Validating the Structure of Narcissism: A Replication and Extension of Crowe et al. (2019)

Narcissism is a personality construct that is linked to dysfunction in several domains and has been studied across many subfields of psychology for a century. Narcissism encompasses two primary dimensions: grandiose narcissism and vulnerable narcissism. The grandiose dimension is characterized by grandiosity, assertiveness, manipulativeness, and entitlement, whereas the vulnerable dimension includes features of egocentrism, distrust, negative emotionality, and need for admiration (Miller et al., 2017; Crowe et al., 2019). While the grandiose and vulnerable narcissism dimensions can represent unique presentations of narcissism (although these dimensions can both be elevated in individuals), they are both characterized by antagonistic traits, such as entitlement and callousness (Crowe et al., 2019; Miller et al., 2017; Miller & Campbell, 2008; Wink, 1991). Factor analyses of narcissism measures have uncovered three higher-order factors: agentic extraversion/grandiosity; antagonism/entitlement; neuroticism/vulnerability (Crowe et al., 2019; Krizan & Herlache, 2018; Miller et al., 2016, 2017). Researchers have conceptualized these three factors as the trifurcated model of narcissism (Miller et al., 2016, 2017) and the narcissistic spectrum model (Krizan & Herlache, 2018). These nuanced, multidimensional models of narcissism are useful for understanding the myriad and variable relations found between total narcissism scores and other important constructs (e.g., self-esteem, boldness, psychopathy, aggression; Du et al., 2023; Hyatt et al., 2018; Miller et al., 2016, 2020; Weiss et al., 2021) as well as how extant narcissism scales differ in the conceptualizations of narcissism that they measure (Packer West et al., 2023).

Crowe et al. (2019) conducted a comprehensive examination of the structure of narcissism by subjecting 215 items drawn from 12 of the most commonly used, full-length narcissism measures (e.g., Narcissistic Personality Inventory, Raskin & Terry, 1988; The Five-Factor Narcissism Inventory Short Form, Sherman et al., 2015) to an item-level bass-ackward factor analysis (see Crowe et al., 2019; Goldberg, 2006). This method allows construction of a hierarchical model of personality by conducting stepwise factor analyses, starting from one broad factor to more specific ones. Each level’s factor scores are saved, allowing comparison between adjacent levels. This approach captures the broad-to-narrow emergence of traits. Crowe et al.’s (2019) two‐factor solution revealed two moderately correlated factors consistent with grandiose and vulnerable narcissism, as determined by their highest loading items (e.g., “I am superior” for Grandiose Narcissism; “It’s hard to feel good about myself unless I know other people admire me” for Vulnerable Narcissism) and their relations with external criteria (e.g., low Agreeableness and high Extraversion for Grandiose Narcissism; strong positive associations with Neuroticism and a negative association with Agreeableness for Vulnerable Narcissism). The third level of analysis revealed a factor solution consistent with the trifurcated model of narcissism (and the narcissistic spectrum model, which suggests a similar structure but with different labels), with highest loading items and external criteria relations indicating Self-centered Antagonism (e.g., “It’s fine to take advantage of persons to get ahead;” negative associations with Dutifulness and all facets of Agreeableness), Agentic Extraversion (e.g., “I see myself as a good leader;” strong relations with Modesty and Assertiveness), and Narcissistic Neuroticism (e.g., “I wish I didn’t care so much about what others think of me;” strong relations with anxiety and depression). The present study sought to build upon Crowe et al. (2019) by testing whether the three-factor structure (Antagonism, Agentic Extraversion, and Neuroticism) can be replicated using these short-form measures. Uncovering this same factor structure of narcissism found by Crowe et al. (2019) in short forms, which can leave out important content, would be an important test of the robustness of this structure.

In an online community sample, the current study employed the same bass-ackward factor analytic approach used by Crowe et al. (2019) to uncover and confirm coverage of narcissism among short forms as a whole as well as to compare resultant factors’ relations to external criteria (e.g., self-esteem, antisocial behavior) and factor loadings for each level to those of Crowe et al. (2019). We hypothesized that the three-factor structure of narcissism found by Crowe et al. (2019; i.e., Antagonism, Agentic Extraversion, and Neuroticism), as well as the two-factor structure (i.e., Grandiose Narcissism and Vulnerable Narcissism), would be replicated utilizing a smaller, possibly impoverished, subset of the narcissism items used by Crowe et al. (2019). Specifically, using widely employed short-form narcissism measures, many of which are brief versions of the full-form measures used in Crowe et al. (2019), we expected that the short-form measures would yield factor loadings and factor intercorrelations similar to those found in the original study. Additionally, we hypothesized that the three-factor structure of the short-form measures would demonstrate similar correlation patterns with external criteria (e.g., self-esteem, externalizing behaviors, emotional distress) as those reported by Crowe et al. (2019).

The primary hypotheses and methodological approach for the present study were preregistered before commencing data collection and can be found at: https://​osf.​io/​xegv3. This sample has been used for previously published, preregistered work for research questions separate from the current study (Packer West et al., 2023). The study was approved by Purdue University’s Institutional Review Board (IRB-2021-347) prior to data collection, and informed consent was obtained from research participants at the beginning of the online survey.

Correlations of the 122 brief narcissism measure items were inspected to find any identical or excessively overlapping (i.e., rs ≥ 0.65) items. One hundred and thirty-six item pairs (out of 14,762 possible pairs) were identified with correlations greater than 0.65 (see Crowe et al. (2018) for explanation of similar method). To minimize bias and maximize the final number of saved items, we deleted the items with the most correlations above 0.65 until there were no correlations greater than 0.65. Forty-three items were removed, leaving 79 items in the pool (see Supplementary Material for items, their factor loadings, and descriptive statistics). Sixty-two out of the 79 items used in the current study’s pool were used in Crowe et al. (2019).

Next, a structural analysis was conducted with the remaining items; all factor solutions were identified employing the principal axis factoring method; when multiple factors were extracted, promax rotation was employed. A single unrotated factor was extracted, which accounted for 27% of the total variance. The first 10 eigenvalues were as follows: 21.46, 10.47, 3.90, 2.08, 1.80, 1.67, 1.54, 1.47, 1.33, and 1.27. Velicer’s minimum average partial (MAP) test and a parallel analysis were conducted to help guide factor extraction (O’Connor, 2000). The MAP test suggested 9 to 10 factors; 10 according to the Original (1976) MAP Test and 9 according to the Revised (2000) MAP Test (Velicer, 1976; Velicer et al., 2000). The parallel analysis (Horn, 1965) suggested 10 factors.

In the five-factor solution, only one item had its highest loading on the fifth factor. Similarly, only one item had its highest loading on the fourth factor in the four-factor solution and thus was not interpretable. The three‐factor solution was therefore the largest model given consideration. The content of each of the factors through the three‐factor solution was interpretable, and the item loadings were generally high. The three‐factor solution was chosen as the base of the narcissism hierarchy. The three‐factor solution accounted for 45% of the variance in narcissism scores; the two-factor solution accounted for 40% of the variance.

Narcissism is connected to dysfunction in several domains, including aggressive and risk-taking behavior (Kealy et al., 2017), cognitive biases (e.g., Rhodewalt & Morf, 1998), social relationships (e.g., Miller et al., 2007), alcohol and drug overuse (Kealy et al., 2017), and psychopathology (Pincus et al., 2009). Recently, three factor models of narcissism have become ascendant as particularly promising in representing the most critical core domains of narcissism in a more articulated manner than two or one factor models (Miller et al., 2021). The present study sought to evaluate whether this newly proposed factor structure replicates in a subset of items from the Crowe et al. (2019) study. Our subset included 79 items from 122 possible items (i.e., before an item from each highly correlated item pair was removed) across 11 measures whereas the final narcissism pool for Crowe et al. (2019) included 215 items out of 280 (i.e., after an item from each identically worded item pair was removed but before an item from each highly correlated item pair was removed) possible items across 12 measures. Sixty-two out of the 79 items used in the current study’s pool were used in the final narcissism pool of Crowe et al. (2019). Our final pool included 17 items not found in that of Crowe and colleagues (2019), but these mostly came from NVS and DD-N, two measures not used in their study.

Both the two and three-factor solutions found by Crowe et al. (2019) using long narcissism measures were very well replicated in our current sample of brief measures, as hypothesized. Our narcissism factors showed similar relations to narcissism measures and to external criteria as those for Crowe et al. (2019), with median ICCs of 0.94 and 0.95 across factor solution levels, respectively. The short-form versus long-form extracted factors between our sample and Crowe et al.’s (2019) sample can be considered equal, as all Tucker’s congruence coefficients were at least 0.95. Finally, the narcissism pool items in common with those used in Crowe et al. (2019) mostly fell in the same factors as in theirs, with 89% agreement at the two-factor level and 85% agreement at the three-factor level.

The items that were retained in our narcissism item pool after removing overlapping items were almost the same as those retained by Crowe et al. (2019), which is impressive given that the narcissism measures used in each study differed to some extent. The short form items used for the current bass-ackward factor analysis successfully replicated the factor structure of narcissism found by Crowe et al. (2019) using longer narcissism measures at both the two- and three-factor level, which suggests that coverage of narcissism in terms of structure and breadth was not adversely impacted by the short form items overall. In fact, the current study’s three-factor solution explained more variance (45%) than Crowe et al.’s (2019) three-factor solution (34%); this may be because the retained items are particularly efficient at capturing narcissism’s core variance while minimizing measurement noise rendered by additional or peripheral items in full-length narcissism measures. The factors generated from the short form items also showed similar relations with narcissism measures and key psychological outcomes, such as self-esteem and externalizing behaviors, mirroring correlational patterns found in Crowe et al. (2019) and lending to the external validity of the factor structure rendered by the short narcissism forms. These findings suggest that narcissism super-short forms collectively retain the multifaceted structure of narcissism, including at the three-factor level. Of note, this full multifaceted structure is only captured by one extant brief narcissism measure by itself, the FFNI-SSF. Thus, we advise that researchers who want to measure the full narcissism construct, and in a very short amount of time, use the FFNI-SSF (although other “stacked” assessment approaches are possible; e.g., Schneider et al., 2023). Future research could also examine the performance of a measure composed from the highest loading items from across multiple brief narcissism measures in our three-factor solution (e.g., Bauditz et al., 2024, 2025).

The current study’s validation of the three-factor structure of narcissism, namely, that narcissism consists of Self-centered Antagonism, Agentic Extraversion, and Narcissistic Neuroticism, is significant for several reasons. First, it supports the use of multidimensional measurement of narcissism. There are several potential benefits of breaking personality constructs, including narcissism, down into their constituent elements, including when studying them in relation to relevant constructs (Lynam & Miller, 2015). First, it enables comparison of extant narcissism measures. Studies on narcissism evince varied relations between narcissism and relevant outcomes depending on which instruments are used—thinking of narcissism multidimensionally and understanding the underlying structure of a given measure helps unites findings across studies using different measures. For example, a common difference in narcissism measures is the degree to which they emphasize vulnerability/hypersensitivity. A multidimensional measurement approach facilitates understanding of the factor structures of extant narcissism measures and therefore how they measure narcissism. Second, this approach provides a parsimonious way to study the epidemiology of narcissism, including comorbidities (e.g., narcissism is comorbid with other personality disorders to the degree that they measure similar traits) and how it differs across important demographics, such as gender and age. Third, it allows for understanding of the underlying mechanisms of narcissism, including deficits (e.g., affective, interpersonal) and advantages that lead some narcissistic individuals to function reasonably well in certain contexts. Fourth, it permits a bottom-up approach to understanding and refining the construct of narcissism through the examination of which elements are unnecessary, peripheral, and central to narcissism (Lynam & Miller, 2015). Fifth, in the case of the FFNI, it connects narcissism to the robust, highly replicated FFM of personality (Costa & McCrae, 1992; Jones et al., 2022) and its abundant literature on pathological and basic personality traits. The three-factor, or trifurcated, model of narcissism replicated in the current study largely aligns with basic trait models of personality disorders (PDs) and supports the theory that PDs are essentially composites of basic traits.

The current study has limitations worth noting. The study used only self-report measures. Especially when measuring a personality construct in which insight may be lacking, such as narcissism, it is valuable to garner informant accounts to corroborate survey ratings. Additionally, the current study sample comprised primarily “WEIRD” (i.e., White, Educated, Industrialized, Rich, and Democratic; Henrich et al., 2010) participants, which may constrain the representativeness of our sample and conclusions. Future research should further examine the factor structure of narcissism in more diverse populations.