A valid adjective-based measure of the BFAS must replicate not only the basic domain and aspect structure but also the complex patterns of relationships among aspects that reveal the nuanced architecture of personality. DeYoung et al. (2007) identified several such patterns that distinguish the aspect level from simpler domain-level models. Replicating these patterns provides strong evidence that an adjective measure captures the same psychological constructs as the original BFAS.
The first, and most obvious, is that the Big Five traits are readily recoverable. Beyond this obvious requirement, there are many instances of high correlations between aspects from different domains. For example, the aspects of Ee and Ap show differential associations whereby Ap is positively correlated with Ee but negatively correlated with Ea. A related example is the high correlation between Ee and Ac, which is approximately the same as the correlation between Ee and Ea – a combination that captures social affiliation. In this case, Ee reflects the rewarding nature of social affiliation and Ac reflects the concern for others (by extension, a person who is high in Ee but low in Ac is one who is excited by the reward of social affiliation but is not likely to ‘care’ about the people with whom they are obtaining their social reward). A third example is the intercorrelation between Oi, Ea, and Ci – a constellation DeYoung et al. (2007) suggest is likely to be predictive of workplace performance. A fourth example is one of suppression, whereby the correlation between N and C is negative (which is a consistent finding in the personality literature; e.g., Mount et al., 2005) and yet the correlation between Co and N is essentially zero with Ci, accounting for the negative N∼C correlation. Going further, when Ci is controlled for (e.g., residualizing Co by regressing Co onto Ci and then using the residuals from this regression as the residualized version of Co) the relationship between N and Co becomes slightly positive, indicating that this relationship is suppressed by the Ci∼N relationship. From a practical perspective, and as noted by DeYoung et al. (2007, p. 892) such a relationship means that “[g]iven two people (or groups) with equal levels of Industriousness, the one higher in Orderliness is likely to show higher levels of Neuroticism”, which, at the extremes, may be measuring perfectionism as a maladaptive form of C.
With the above points in mind, we seek to replicate these patterns in addition to meeting basic psychometric criteria, such as internal consistency, test-retest reliability, convergent and divergent validity, and criterion validity. Study 1 focuses on the initial development of the BFAA and tests whether these specific patterns emerge. Study 2 establishes test-retest reliability and further validates these complex relationships.
Using the BFAA to Measure Personality Aspects at the State Level
Beyond replicating the BFAS structure and relationships, a critical test of validity is demonstrating trait-state correspondence at the aspect level. Because no prior research has validly measured aspects as states, we draw from Big Five research on states, initiated by Fleeson (see Fleeson, 2001; Fleeson & Gallagher, 2009). Fleeson (2001) used an experience sampling method to establish that Big Five traits strongly predict mean personality states across time despite substantial within-person variability. Participants’ personality states were measured over 13 days, five times per day using Big Five adjectives taken from Goldberg’s markers (Goldberg, 1992). Participants were asked to describe their states over the previous hour, which in turn provided evidence of mean states across time. Whilst states in any given hour had a correlation of only .29 with any other hour, when comparing a randomly selected half of the dataset to the other half, the correlation of mean behavior increased to .90.
Fleeson and Gallagher (2009) meta-analyzed 15 studies incorporating more than 20,000 reports of states to assess the degree to which traits correlated with states. They showed that trait-state correlations varied depending on timeframe and measurement reliability. When participants described their states over shorter timeframes (20 or 30 min), correlations between traits and mean states were between .23 (for conscientiousness) and .44 (for agreeableness). When timeframes were extended to be at least 1 hr, the correlations increased to between .41 (for extraversion) and .67 (for openness to experience). Increasing the timeframe further (e.g., to 3 hr) did not meaningfully change the magnitude of these effects. Studies with higher reliabilities had correlations more consistently around .60. In terms of single-state correlations, there was significant variation between studies. For example, in one study, all five traits had single-state correlations of .46 to .63 compared to most other studies being closer to .30 to .40. Across all studies, the meta-analyzed single-state correlations ranged from .21 (for extraversion) to .37 (for openness to experience). Notwithstanding this variability in correlations, the most important observation from the meta-analyzed data was that each trait was the dominant predictor of its corresponding state as measured either by single-state correlations or mean states over multiple within-person reports. Therefore, in our assessment of the validity of the BFAA as a state measure, we would expect to see correlations between the aspect trait and its corresponding aspect state fall within the range of .20 to .60. As a more rigorous test of this trait-state correspondence, we would further expect to see that each state is predicted most strongly by its corresponding trait (i.e., Ac at the trait level should be the dominant predictor of Ac at the state level). A unique challenge of validating a measure of aspects for state-based research is that each aspect must ‘compete’ with its complementary aspect, and so to find direct trait-state relationships using aspects when controlling for all other aspects is strong evidence of the validity of the state-based measure and subsequent trait-state correspondence. We address this challenge in Study 3 and Study 4.
Criterion Validity: Dark Triad Traits
Beyond structural validity and trait-state correspondence, we also assess criterion validity by examining relationships with theoretically relevant external constructs. The Dark Triad personality traits—narcissism, Machiavellianism, and psychopathy—have been studied alongside the Big Five traits for more than two decades (Muris et al., 2017). The Dark Triad represents a constellation of traits that collectively capture a disposition towards exploitativeness. More specifically, narcissism is characterized by grandiosity, pride, and egotism, Machiavellianism is characterized by manipulation and exploitation of others, a cynical disregard for morality, and a focus on self-interest and deception, and psychopathy is characterized by enduring antisocial behavior, impulsivity, selfishness, callousness, and remorselessness. Some research has investigated the relationships between the BFAS and the Dark Triad. The best example is Jonason et al.’s study (2013), which showed certain correlations with a short measure of the Dark Triad (the Dirty Dozen; Jonason & Webster, 2010). Some examples of these correlations include narcissism with Ap (r = −.33), Ea (r = .20), and Nv (r = .18), Machiavellianism with Ac (r = −.14), Ap (r = −.43), and Co (r = −.18), and psychopathy with Ac (r = −.30), Ap (r = −.32), Ee (r = −.24), and Nv (r = .22).
Building on these findings, we seek to adopt a more stringent test of validity by assessing the lower-level constructs of the Dark Triad with each of the BFAA aspects at the trait level. The first set divides narcissism into grandiose and vulnerable forms, which are measured by the Brief Pathological Narcissism Inventory (Schoenleber et al., 2015). Grandiose narcissism captures the degree to which someone is arrogant, fantasizes about unlimited power, is manipulative, and engages in self-enhancement to appear altruistic. Vulnerable narcissism captures the degree to which someone is dependent on external validation for their self-worth, conceals their personal flaws, and is irrationally angered when they don’t get what they want. Similarly, psychopathy can be divided into facets. The Levenson Self-Report Psychopathy Scale (Levenson et al., 1995), for example, distinguishes between primary and secondary psychopathy, where primary psychopathy is characterized by a lack of compassionate emotions that often result in a callous disregard for others and superficial charm, and secondary psychopathy is marked by negative emotional states such as anger and an associated lack of self-control. Machiavellianism (e.g., as measured by the Mach-IV; Christie & Geis, 1970), which lacks established lower-level facets, refers to a proclivity to be manipulative, deceitful, and to use others for one’s own gain. To date, these specific relationships have not been tested and so our approach is to assess the correlation structure of the BFAS aspects with each of these Dark Triad facets and then to compare the corresponding aspects of the BFAA with the same Dark Triad facets. If the BFAA is to behave similarly to the BFAS, then it should be able to replicate the same correlation structure. This validation check is tested in Study 1.
Criterion Validity: Affect and People Interacted With
Positive and negative affect, often measured with the Positive and Negative Affect Schedule (PANAS; Thompson, 2007) have long been associated with trait extraversion and trait neuroticism, respectively (Lucas & Fujita, 2000). At the aspect level, however, previous research using the BFAS (e.g., Sun et al., 2018) shows that Ee is the dominant aspect predicting positive affect with Nw having a secondary effect, and Nw is the dominant aspect predicting negative affect with Nv having a secondary effect. Oi is also a negative predictor of negative affect and, to a lesser extent, a positive predictor of positive affect, although the relationship with positive affect may disappear after controlling for the other aspects (Zajenkowski & Matthews, 2019) and so Oi is likely weak. If the BFAA captures the same constructs as the BFAS, these patterns should emerge with the BFAA. The number of people interacted with is a single-item state variable that we reasoned should be related to Ee primarily, and Ea secondarily, because it captures an approach-oriented motivation to interact with others. These relationships are tested in Study 3 and Study 4.
Criterion Validity and Controls: Age and Gender
Age is a consistent covariate with personality at the Big Five and facet level (see Soto et al., 2011), however there is relatively little research at the aspect level, which limits identification of specific theoretical relationships to test. In terms of gender, Weisberg et al. (2011) reported Cohen’s d effect sizes between men and women using the BFAS of approximately 0.20 to 0.40, which implies an approximate sample size range of 320 to 1,300 at 95% power to reliably detect. Therefore, rather than testing for mean differences, our approach is to test whether the correlation pattern between the BFAS and both age and gender was similar to the correlation pattern between the BFAA and both age and gender, which we test in Study 1. In addition, because of the consistent covariance age and gender have with personality, and because they are excellent exogenous variables (their errors are not correlated with the errors of the outcome variable), they are included as controls when conducting multivariate analyses so that the personality estimates are correctly partialed (see Antonakis, 2011; Antonakis et al., 2010). These analyses are conducted in Studies 3 and 4.
Outline of Studies
In summary, the present research includes four studies structured such that the initial validation of the BFAA at the trait level is conducted in Study 1, which includes the selection of initial adjectives from a candidate item pool and a detailed comparison of the uniquely complex relationships between aspects and Big Five domains found in the BFAS. Evidence of criterion validity is also presented using the Dark Triad traits, age, and gender. Study 2 establishes test-retest reliability along with replication of the relationships between aspects and Big Five domains. Having established evidence for the functionality of the BFAA at the trait level in Studies 1 and 2, Study 3 tests trait-state correspondence using a single-timepoint state assessment. Finally, Study 4 extends Study 3 by using an experience sampling design across 24 surveys, providing stronger evidence for trait-state correspondence through repeated within-person measurement. All data and scripts for all studies are available on OSF (https://osf.io/xu5rt/).
Study 1
Method
Purpose and Design
The purpose of Study 1 was to develop and validate the BFAA against the BFAS at the trait level. We started with an adjective list of putatively relevant items and conducted several analyses (e.g., principal components analysis, factor analyses, convergent and discriminant validity tests, etc.), and then conducted zero-order and residualized correlation tests comparing the BFAA to the BFAS. Finally, we checked criterion validity using the Dark Triad traits, age, and gender. A benefit of the Dark Triad for use in this study is that it, like the Big Five aspects, has two levels of analysis with a complex structure. The ability to measure such complex structure with theoretically relevant constructs simultaneously (i.e., running fully-partialed correlation) is a significant advantage over measuring individual traits one by one. This study was not preregistered. Ethics approval for this study was granted by The University of Sydney Human Research Ethics Committee.
Participants
Despite heuristics for sample size calculations in factor analyses (e.g., 5–10 per item being common), the work by Mundfrom et al. (2005) suggests that if the variables-to-factor ratio is greater than 7, the sample rarely needs to be greater than n = 180. Given that we did not know how many items per factor we would require – although approximately 7 seemed reasonable given that Saucier’s (1994) scales use 8 items per factor – we aimed for a doubling of this n = 180 recommendation to be conservative. Hence, 352 United States participants were recruited from Prolific Academic. There were 173 men and 173 women and 6 who did not disclose their gender. Participants were aged from 18 to 77 years with a mean of 36.46 years (SD = 12.34). Automated/bot responses were screened out by Prolific prior to data collection and, following data collection, data were inspected for aberrant/random responding using variable distribution checks, histogram analysis and sample descriptive statistics. No data quality issues were present.
Measures
Age and Gender
Age was measured in years, and gender was coded as 0 = men, 1 = women.
Big Five Aspect Scales
The Big Five aspects were measured with the BFAS (DeYoung et al., 2007) on a 1 to 5 Likert scale where 1 = Strongly disagree and 5 = Strongly agree. Mean scores were calculated for each aspect and Big Five trait. Omega total (ω) estimates for the aspects were Ac = .90, Ap = .77, Ci = .88, Co = .81, Ea = .88, Ee = .82, Nv = .93, Nw = .88, Oi = .86, and Oo = .76.
Adjectives Pool
The adjective pool consisted of 169 adjectives (see Supplemental Tables) and was derived from several sources explained below. Participants were asked to rate the degree to which each adjective described them in general. A 1 to 6 Likert scale was used where 1 = Strongly disagree and 6 = Strongly agree. Mean scores were used for the final BFAA inventory. Items were reverse scored where appropriate.
Brief Pathological Narcissism Inventory (B-PNI)
We used the B-PNI (Schoenleber et al., 2015) to measure the two dimensions of narcissism. Both had good omega total scores (grandiose narcissism, ω = .83; vulnerable narcissism, ω = .90). A 1 to 5 Likert scale was used where 1 = Strongly disagree and 5 = Strongly agree.
Levenson Self-Report Psychopathy Scale (LSRP)
The LSRP (Levenson et al., 1995) measures both primary psychopathy and secondary psychopathy. Primary psychopathy (ω = .87) is focused on a callous disregard for others and is characterized by coldness, superficial charm, and a lack of anxiety. It is measured with 16 items, examples of which include Looking out for myself is my top priority and People who are stupid enough to get ripped off usually deserve it. Secondary psychopathy (ω = .74) is characterized by impulsive-like behavior and hostility. It is measured with 10 items, examples of which include I have been in a lot of shouting matches with other people and I don’t plan anything very far in advance. A 1 to 4 Likert scale was used where 1 = Strongly disagree and 4 = Strongly agree.
Mach-IV
Machiavellianism was measured using the MACH-IV (Christie & Geis, 1970) which consists of 20 items. Example items include The best way to handle people is to tell them what they want to hear and It is wise to flatter important people. A 1 to 5 Likert scale was used where 1 = Disagree strongly and 5 = Agree strongly. The omega total score was good (ω = .79).
Procedure and Estimation
To arrive at the final set of adjectives, we completed five steps. At step 1, we selected 169 adjectives that we aligned with the BFAS items in terms of content. These items served as our initial item pool. Our criteria for choosing these items were based on the requirement that they should relate to the aspects as highlighted by DeYoung et al. (2007), without being clearly a better marker of the Big Five trait instead of the aspect. For example, the adjective extraverted is an excellent marker for E but because it does not clearly relate to either Ea or Ee, we opted to exclude it from the initial item pool. In selecting our adjectives, we drew from Goldberg’s markers (Goldberg, 1992), Saucier’s Mini-Markers (Saucier, 1994), Thompson’s markers (Thompson, 2008), and the AB5C circumplex (Hofstee et al., 1992). Where there were gaps in the putative factor space, we added other adjectives based on synonym searches in English dictionaries.
At step 2, we adjusted the item scores for acquiescence bias (for details, see Primi et al., 2020) and then conducted principal components analysis with varimax rotation to extract the Big Five (all analyses were conducted in R; R Core Team, 2018). This step served as a check that the items loaded on the expected domain. For those items that did not load on the expected domain, we made a judgment as to whether to discard the item or retain it. In general, we dropped items that failed to load on the expected domain at |.35| or greater or that cross-loaded on several domains that were not sensible (a potentially sensible cross-load would be one that loads onto both E and O because E and O are related via trait plasticity). However, we were more lenient with reverse-scored items as we wanted to retain reverse-scoring items to remain consistent with DeYoung et al.’s (2007) approach.
At step 3, we conducted a factor analysis (principal axis factoring with direct oblimin rotation, Δ = 0) for each Big Five domain using the candidate items from step 2. For each domain, we entered the candidate items and the items from the original BFAS. We reduced the candidate items based on a rule of a loading of less than |.40| but being more lenient on reversed-scored items. In line with DeYoung et al.’s (2007) procedure, we selected the items that (i) were from the correct domain as per the factor analysis in step 2, (ii) were highly correlated with the aspects, and (iii) were correlated with one aspect more than the other aspect by at least .10. We had no requirement as to how many items were to be retained at this step.
At step 4, we conducted an analysis using the OASIS tool described in Cortina et al. (2020). The OASIS tool is designed to test every possible combination of items specified by the user and produces an internal consistency term (an alpha coefficient), a convergent validity score (a correlation with a target variable), and a divergent validity score (a correlation with a second target variable). The procedure we adopted was the same for every aspect. That is, for each aspect we selected the calculated variable from the BFAS as the convergent validity target variable and the complementary aspect from the BFAS as the divergent validity target variable (e.g., if Ee was the convergent target variable, then Ea was the divergent target variable). We applied filters to the output based on criteria for the convergent validity, divergent validity, and alpha coefficient. The first criterion was that alpha needed to be a minimum of .70, the second criterion was that the convergent validity be as high as possible, and the third being that the divergent validity should be within a specified range based on the correlations between complementary aspects as presented in DeYoung et al. (2007) and Weisberg et al. (2011). The target ranges for the divergent validities were Ac and Ap = [.36, .44], Ci and Co = [.39, .45], Ea and Ee = [.42, .50], Nv and Nw = [.58, .61], and Oi and Oo = [.35, .40]. Selecting items that resulted as close as possible in these divergent validities, whilst at the same time converging on the target aspect, maximized our ability to find an appropriate combination of items that behaved like the BFAS.
In the fifth and final step, we conducted validation testing using the raw scores (we reverted to raw scores at this step because it is important that we confirmed that the final scale behaves correctly without any adjustments). In conducting validation, it was important to not just inspect the correlations between the BFAS and BFAA, but also to analyze the behavior of the residualized aspects, as noted in our introduction. Thus, validation checks were conducted such that (i) the expected raw correlations between the aspects and domains were as expected, (ii) the raw correlations between the aspects were expected, (iii) the residualized correlations between the aspect were as expected, (iv) the aspects and domains correlated with the BFAS aspects and domains, and (v) that correlations with the Dark Triad traits, age, and gender were similar across the two inventories.
Consistent with DeYoung et al.’s (2007) approach and as is common practice in personality research, we did not conduct a confirmatory factor analysis (CFA). CFA assumes simple structure with exclusive item loadings, which is unrealistic for personality models (Mooradian & Nezlek, 1996). Although obtaining ‘good fit’ on standard CFA metrics can often be achieved by correlating the error terms of items across traits to better account for a lack of simple structure, or in more recent examples using novel optimizing techniques (e.g., ‘brute force’ optimization; Gallagher et al., 2022), we suggest that these techniques effectively result in a quasi-exploratory approach and thus negate the original purpose of a CFA. We therefore focused on replicating DeYoung et al.’s approach as much as practicable.
Results
The first step was to run several principal component analyses (varimax rotation) to filter out the initial adjectives that did not load well on their putative Big Five domain. The final solution showed a clean solution (see Supplemental Tables). Because of a tendency for negative adjectives to load onto a common factor when using principal components, we made a judgment to retain adjectives that cross-loaded with N, namely Aggressive, Inefficient, Lazy, and Unproductive. In addition, it makes theoretical sense for Inefficient, Lazy, and Unproductive to cross-load onto N because C and N are typically negatively correlated. These three items were also retained because they are reverse-score items for C and we wanted to retain reverse-score items where possible. Effective cross-loaded with E, which can be expected theoretically because of the constellation between Ea and Ci. Unintelligent and Unimaginative were similarly retained to keep a sufficient number of reverse-scored items for Oi. Other cross-loading items were retained as they loaded on their primary factor.
Study 1 Principal Axis Factoring for Each Big Five Domain
Note. The original BFAS are shown at the top of each factor analysis and are ordered by question number (see Supplemental Tables for question mapping). The candidate adjectives are ordered by aspect and then by loading size. Direct oblimin rotation (Δ = 0). Loadings at |.35| or greater are shown in bold.
Study 1 Item Combinations From OASIS and the Final BFAA Scales
Note. BFAS = Big Five Aspect Scales, Ac = compassion, Ap = politeness, Ci = industriousness, Co = orderliness, Ea = assertiveness, Ee = enthusiasm, Nv = volatility, Nw = withdrawal, Oi = intellect, Oo = openness. The items are ordered based on the factor loadings from the principal axis factoring analysis. The items selected to form the BFAA are bolded.
Study 1 Correlations, Means, Standard Deviations and Omega Scores of the BFAS and BFAA
Note. n = 352. BFAS = Big Five Aspect Scales, BFAA = Big Five Aspect Adjectives, Ac = compassion, Ap = politeness, Ci = industriousness, Co = orderliness, Ea = assertiveness, Ee = enthusiasm, Nv = volatility, Nw = withdrawal, Oi = intellect, Oo = openness. Omega total scores shown in bold on the diagonal. Zero-order correlations are under the diagonal and residualized correlations are above the diagonal. Correlations above |.10| are significant at the .05 level and above |.13| are significant at the .01 level.
Study 1 BFAS and BFAA Fully Partialed Correlations With Dark Triad Traits, Age, and Gender
Note. n = 344 after listwise deletion. Each correlation matrix (A and B) was fully partialed and run separately, and included all aspects (not shown). BFAS = Big Five Aspect Scales, BFAA = Big Five Aspect Adjectives. Ac = compassion, Ap = politeness, Ci = industriousness, Co = orderliness, Ea = assertiveness, Ee = enthusiasm, Nv = volatility, Nw = withdrawal, Oi = intellect, Oo = openness. Grand = grandiose narcissism, Vuln = vulnerable narcissism, Primary = primary psychopathy, Second = secondary psychopathy, Mach = Machiavellianism, Gender coded as men = 0, women = 1. Correlation pairs where at least one correlation is significant at .05 and where there is a significant difference between the pair (p < .05, Williams t-test adjusted for false discovery) are bolded.
Study 2
Method
Purpose and Design
The purpose of Study 2 was to provide further evidence of the validity of the BFAA at the trait level. We conducted test-retest reliability analysis in a two-wave repeated-measures design and compared zero-order and residualized correlations between the BFAA and BFAS as per Study 1. This study was not preregistered. Ethics approval for this study was granted by The University of Sydney Human Research Ethics Committee.
Participants
For two-wave studies seeking to assess test-retest reliability with 90% power, Bujang and Baharum (2017) recommend a minimum sample size of 50, 30, 20, and 13 for intraclass correlation coefficients (ICC; explained below) of .40, .50, .60, and .70 respectively. We recruited 144 Australian university students, of whom 59 completed a second wave. To reduce participant burden, the 59 were only required to complete the BFAA at baseline. Completion was for partial course credit. Thirty-nine participants were men and 105 were women. Participants were aged from 17 to 50 years with a mean of 20.47 years (SD = 4.30). The average time between waves was 25.50 days (SD = 10.20). Data were inspected for aberrant/random responding using variable distribution checks, histogram analysis and sample descriptive statistics. No data quality issues were present.
Measures
Big Five Aspect Scales
The Big Five aspects were measured with DeYoung et al.’s (2007) BFAS as per Study 1. Omega total scores for the aspects were Ac = .91, Ap = .76, Ci = .86, Co = .82, Ea = .85, Ee = .89, Nv = .91, Nw = .86, Oi = .82, and Oo = .86.
Big Five Aspect Adjectives
The six-item scale constructed in Study 1 was used (see Table 2) with a 6-point Likert scale where 1 = “Strongly disagree” and 6 = “Strongly agree”.
Procedure and Estimation
In the baseline survey, participants completed basic demographic information, the BFAA, and the BFAS (for those not doing a second wave). The baseline survey was used to conduct the same analyses as per Study 1 with respect to the correlations (both zero-order and residualized) of the BFAA with the BFAS. Participants who did not complete the BFAS at baseline, were invited back for a follow-up assessment of the BFAA approximately four weeks later. For the test-retest reliabilities, the formula for ICC (2,k) was used as per the flowchart in Koo and Li (2016), which is appropriate for test-retest reliability based on the mean from k = 2 measurements with absolute agreement (i.e., seeking the degree of agreement on the same scale for the k = 2 time points). Koo and Li suggest that ICC values less than 0.50 are ‘poor’, values between 0.50 and 0.75 are ‘moderate’, values between 0.75 and 0.90 are ‘good’, and values greater than 0.90 are ‘excellent’. R (R Core Team, 2018) was used for ICC calculations. The primary ICC results set was calculated on the test-retest subsample (n = 58 after listwise deletion) using restricted maximum likelihood estimation. With the ability to exploit full information maximum likelihood estimation for missing data (see Revelle et al., 2021) coupled with BFAA data at baseline, a second ICC results set could also be calculated on the full dataset (n = 144). This second result set was used for additional power in estimating confidence intervals given that concordance between the two result sets was almost perfect.
Results
Study 2 Test-Retest Reliability of the BFAA
Note. BFAA = Big Five Aspect Adjectives. ω = omega total. ICC (2,2) = intraclass correlation coefficient based on absolute agreement across 2 measurements. 95% confidence intervals shown in square brackets. Full information maximum likelihood estimation used for ICC calculations where n = 144, and restricted maximum likelihood estimation where n = 58 after applying listwise deletion.
Study 2 Wave 1 Correlations, Means, Standard Deviations and Omega Scores of the BFAS and BFAA
Note. n = 144. BFAS = Big Five Aspect Scales, BFAA = Big Five Aspect Adjectives, Ac = compassion, Ap = politeness, Ci = industriousness, Co = orderliness, Ea = assertiveness, Ee = enthusiasm, Nv = volatility, Nw = withdrawal, Oi = intellect, Oo = openness. Omega total scores shown in bold on the diagonal. Zero-order correlations are under the diagonal and residualized correlations are above the diagonal. Correlations above |.16| are significant at the .05 level and above |.21| are significant at the .01 level.
Study 3
Method
Purpose and Design
The primary purpose of Study 3 was to test for initial evidence of the BFAA’s validity as a state measure by conducting a multivariate analysis using the BFAS at the trait level to predict the corresponding BFAA aspect at the state level in a single time point, controlling for age and gender. A similar analysis was conducted to show that the Big Five domains can be recovered at the state level. These analyses also provide evidence of cross-instrument trait-state correspondence. A secondary purpose of Study 3 was to test the BFAA’s ability at the state level to predict momentary positive and negative affect, and the number of people interacted with over the last 3 hr. This study was not preregistered. Ethics approval for this study was granted by The University of Sydney Human Research Ethics Committee.
Participants
We are not aware of any readily available tools for guiding sample size decisions in the case of multivariate regression with 12 predictor variables and 10 outcome variables, and so we considered the research on statistical power in relation to linear multiple regression and structural equation modeling with single-item variables (i.e., path models). This research suggests that multiple regression requires 15 to 20 participants per predictor, whereas for a structural equation model, 15 participants per variable should be sought (Siddiqui, 2013). Other research suggests that 440 participants may be a safe upper limit in the case of structural equation models with single-item variables, but that if those variables have been properly adjusted for reliability, the number is likely much lower (Wolf et al., 2013). Given this lack of clarity, we erred on the side of caution and sought 550 participants. After data collection, 549 United States participants were recruited via Prolific Academic, where 271 were men, 270 were women, and 8 did not disclose their gender. Participants were aged from 18 to 85 years with a mean of 41.18 years (SD = 14.98). Participants were compensated with monetary reward upon successful completion of the questionnaire, which was conducted online. The same data screening and data check procedures as Study 1 were conducted. No data quality issues were present.
Measures
Age and Gender
Age was measured in years, and gender was coded as 0 = men, 1 = women.
Big Five Aspect Scales
The Big Five aspects were measured with DeYoung et al.’s (2007) BFAS as per Study 1. Omega total scores for the aspects were Ac = .90, Ap = .81, Ci = .91, Co = .86, Ea = .91, Ee = .90, Nv = .93, Nw = .92, Oi = .89, and Oo = .83.
Big Five Aspect Adjectives
The BFAA instructions were worded to measure personality states by asking participants to rate how each adjective described them during the last 3 hours using a 1 to 6 Likert scale where 1 = Strongly disagree and 6 = Strongly agree.
Positive and Negative Affect (State)
The PANAS (Thompson, 2007) was used to measure positive and negative affect at the state level asking participants to rate how they felt right now on a 1 to 6 Likert scale. Omega total scores were PA = .88, NA = .90.
Number of People Interacted With
We asked participants to specify how many people they interacted with in the last 3 hr.
Procedure and Estimation
Participants first completed their demographic information, then their trait measures, and then their state measures. A 3-hr time window for the states was selected. We chose to use the BFAS for traits instead of the BFAA because, first, the previous studies already showed excellent convergence of the BFAA traits with the BFAS traits and, second, providing evidence of a cross-instrument correspondence is a stricter test of correspondence than using the same instrument. Indeed, if each BFAA state is best predicted by its corresponding aspect using the original BFAS inventory, then this is powerful evidence that the BFAA is behaving as desired at the state level. We also conducted a similar analysis at the Big Five domain level. Multivariate analysis controlling for age, gender, and BFAS trait measurement error using Bollen’s (1996) method in R (R Core Team, 2018) using the lavaan package (Rosseel, 2012) was conducted for this study. For the additional validation using positive and negative affect, and number of people interacted with, we predicted all three outcome variables in a multivariate model using the BFAA aspect states (measurement error corrected) whilst controlling for the BFAS aspects (measurement error corrected), age, and gender. Including the BFAS in this model accounted for trait effects.
Results
Study 3 Correlations, Means, Standard Deviations and Internal Consistency Estimates of the BFAS and State-Based BFAA
Note. n = 549. BFAS = Big Five Aspect Scales, BFAA = Big Five Aspect Adjectives (state). Gender coded as 0 = men, 1 = women. Ac = compassion, Ap = politeness, Ci = industriousness, Co = orderliness, Ea = assertiveness, Ee = enthusiasm, Nv = volatility, Nw = withdrawal, Oi = intellect, Oo = openness. Omega total scores shown in bold on the diagonal. Correlations above |.09| are significant at the .05 level and above |.11| are significant at the .01 level.
Study 3 Summary of Multivariate Results at Each of the Big Five Domain and Aspect Levels
Note. n = 541 after listwise deletion. Robust maximum likelihood estimation. BFAS = Big Five Aspect Scales, BFAA = Big Five Aspect Adjectives. b = unstandardized regression coefficient, SE = standard error of the estimate, β = standardized regression coefficient. All BFAS traits are corrected for measurement error. This table is a summary of the dominant predictors for each outcome shown in the full multivariate models in the Supplemental Tables.
Study 3 Summary of Significant Multivariate Results for Outcome Variables at the Aspect Level
Note. n = 541 after listwise deletion. Robust maximum likelihood estimation. BFAS = Big Five Aspect Scales, BFAA = Big Five Aspect Adjectives. b = unstandardized regression coefficient, SE = standard error of the estimate, β = standardized regression coefficient. All BFAS traits and BFAA states are corrected for measurement error. Ordered by standardized regression coefficient within each outcome. This table is a summary of the full multivariate model in the Supplemental Tables.
Study 4
Method
Purpose and Design
The purpose of Study 4 was to build on the evidence from Study 3 to test the validity of the BFAA states when measured across multiple time points as per typical experience sampling studies. A specific focus was on the correspondence between BFAA traits and BFAA states (where Study 3 tested for cross-inventory correspondence). Participants first completed a baseline survey with their demographic information, followed by the BFAA (trait version). After the baseline survey was completed, participants were invited to complete up to 3 follow-up surveys per day over consecutive days (up to 24 surveys in total), which were scheduled to be released randomly between 08:00 to 10:00, 12:00 to 14:00, and 17:00 to 19:00 hours GMT. The state version asked participants to focus on the previous 2 hr. State positive and negative affect, and the number of other people interacted with over last the 2 hr, were also included for validation purposes, however the primary validation was the extent to which the BFAA traits predicted the BFAA states as per density distribution theory under Whole Trait Theory (Jayawickreme et al., 2019). This study was not preregistered. Ethics approval for this study was granted by Durham University Education Ethics Committee.
Participants
With the same power analysis limitations noted in Study 3 but with the additional complexity of multilevel modeling, using guidance from several sources (see Maas & Hox, 2005; Siddiqui, 2013; Wolf et al., 2013), we balanced power requirements for between-person effects and within-person effects and targeted 270 participants across a maximum of 24 follow-up surveys. With an expected 65% retention between baseline survey and follow-up survey engagement, and assuming approximately 17 surveys per person who engaged with the follow-up portion, we targeted a total observation count of approximately 3,000. After data collection, 270 United Kingdom participants were recruited via Prolific Academic, and after 5 were removed from the dataset due to not having completed any follow-up surveys, the remaining sample included 134 men and 131 women across a total of 3,999 observations (per person observations: mean = 15.09, SD = 6.45, median = 16, min = 1, max = 24). Participants were aged from 18 to 77 years with a mean of 45.94 years (SD = 12.54). Participants were compensated with monetary reward upon successful completion of each survey (conducted online). The same data screening and data check procedures as Study 1 were conducted. No data quality issues were present.
Measures
Age and Gender
Age was measured in years, and gender was coded as 0 = men, 1 = women.
Big Five Aspect Adjectives (Trait)
The BFAA trait measure and scale was as per the previous studies.
Survey Number
A within-person variable which represents the survey number from 1 to 24.
Big Five Aspect Adjectives (State)
The BFAA instructions and scale were as per Study 3 but for a 2-hr time window and measured at each follow-up survey.
Positive and Negative Affect (State)
The PANAS (Thompson, 2007) was used as per Study 3 at each follow-up survey. Omega total scores were PA = .91, NA = .94.
Number of People Interacted With
As per Study 3 but for a 2-hr time window and measured at each follow-up survey.
Estimation Procedure
The primary analysis used a multilevel multivariate model, clustered by participant, controlling for survey number at the within-person level (to account for survey fatigue), and age and gender at the between-person level. Measurement error of the BFAA traits was purged using the method described by Bollen (1996). As a follow-up test of the multilevel multivariate model, equality constraint testing was used to assess whether each trait aspect was in fact the dominant predictor of its associated state after competing with the complementary trait aspect. The constrained models forced each complementary aspect pair to be equal within each of the two equations predicting the same state pairs. For example, the first constrained model forced trait Ac and trait Ap to be equal within the equation predicting state Ac, and simultaneously forced trait Ac and trait Ap to be equal within the equation predicting state Ap. These constrained models were then compared to the unconstrained model using a chi-square test with an AIC comparison to identify which of the unconstrained versus constrained model had the better fit. Five such tests were performed, one being for each pair of aspects within each of the Big Five domains (i.e., the first forcing trait Ac to equal trait Ap, the second forcing trait Ci to equal trait Co, and so forth). For the additional validation using positive and negative affect, and number of people interacted with, we predicted all three outcome variables in a multilevel multivariate model (clustered by participant) using only the BFAA aspect states (measurement error corrected) separating variance into within-person and between-person components. Survey number, age and gender were controls. All models were constructed in R (R Core Team, 2018) using the lavaan package (Rosseel, 2012).
Results
Study 4 Correlations, Means, Standard Deviations and Internal Consistency Estimates
Note. n = 265, i = 3,999. Means and SDs calculated with all observations. Raw correlations shown below the diagonal and within-persons state correlations shown above the diagonal. BFAA = Big Five Aspect Adjectives. Gender coded as 0 = men, 1 = women. Ac = compassion, Ap = politeness, Ci = industriousness, Co = orderliness, Ea = assertiveness, Ee = enthusiasm, Nv = volatility, Nw = withdrawal, Oi = intellect, Oo = openness. Omega total scores shown in bold on the diagonal. Correlations above |.03| significant at the .05 level, and above |.04| significant at .01 level.
Study 4 Summary of Multilevel Multivariate Results at Each of the BFAA Big Five Domain and Aspect Levels
Note. Robust maximum likelihood estimation. b = unstandardized regression coefficient, SE = standard error of the estimate, β = standardized regression coefficient. ICC = intraclass correlation coefficient. All BFAA traits are corrected for measurement error. This table is a summary of the dominant predictors for each outcome shown in the full multilevel multivariate models in the Supplemental Tables.
Study 4 Summary of Significant Multilevel Multivariate Results for Outcome Variables at the BFAA State Aspect Level
Note. n = 264, i = 3,987 after listwise deletion. Robust maximum likelihood estimation. BFAA = Big Five Aspect Adjectives. (wp) = within-person, (bp) = between-person. b = unstandardized regression coefficient, SE = standard error of the estimate, β = standardized regression coefficient. All BFAA states are corrected for measurement error. Ordered by standardized regression coefficient within each outcome. This table is a summary of the full multilevel multivariate model in the Supplemental Tables.
Discussion
The present research establishes the Big Five Aspect Adjectives (BFAA) as the first validated measure of personality aspects in both trait and state forms (see Appendix for the full BFAA inventory). Critically, our findings provide the first direct empirical evidence that Big Five aspects represent psychologically meaningful units of momentary experience, not merely taxonomic conveniences for describing stable traits. This validation extends Whole Trait Theory (Fleeson, 2001; Fleeson & Gallagher, 2009; Jayawickreme et al., 2019) to the aspect level and establishes the foundation for investigating within-person personality dynamics at levels of specificity not previously accessible.
Summary of Validation Evidence for the Big Five Aspect Adjectives
In Study 3, we provided initial evidence that the BFAA could be converted to measure states using a single-occasion, between-person design (over the last 3 hr). We conducted cross-instrument validation using multivariate analysis where we showed that each of the BFAA states were best predicted by their corresponding BFAS aspect, providing powerful evidence that the BFAA is indeed a valid measure of aspects at the state level and indicating trait-state correspondence. We were also able to show the same functionality of state validity at the Big Five domain level. Study 3 also provided evidence for criterion validity in the form of momentary positive and negative affect, and the number of people interacted with in the last 3 hours. Using a 2-hr time window, Study 4 replicated the findings in Study 3 by using a full experience sampling design across up to 24 time points (3 measures per day) using both the BFAA for traits and states. Using multivariate multilevel modeling and equality constraints testing, results showed that each BFAA state was best predicted by the associated BFAA trait and that this correspondence applied at both the aspect and Big Five levels. In a further replication of Study 3, criterion validity using momentary positive and negative affect, and number of people interacted with in the last 2 hr showed that BFAA states predicted these outcomes in theoretically meaningful ways (e.g., positive affect was best predicted by Ee, and negative affect was best predicted by Nw). Consistent with Big Five trait-state literature (e.g., Fleeson & Gallagher, 2009; Jayawickreme et al., 2019), we found that despite the trait-state correspondence, meaningful variance at the within-person level remains.
Implications
There are three key implications stemming from the development of the BFAA. The first is that adjectives lend themselves well to state-based transformations. For example, the instructions can be easily changed from Please rate the degree to which the following words describe you in general followed by the adjective, to Please rate the degree to which the following words described you in the last 2 hr. Such state conversions have been used at the Big Five level for over two decades with much success (e.g., Fleeson, 2001; Jacques-Hamilton et al., 2019; McNiel & Fleeson, 2006; Margolis & Lyubomirsky, 2020; Spark & O'Connor, 2021; Zelenski et al., 2013), although often without explicit validity testing (Horstmann & Ziegler, 2020). In the current paper, we not only found good psychometric support for our novel measure but also produced evidence that dimensions of the BFAA (at both Big Five and aspect levels) are valid and meaningful at the state level. Importantly, aspects within the same domain showed differentiated patterns of prediction, for instance, Ee and Nw emerged as dominant predictors of positive and negative affect respectively, with Nv showing secondary effects, demonstrating that aspect-level measurement captures dynamics that would be obscured by domain-level analysis. Researchers interested in exploring state forms of Big Five aspects now have a method of doing so.
The second implication is that the BFAA is shorter and faster to administer than the BFAS and therefore may better enable researchers to measure all 10 aspects (as well as the Big Five traits) with minimal trade-offs. Indeed, the total number of words contained in the BFAS is 463 plus instructions, and even the shorter version, the BFAS-40, has 183 words in total. The BFAA has just 60 words.
The third implication is derived from a point made by DeYoung et al. (2007) and was the basis for many of our validation checks. That is, certain aspects appear to suppress other aspects and hence states may be able to be switched ‘on’ or ‘off’ in a targeted fashion (e.g., for a clinical intervention). Indeed, as part of our validation checks, we noted that N, despite being negatively correlated with C, has a negative relationship with Ci and a positive relationship with Co. The implication being that if Ci (but not Co) states can be enacted, then neurotic experience may decrease. Another example is E and N. E and N are negatively correlated, however the zero-order correlation between E and Nv is approximately zero, whereas the correlation between E and Nw is approximately −.40. When residualized, the correlation between E and Nv becomes positive and the correlation with Nw remains negative. Given that Nv is an outward, approach-oriented aspect and Nw is inwardly focused, it makes sense that Nv is positively correlated with E given E’s approach-oriented nature. The implication of such a relationship is that if an individual is experiencing withdrawal (depression, anxiety, etc.), a path to state E (with its associated well-being benefits; see Jacques-Hamilton et al., 2019; Margolis & Lyubomirsky, 2020) may be through Nv. That is, Nv may be relatively easy to generate in high Nw individuals, which may then be used to generate an approach-oriented behavioral strategy to generate extraverted behavior. Such a strategy is not without its risks, and we do not suggest that we have identified (and understood) these paths here. The point we make is that the relationships between aspects may create a road map that, if studied at a state level, may provide for behavioral interventions to improve outcomes for individuals in any domain of life (e.g., the workplace, romantic relationships, friendships, parenting, etc.). These are but two examples – inspection of the correlation tables reveals many more. With the ability of the BFAA to measure states conveniently, within-person testing of these nuanced relationships can now be conducted.
Limitations
We note several limitations. First, although our initial adjective pool was reasonably large and carefully selected, we may have missed other adjectives that serve as equally valid measures of the Big Five aspects. Additional testing of other adjectives is welcome. Second, despite showing similar psychometric properties in US, Australian, and UK samples, replication in other English-speaking countries may further enhance the function of the BFAA across English-speaking cultures. Third, and as an extension of our second limitation, the BFAA is only appropriate for English-speaking participants and therefore further research should be conducted to translate and extend the BFAA to non-English languages. Fourth, we relied on self-report data, which was unavoidable given that both the BFAS and BFAA are self-report instruments by design. Nevertheless, several steps were taken to mitigate self-report method risks: (i) acquiescence was adjusted for in Study 1 when factoring the initial item pool, (ii) longitudinal designs were used in Study 2 and Study 4, and (iii) cross-instrument and cross-level (i.e., trait vs state) designs were employed in Study 3 and Study 4. Fifth, as noted by DeYoung et al. (2007), we have retained the labels of the BFAS but whether they are the best labels to represent the constructs remains open to interpretation.
Conclusion
Across a series of four studies, we have presented the BFAA as an adjective-based alternative to the BFAS and we have shown that the BFAA is a reliable and valid inventory measuring the ten personality aspects within the Big Five at both trait and state levels. These findings address a fundamental question in personality science: whether aspect-level distinctions observed in stable traits represent psychologically meaningful units that operate in momentary experience. Our results, particularly the trait-state correspondence demonstrated through equality constraint testing and the differential prediction of affect by aspects within the same domain, provide the first direct evidence that aspects function as meaningful constructs at the state level, not merely as taxonomic conveniences for describing traits. By equipping researchers with this tool, we hope that the BFAA will facilitate new lines of research that allow for greater insight as to how personality influences psychological phenomena at a more fine-grained level not previously accessible.
Supplemental Material
Supplemental Material - Measuring Personality States Below the Big Five: Development and Validation of the Big Five Aspect Adjectives for Traits and States
Supplemental Material for Measuring Personality States Below the Big Five: Development and Validation of the Big Five Aspect Adjectives for Traits and States by Andrew Spark, Sarah A. Walker, Peter J. O’Connor, and Nerina L. Jimmieson in Personality Science
Supplemental Material
Supplemental Material - Measuring Personality States Below the Big Five: Development and Validation of the Big Five Aspect Adjectives for Traits and States
Supplemental Material for Measuring Personality States Below the Big Five: Development and Validation of the Big Five Aspect Adjectives for Traits and States by Andrew Spark, Sarah A. Walker, Peter J. O’Connor, and Nerina L. Jimmieson in Personality Science
Footnotes
Author Note
Study 1, Study 2, and Study 3 ethics approval granted by The University of Sydney Human Research Ethics Committee [2020/307]. Study 4 ethics approval granted by Durham University Education Ethics Committee [EDU-2025-9876-9901].
Acknowledgements
Not applicable.
Author Contributions
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was partially funded by the Australian Government through the Australian Research Council’s Discovery Projects funding scheme (project DP190100848), and the Queensland University of Technology School of Management.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Open Science Statement
This article has received the badge for Open Data and Open Materials. No studies were preregistered.
Not applicable.
Supplemental Material
Supplemental material for this article is available online. Depending on the article type, these usually include a Transparency Checklist, a Transparent Peer Review File, and optional materials from the authors.
Appendix
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
