A Multivariate Analysis of the Pathological Narcissism Inventory’s Nomological Network

Abstract

The Pathological Narcissism Inventory (PNI) has enjoyed widespread use in the study of the narcissism. However, questions have been raised about whether the PNI’s grandiosity scale adequately captures narcissistic grandiosity as well as other popular measures do. Specifically, some have noted that PNI grandiosity shows a pattern of external associations that diverges from patterns for narcissistic grandiosity predicted by experts, and is more similar to the predictions for the vulnerability scale than is desirable. Previous research driving these critiques has relied on patterns of zero-order correlations to examine the nomological networks of these scales. The present study reexamines the nomological networks of PNI grandiosity and vulnerability scales using hierarchical regression. Results indicate that once accounting for overlapping variance of vulnerability and grandiosity, the unique variance in the PNI’s grandiosity scale closely matches contemporary expert conceptualizations of narcissistic grandiosity based on expected associations with other personality variables.

Keywords

Pathological Narcissism Inventory narcissistic grandiosity narcissistic vulnerability narcissistic personality disorder narcissism measurement

In recent years, there has been a surge in the number of studies examining narcissism and its measurement (Miller, Lynam, Hyatt, & Campbell, 2017; Wright & Edershile, 2018). Although the precise definition of narcissism and how best to study the concept remains somewhat unclear, it is generally accepted that narcissism consists of both grandiose and vulnerable elements. Narcissistic grandiosity encompasses an inflated sense of self, lack of empathy, and entitlement (e.g., Cain, Pincus, & Ansell, 2008). Narcissistic vulnerability, on the other hand, is characterized by feelings of inferiority, fragile and contingent self-esteem, and a constant need for validation (e.g., Cain et al., 2008; Morf, 2006). A host of measures have been developed to capture the diverse components of narcissistic grandiosity and vulnerability (e.g., Back et al., 2013; Glover, Miller, Lynam, Crego, & Widiger, 2012; Pincus et al., 2009). However, due to the lack of clarity of the definition of narcissism, the ways in which these measures define narcissism’s grandiose and vulnerable components differ, leaving the field with a lack of convergence among measures.

The Pathological Narcissism Inventory (PNI; Pincus et al., 2009) is a popular measure often used to assess pathological narcissism. Recently, questions have been raised regarding how well the PNI scales differentiate between the grandiose and vulnerable aspects of narcissism (Miller et al., 2014; Miller, Lynam, & Campbell, 2016; Thomas, Wright, Lukowitsky, Donnellan, & Hopwood, 2012, 2016). Specifically, the critiques have focused on the way in which PNI operationalizes narcissistic grandiosity (Wright, Lukowitsky, Pincus, & Conroy, 2010). For example, Miller et al. (2014, 2016) have suggested that, because the PNI was developed to target narcissism as manifested in clinical settings, the measure does not represent narcissistic grandiosity in its prototypical manifestation (e.g., inflated sense of self, high self-esteem), but rather overemphasizes fragility and underrepresents antagonism (Miller et al., 2014). In capturing the maladaptive presentation of narcissism, as Miller et al. (2014, 2016) note, the PNI’s grandiosity (PNI-G) and vulnerability (PNI-V) scales covary excessively, resulting in a nomological network that lacks discriminant validity. PNI-G has received the brunt of the criticism because experts argue that grandiosity should have a well-demarcated pattern of associations with basic and maladaptive personality traits, emphasizing interpersonal antagonism, and deemphasizing distress. Narcissistic vulnerability, on the other hand, is accepted to have a diffuse pattern of associations with distress and maladaptivity, regardless of the measure used (Miller et al., 2014), including the PNI-V scale.

These critiques arise from examining correlations between PNI-G and a host of personality variables. For the purposes of the present study, we will focus our review of prior findings on the PNI-G scale’s pattern of association with personality pathology as well as the five-factor model (FFM) traits. In a pair of studies using undergraduate participants, Miller et al. (2014) compared the observed patterns of correlations of the PNI-G scale and several measures of personality and its pathology and expert ratings of narcissistic grandiosity’s association with those measures. Based on modest profile correlations (e.g., −.03, −.28 on the Big Five Inventory) between the observed and predicted values, they concluded that the PNI does not adequately capture narcissistic grandiosity. Moreover, Miller et al. suggested that the PNI-G scale better reflects a mixture of prototypical grandiosity and vulnerability that is more evident in other personality disorders (PDs), such as borderline personality disorder (BPD) rather than narcissism per se. If the PNI adequately reflects grandiosity and vulnerability, the grandiosity component should mostly be unrelated to other forms of personality pathology (Miller et al., 2014). Furthermore, they demonstrate that the PNI-G scale’s pattern of association with other personality scales is atypical of grandiose narcissism. Specifically, it does not align with the FFM domains and the 25 DSM-Section III PD traits (Krueger, Derringer, Markon, Watson, & Skodol, 2012) in ways that one would intuit. A similar pattern was independently found by Thomas et al. (2012, 2016), who found that expert and nonexpert raters both were much better at predicting PNI-V correlations with a variety of different personality measures than they were at predicting PNI-G’s associations with the same measures.

The above-referenced literature relied on patterns of zero-order correlations to understand the association among PNI-G, PNI-V, and normal and maladaptive personality. However, univariate analytic approaches may fail to reveal the full picture of the PNI’s scales’ performance. Because the PNI was developed to capture maladaptive manifestations of narcissism, it likely captures not only prototypical grandiosity but also its concomitant dysfunction including distress; that is, the PNI-G scale is complex, with multiple sources of variance contributing to its total score: grandiosity and its resultant impairments. When relying on univariate analyses, this yields a pattern of broader associations between the PNI-G scale and other forms of personality pathology (including vulnerability). Alternatively, it is possible, as Miller et al. (2014, 2016) suggest, that the PNI-G scale does not contain the core of grandiosity and that the observed pattern of correlations reflects a less well differentiated construct that departs from common themes of grandiosity. To adjudicate between these possibilities, we adopted a multivariate analytic approach to isolate the patterns of unique associations of both the PNI’s domain scales (grandiosity and vulnerability), accounting for their shared maladaptive variance. Thus, to clarify the unique contributions of the PNI’s articulation of grandiose and vulnerable narcissism, we ran univariate and multivariate regressions predicting each of the 10 Diagnostic and Statistical Manual of Mental Disorders–Fifth edition (DSM-5) Section II PDs, each of the FFM domain scores, and each of the 25 DSM-Section III PD traits (Krueger et al., 2012). We regressed criterion counts from each of the DSM-5 Section II PDs, each of the FFM domain scores, and each of the 25 DSM-5 Section III facets on PNI-V and PNI-G, first with each as a single predictor and then both entered together as predictors, to directly address previous findings (Miller et al., 2014, Miller et al., 2016; Thomas et al., 2012, 2016) in which PNI-G did not behave as anticipated for a grandiosity scale.

We hypothesized that the pattern of coefficients that would emerge from the univariate regressions would replicate the observed profiles of correlations in Miller et al. (2014) and Thomas et al. (2016; e.g., PNI-G will have positive associations with BPD and Neuroticism). Specifically, we predicted that both PNI-G and PNI-V will have a relatively diffuse pattern of association with the various outcomes. In contrast, we hypothesized that the pattern of coefficients that would emerge from multivariate regressions will provide a closer match to the expected patterns of association for grandiosity. We hypothesized that once controlling for the shared variance in PNI-G and PNI-V, PNI-G would have little association, or even negative associations, with most PDs. The exceptions, we hypothesized, would be positive associations between PNI-G and narcissistic and histrionic PD dimensions. We expected that the need for attention, common in both disorders, would associate with the grandiose themes from the PNI-G scale. Furthermore, we hypothesized the multivariate coefficients when predicting the FFM domains would be more similar to expert generated profiles. We predicted the same trend for the 25 scales and the DSM-5 Section III traits. Finally, we predicted PNI-V to maintain a relatively undifferentiated pattern of association in the multivariate analyses.

Method

Two data sets were used to address the research questions. A clinical data set was used to examine the associations among the PNI-G, PNI-V, FFM domains, and each of the DSM-5 Section II 10 PDs. For complete methods and procedures of this data collection, please refer to Wright and Simms (2014, 2015). The second data set was an undergraduate student sample that included a measure of the DSM-5 Section III traits. For complete methods and procedures of this data collection, refer to Wright et al. (2013).

Participants and Procedures

In the clinical data set, participants were recruited using flyers posted around mental health clinics across Western New York. To participate, participants had to be at least 18 years of age and had received psychiatric treatment within the past 2 years. The final sample included 628 participants (M_age = 43.2; 63.5% female). Eighty percentage of the individuals in the sample were currently in treatment, 10% had sought treatment within the last year, and 5% had been in treatment within the past 2 years. The measures of interest fell toward the end of a lengthy clinical interview and self-report protocol. Thus, when participants progressed at a slow pace, at times they were unable to complete the study procedures. Participants were removed from analysis if they did not complete any of the measures used in analyses. Of the original 628 participants, 288 had complete data and were included in the regression models. Thus, due to time intensive procedures, there were large amounts of missing data in the psychiatric sample.¹ Of the 288 participants, 67% were female and the average age was 38.6.

The undergraduate sample was collected at a large public university. Participants (N = 1,927) completed self-report questionnaires online for course credit. A total of 1,653 participants returned questionnaires with fewer than 10% missing items and yielded scores within 2.5 standard deviations of the community average on a measure of random or careless responding (Personality Assessment Inventory Infrequency Scale; Morey, 1991). These participants were retained for analyses. The average age was 18.8 years (range = 18-56 years) and 66% female. Data and syntax are openly available at this link: https://osf.io/9wdju/

Measures

NEO Personality Inventory–3–First Half (NEO-PI-3FH; McCrae & Costa, 2007)

This measure was administered in the clinical sample. The NEO-PI-3 (McCrae, Costa, & Martin, 2005) is a more recent version of the NEO-PI-R (Costa & McCrae, 1992). Like the NEO-PI-R, The NEO-PI-3 is a 240-item personality measures that assess the FFM personality factors: Openness to Experience, Agreeableness, Extraversion, Neuroticism, and Conscientiousness. For the purposes of this study, a short version of the NEO-PI-3 was used. This measure is the first half of the NEO-PI-3 (NEO-PI-3FH; McCrae & Costa, 2007) and contains 120 items. The five domain scales were scored for use in this study, with α values that ranged from .80 to .90.

Structured Clinical Interview for DSM-IV-TR Personality Disorders (SCID-II; First, Spitzer, Gibbon, & Williams, 2002)

Dimensional counts of PD diagnostic features were evaluated in the clinical sample by using a modified protocol of the SCID-II (First et al., 2002). This structured interview is based on the DSM-5; American Psychiatric Association, 2013). Participants first completed the SCID-II personality questionnaire and then were interviewed to verify the presence of individual criteria. Interviews were administered by trained clinical psychology doctoral students. Kappas were evaluated at the disorder level in Wright and Simms (2015; Mdn = .96; range = .66-1.00). In the current study, dimensional criteria counts for each diagnosis were used.

The Pathological Narcissism Inventory (Pincus et al., 2009)

The PNI is a 52-item self-report measure. The current study used the higher order dimensions (Wright et al., 2010) of narcissistic grandiosity (PNI-G; Exploitativeness, Grandiose Fantasy, Self-sacrificing Self-Enhancement) and narcissistic vulnerability (PNI-V; Entitlement Rage, Contingent Self-esteem, Hiding the Self, Devaluing). Internal consistencies for the PNI-G and PNI-V scales were high (αs = .88 and .95, respectively). This measure was used in the clinical and undergraduate sample.

The Personality Inventory for DSM–5 (Krueger et al., 2012)

The Personality Inventory for DSM–5 (PID-5) is a 220-item questionnaire with a 4-point response scale that measures the proposed DSM-5 traits and was administered in the undergraduate sample. Of interest was the 25 primary scales. Internal consistencies of the scales were adequate to high in the current sample (Mdn α = .86; range = .73-.95).

Data Analytic Plan

For all outcomes, which included the 10 DSM-5 Section II PDs, each of the FFM domain scores, and each of the 25 DSM-Section III PD traits (Krueger et al., 2012), we ran univariate and multivariate regressions with PNI-G and PNI-V as predictors. In the univariate models, PNI-G and PNI-V were entered in separate models. The multivariate model entered PNI-G and PNI-V simultaneously, thereby adjusting for shared variance in their scores. Multivariate models were run twice to determine the unique contribution of each scale as reflected in R² changes from the two univariate models (e.g., Model 1: PNI-G; Model 2: PNI-G + PNI-V). As an additional metric of profile match, profile correlations were calculated for the FFM and PID-5. Profile correlations for the FFM and PID-5 were taken from Thomas et al., 2012 and Samuel, Lynam, Widiger, and Ball (2012), respectively.

Results

Univariate and Multivariate Regression Results

Though PNI-G and PNI-V were strongly correlated in both the clinical and undergraduate sample (rs = .63 in both samples), the results suggest that there is enough independent variance between the two to accurately capture the differences. Specifically, this means that regression analyses will allow us to examine the individual effect of each as they predict the 10 PDs dimensional counts, the FFM domains, and the 25 PID-5 scales.

Regression Results for Predicting the 10 PDs

Standardized regression coefficients (βs) for PNI-G and PNI-V, 95% confidence intervals, R² values, and R² change values (i.e., unique variance accounted for) for the models predicting the 10 DSM-5 Section II PD dimensional counts are reported in Table 1. Due to the large number of results, we will highlight certain findings in the text. However, please refer to the tables for complete patterns of associations.

Table 1.

Hierarchical Model Building With PNI Grandiosity and Vulnerability Predicting PD Criteria Dimensions.

Predictor/disorder	Model 1			Model 2
Predictor/disorder	β	CI (β)	R ²	β	CI (β)	ΔR²
Avoidant
Grandiosity	.05	[−.07, .17]	.00	−.36***	[−.49, −.23]	.08***
Vulnerability	.43***	[.32, .53]	.18***	.65***	[.52, .78]	.26***
Dependent
Grandiosity	.18**	[.06, .29]	.03**	−.07	[−.21, .08]	.00
Vulnerability	.35***	[.24, .46]	.12***	.39***	[.25, .53]	.09***
Obsessive–Compulsive
Grandiosity	.18**	[.07, .30]	.03**	−.03	[−.17, .11]	.00
Vulnerability	.32***	[.21, .43]	.10***	.34***	[.19, .48]	.07***
Narcissism
Grandiosity	.48***	[.38, .59]	.23***	.33***	[.20, .46]	.07***
Vulnerability	.45***	[.35, .55]	.20***	.24***	[.12, .37]	.04***
Borderline
Grandiosity	.30***	[.19, .42]	.09***	−.07	[−.20, .05]	.00
Vulnerability	.55***	[.45, .65]	.30***	.60***	[.47, .72]	.21***
Histrionic
Grandiosity	.49***	[.39, .59]	.24***	.50***	[.37, .63]	.15***
Vulnerability	.30***	[.19, .41]	.09***	−.02	[−.15, .11]	.00
Antisocial
Grandiosity	.24***	[.12, .35]	.06***	.16*	[.02, .31]	.02*
Vulnerability	.22***	[.10, .33]	.05***	.12	[−.03, .26]	.01
Paranoid
Grandiosity	.24***	[.13, .36]	.06***	−.14*	[−.26, −.01]	.01*
Vulnerability	.52***	[.42, .62]	.27 ***	.60***	[.47, .73]	.22***
Schizotypal
Grandiosity	.35***	[.24, .46]	.12***	.12	[−.01, .25]	.01
Vulnerability	.44***	[.34, .55]	.20***	.37***	[.23, .50]	.08***
Schizoid
Grandiosity	−.05	[−.17, .07]	.00	−.20**	[−.35, −.05]	.02**
Vulnerability	.11	[−.01, .23]	.01	.24**	[.09, .38]	.03**

Note. N = 288. PNI = Pathological Narcissism Inventory; PDs = personality disorders; CI = confidence interval. Model 1 reflects the standardized betas, CIs for the unstandardized betas, and R² values when only the given predictor was entered into the model; Model 2 reflects the standardized betas, CIs for the unstandardized betas, and R² change values when both grandiosity and vulnerability were entered into the model at the same time. The R² change value is the value when the given predictor is added to the model.

p < .05. **p < .01. ***p < .001.

PNI-G yielded significant positive associations with each of the PD dimensions, except for avoidant PD (β = .05, p = .412) and schizoid PD (β = −.05, p = .390) when entered into the regressions as the sole predictor. However, when PNI-V was added to the models in multivariate analyses, PNI-G showed a more circumscribed pattern of associations. Specifically, significant positive associations emerged only with narcissism, histrionic, and antisocial PD dimensions (β = .33, p < .001; β = .50, p < .001; β = .16, p = .028, respectively). Also, suppression effects emerged such that some significant positive associations or nonsignificant associations in the univariate models changed to significant negative associations in the multivariate models. This was true for avoidant, paranoid, and schizoid (β = −.36, p < .001; β = −.14, p = .037; β = −.20, p = .008, respectively).

PNI-V showed similar patterns of association as PNI-G in the univariate models. Specifically, significant associations were found for each of the PDs except for schizoid (β = .11, p = .062). When PNI-G was added into the regression equations in multivariate models, the PNI-V associations with histrionic and antisocial (β = −.02, p = .775; β = .12, p = .119) were no longer significant. Furthermore, a significantly positive association between PNI-V and schizoid (β = .24, p = .002) emerged.

In sum, after accounting for the shared variance between PNI-V and PNI-G (by taking a multivariate approach), PNI-V continues to be positively associated with a wide range of PDs, whereas PNI-G has only positive associations with narcissism, histrionic, and antisocial PDs.

At the request of an anonymous reviewer, in a separate set of analyses, we reran regressions with PNI-G and Neuroticism as predictors in multivariate regressions. Patterns emerge such that PNI-G has similar associations to that of the univariate model. For example, PNI-G continues to have significant positive associations with obsessive–compulsive personality disorder (β = .15, p = .010), BPD (β = .19, p < .001), Schizotypal (β = .29, p < .001), and others for which the significant positive associations disappear once accounting for the shared variance between PNI-G and PNI-V. Thus, the change in effects we observe when controlling for PNI-V do not appear to extend to all measures of dispositional distress/negative emotionality.

Regression Results for the FFM

We next examined PNI-G and PNI-V as they predicted each of the FFM domains. Regression results can be found in Table 2. Examining PNI-G’s associations from a univariate approach, we see significant positive associations with Extraversion, Neuroticism, and Openness (β = .16, p = .007; β = .19, p = .001; β = .14, p = .016, respectively), and significant negative associations with Conscientiousness and Agreeableness (β = −.19, p = .002; β = −.42, p < .001, respectively). Moving to the multivariate models, apart from Conscientiousness (ΔR² = .01, p = .061), the effect size for the unique association between PNI-G and each of the FFM increases relative to the univariate models, now exhibiting a marked positive association with Extraversion. Additionally, a now negative significant association with Neuroticism (β = −.33, p < .001) emerges.

Table 2.

Hierarchical Model Building With PNI Grandiosity and Vulnerability Predicting Each of the FFM.

	Model 1			Model 2
Predictor/outcome	β	CI (β)	R ²	β	CI (β)	ΔR²
Extraversion
Grandiosity	.16**	[.04, .27]	.03**	.56***	[.43, .69]	.19***
Vulnerability	−.29***	[−.40, −.17]	.08***	−.64***	[−.77, −.51]	.25***
Conscientiousness
Grandiosity	−.19***	[−.30, −.07]	.04**	.13	[−.01, .27]	.01
Vulnerability	−.42***	[−.53, −.31]	.18***	−.50***	[−.64, −.37]	.15***
Agreeableness
Grandiosity	−.42***	[−.52, −.31]	.17***	−.22**	[−.36, −.09]	.03**
Vulnerability	−.45***	[−.55, −.34]	.20***	−.31***	[−.44, −.17]	.06***
Neuroticism
Grandiosity	.19**	[.08, .31]	.04**	−.33***	[−.44, −.21]	.06***
Vulnerability	.62***	[.53, .71]	.38***	.82***	[.71, .94]	.41***
Openness
Grandiosity	.14*	[.03, .26]	.02*	.34**	[.20, .49]	.07***
Vulnerability	−.11	[−.22, .01]	.01	−.32**	[−.46, −.18]	.06***

Note. N = 288. PNI = Pathological Narcissism Inventory; FFM = five-factor model; CI = confidence interval. Model 1 reflects the standardized betas, CIs for the unstandardized betas, and R² values when only the given predictor was entered into the model; Model 2 reflects the standardized betas, CIs for the standardized betas, and R² change values when both grandiosity and vulnerability were entered into the model at the same time. The R² change value is the value when the given predictor is added to the model.

p < .05. **p < .01. ***p < .001.

In univariate models, PNI-V has significant negative associations with each of the FFM except for Neuroticism (β = .62, p < .001). Moving to a multivariate approach, PNI-V maintains the univariate associations and takes on a significant negative association with openness (β = −.32, p < .001).

In sum, using a multivariate approach to examine the association between PNI-V, PNI-G, and the FFM yields a more distinct pattern of association with PNI-G that has thus far been overlooked in prior univariate analyses.

Regression Results for the PID-5

Patterns of associations between PNI-G and PNI-V and the 25 PID-5 scales can be found in Table 3. With respect to PNI-G, the univariate analyses demonstrate significant positive associations with each of the 25 scales except Intimacy Avoidance (β = .04, p = .112). Moving to a multivariate approach, holding PNI-V constant, a less diffuse pattern emerges such that many of the 25 scales no longer have a significant positive association with PNI-G (e.g., Restricted Affectivity; β = −.02, p = .607). Additionally, many of the effects exhibit suppression and take on a significantly negative association with PNI-G (e.g., Anhedonia; β = −.32,p < .001).

Table 3.

Hierarchical Model Building With PNI Grandiosity and Vulnerability Predicting Each of the PID-5 Scales.

Predictor/PID-5 Scale	Model 1			Model 2
Predictor/PID-5 Scale	β	CI (β)	R ²	β	CI (β)	ΔR²
Submissive
Grandiosity	.24***	[.19, .28]	.06***	−.04	[−.10, .01]	.00
Vulnerability	.39***	[.35, .43]	.18***	.42***	[.37, .47]	.12***
Depressivity
Grandiosity	.17***	[.13, .22]	.03***	−.25***	[−.30, −.20]	.04***
Vulnerability	.49***	[.45, .52]	.28***	.64***	[.59, 69]	.29***
Separation Insecurity
Grandiosity	.30***	[.25, .34]	.09***	−.06*	[−.11, −.01]	.00*
Vulnerability	.50***	[.46, .54]	.29***	.54***	[.49, .59]	.20***
Anxiousness
Grandiosity	.28***	[.23, .32]	.08***	−.13***	[−.18, −.08]	.01***
Vulnerability	.54***	[.50, .57]	.33***	.62***	[.57, .66]	.26***
Emotional Lability
Grandiosity	.28***	[.23, .32]	.08***	−.09**	[−.14, −.04]	.01**
Vulnerability	.50***	[.46, .53]	.29***	.55***	[.50, .60]	.21***
Suspiciousness
Grandiosity	.27***	[.22, .31]	.08***	−.11***	[−.16, −.05]	.01***
Vulnerability	.50***	[.46, .54]	.29***	.56***	[.51, .61]	.22***
Restricted Affectivity
Grandiosity	.10***	[.05, .15]	.01***	−.02	[−.08, .04]	.00
Vulnerability	.17***	[.12, .21]	.03***	.18***	[.12, .23]	.02***
Withdrawal
Grandiosity	.09***	[.04, .14]	.01***	−.26***	[−.31, −.20]	.04***
Vulnerability	.36***	[.32, .40]	.15***	.52***	[.47, .57]	.19***
Intimacy Avoidance
Grandiosity	.04	[−.01, .09]	.00	−.15***	[−.21, −.10]	.02***
Vulnerability	.20***	[.15, .24]	.05***	.29***	[.23, 35]	.06***
Anhedonia
Grandiosity	.10***	[.05, .14]	.01***	−.32***	[−.37, −.27]	.07***
Vulnerability	.44***	[.40, 48]	.22***	.63***	[.58, .68]	.28***
Manipulativeness
Grandiosity	.47***	[.43, .51]	.23***	.47***	[.41, .52]	.14***
Vulnerability	.28***	[.24, .33]	.09***	.00	[−.05, .05]	.00
Deceitfulness
Grandiosity	.42***	[.38, .46]	.19***	.22***	[.17, .28]	.03***
Vulnerability	.43***	[.39, .47]	.21***	.29***	[.24, .34]	.06***
Hostility
Grandiosity	.29***	[.24, .33]	.09***	−.04	[−.09, .02]	.00
Vulnerability	.46***	[.42, .50]	.25***	.48***	[.43, .53]	.16***
Callousness
Grandiosity	.15***	[.10, .19]	.02***	−.06	[−.12, .00]	.00
Vulnerability	.28***	[.23, .32]	.09***	.31***	[.26, .37]	.07***
Attention Seeking
Grandiosity	.47***	[.43, .51]	.23***	.40***	[.34, .45]	.10***
Vulnerability	.35***	[.31, .39]	.14***	.11***	[.06, .16]	.01***
Grandiosity
Grandiosity	.33***	[.29, .38]	.12***	.32***	[.26, .38]	.07***
Vulnerability	.22***	[.17, .26]	.05***	.02	[−.03, .08]	.00
Irresponsibility
Grandiosity	.16***	[.11, .21]	.03***	−.13***	[.19, −.08]	.01***
Vulnerability	.36***	[.32, .40]	.15***	.44***	[.39, .49]	.14***
Impulsivity
Grandiosity	.21***	[.17, .26]	.05***	.09**	[.04, .15]	.01**
Vulnerability	.23***	[.19, .28]	.06***	.18***	[.12, .23]	.02***
Distractibility
Grandiosity	.24***	[.19, .29]	.06***	−.09**	[−.14, −.03]	.01**
Vulnerability	.44***	[.40, .48]	.22***	.49***	[.44, .54]	.17***
Perseveration
Grandiosity	.34***	[.30, .38]	.12***	−.05*	[−.10, −.00]	.00*
Vulnerability	.56***	[.52, .59]	.36***	.59***	[.54, .63]	.24***
Rigid Perfectionism
Grandiosity	.28***	[.24, .33]	.09***	.10**	[.04, .16]	.01**
Vulnerability	.34***	[.30, .38]	.13***	.28***	[.22, .33]	.05***
Risk Taking
Grandiosity	.17***	[.12, .21]	.03***	.27***	[.21, .33]	.05***
Vulnerability	.01	[−.03, .06]	.00	−.15***	[−.21, −.10]	.02***
Eccentricity
Grandiosity	.31***	[.26, .35]	.10***	.08**	[.02, .13]	.00**
Vulnerability	.40***	[.36, .44]	.18***	.35***	[.30, .40]	.09***
Perceptual Dysregulation
Grandiosity	.33***	[.29, .38]	.12***	.02	[−.03, .08]	.00
Vulnerability	.48***	[.44, .52]	.27***	.46***	[.42, .51]	.15***
Unusual Beliefs
Grandiosity	.31***	[.26, .35]	.10***	.16***	[.10, .22]	.02***
Vulnerability	.32**	[.28, .36]	.12***	.22***	[.17, .28]	.03***

Note. N = 1,653. PNI = Pathological Narcissism Inventory; PID-5 = The Personality Inventory for DSM-5; CI = confidence interval. Model 1 reflects the standardized betas, CIs for the unstandardized betas, and R² values when only the given predictor was entered into the model; Model 2 reflects the standardized betas, CIs for the unstandardized betas, and R² change values when both grandiosity and vulnerability were entered into the model at the same time. The R² change value is the value when the given predictor is added to the model.

p < .05. **p < .01. ***p < .001.

PNI-V from a univariate perspective yields significant positive associations with each of the 25 scales except for Risk Taking (β = .01, p = .620). Moving to a multivariate approach, a significant negative association arises with Risk Taking (β = −.15, p < .001) and the association with Manipulativeness and Grandiosity is no longer significant (β = .00, p = .957; β = .02, p = .413).

Again, examining the differences in associations from a univariate and multivariate approach, we see that PNI-G initially has a wide range of associations but takes on very specific associations in the multivariate approach. PNI-V, on the other hand, continues to have a diffuse pattern of association from both a univariate and multivariate perspective.

Profile Correlations

Because profiles of correlations are difficult to interpret by eye, as has been done in prior research, we examined the Pearson correlations between the profiles of expert ratings and observed βs for PNI-G and PNI-V predicting the FFM domains and for PID-5 scales. We calculated correlations for both the univariate βs as well as the multivariate βs, and we calculated correlations within domain (e.g., grandiosity βs correlated with expert ratings of grandiosity) as well as across domain (e.g., grandiosity βs correlated with expert rating of vulnerability). Results from the profile correlations can be found in Table 4.

Table 4.

Profile Correlations Between Expert Ratings and Observed βs for PNI-G and PNI-V.

	Matched		Crossed
	Grandiosity	Vulnerability	Grandiosity × Vulnerability	Vulnerability × Grandiosity
FFM
Univariate	.81	.98	.63	.14
Multivariate	.84	.92	−.48	−.28

Note. PNI = Pathological Narcissism Inventory; G = Grandiosity; V = Vulnerability; FFM = five-factor model. FFM expert ratings were based on those reported in Thomas et al. (2012). Grandiosity × Vulnerability is the observed βs for grandiosity × expert ratings of vulnerability. Vulnerability × Grandiosity is the observed βs for vulnerability × expert ratings of grandiosity.

Profile correlations from the univariate FFM models were based on expert ratings reported in Thomas et al. (2012) for correlations of grandiose and vulnerable narcissism across the FFM, and yielded within-domain correlations of r_{expertG.PNI-G} = .81 for grandiosity and r_{expertV.PNI-V} = .98 for vulnerability. In the across-domain correlations, the univariate grandiosity βs were correlated at r_{expertV.PNI-G} = .63 with expert ratings of vulnerability, whereas univariate vulnerability βs were correlated at r_{expertG.PNI-V} = .14 with expert ratings of grandiosity. The within-domain correlation for multivariate effects between grandiose βs and expert ratings of grandiosity was r_{expertG.PNI-G} = .84. The corresponding correlation for vulnerability was r_{expertV.PNI-V} = .92. The across-domain correlation between the βs for grandiosity and the expert ratings of vulnerability was r_{expertV.PNI-G} = −.48 and the correlation between the βs for vulnerability and expert ratings of grandiosity in the multivariate model was r_{expertG.PNI-V} = −.28. These results demonstrate that the multivariate approach yields a more distinct pattern of associations (i.e., profile) across domains, even when the univariate approach exhibits strong within-domain correlations with the expected pattern in this data set.

For the PID-5 expert ratings, we used the trait elevations for narcissistic PD (Samuel et al., 2012). In the univariate model, expert ratings of narcissistic PD were correlated at r_expert.PNI-G = .29 with the βs for PNI-G. The expert ratings of narcissistic PD were correlated at r_expert.PNI-V = −.42 with the βs of PNI-V. In the multivariate approach, PNI-G βs and expert ratings of narcissistic PD were correlated atr_expert.PNI-G = .59, while βs for PNI-V were correlated at r_expert.PNI-V = −.59. When adjusting for PNI-V, PNI-G exhibits strong positive correlation with the expert generated profile of narcissistic PD in the PID-5 traits.

Discussion

The present research aimed to clarify the unique nomological networks of the PNI-G and PNI-V scales. Previous research has relied on univariate models (i.e., zero-order correlations), finding that both scales yield a diffuse pattern of association to a broad range of maladaptive personality measures. This pattern of associations has raised questions about the validity of the PNI-G domain scale, as it departs from expert ratings for a measure of grandiose narcissism. For example, Miller et al. (2014) note that PNI-G tends to have associations with negative emotionality/fragility and does not adequately assess antagonism. We suggested that moving to a multivariate approach, which reveals the unique associations among variables, should reveal that the PNI-G scale does conform with patterns suggested by expert ratings once the shared maladaptive variance with PNI-V is statistically controlled.

Indeed, when evaluating the PNI-G and PNI-V from a multivariate approach, both scales aligned closely to contemporary conceptualizations of grandiosity and vulnerability, respectively. PNI-V continues to have associations with a wide range of personality scales. PNI-G, on the other hand, has more limited associations, and specifically aligns with scales thought to be marked by grandiose themes (e.g., narcissistic PD and histrionic PD). This demonstrates that the PNI-G scale does, in fact, capture the central elements of narcissistic grandiosity and should not be discounted as a scale capturing core features of narcissism.

We hypothesized that results from univariate regression analyses would look similar to previous studies examining patterns of correlations between the PNI-G and PNI-V scales and various personality measures (e.g., Miller et al., 2014; Thomas et al., 2012, 2016). Across the 10 PDs, FFM domains, and the DSM-5 Section III traits, univariate results were similar to the prior studies that led to criticisms of the PNI-G scale. Specifically, PNI-G had positive associations with 8 of the PDs, positive associations with Neuroticism, and positive associations with almost all of the 25 DSM-5 Section III traits. Admittedly, from this perspective, PNI-G looks much more like vulnerability than a construct marked by entitlement and a grandiose sense of self would be expected to.

However, we also hypothesized that moving to a multivariate approach would result in patterns of associations that clearly differentiated the PNI-G and PNI-V scales. Across both measures of personality pathology, PNI-V continues to have a wide range of positive associations, as well as a pattern of FFM domain associations consistent with general personality pathology (i.e., positive with Neuroticism, negative with Agreeableness and Conscientiousness). PNI-G, on the other hand, largely only associated significantly with those scales defined by grandiose themes. Specifically, across the 10 PD counts, significant positive associations emerged only with narcissistic PD, histrionic PD, and antisocial PD. In the FFM models, in evidence of suppression, PNI-G’s association with Neuroticism became negative, and its association with Extraversion strengthened, while a significant negative association with Agreeableness remained. Finally, with respect to the DSM-5 Section III traits, PNI-G had more circumscribed associations overall, with marked associations remaining for aspects of antagonism (Manipulativeness, Grandiosity, Attention Seeking) and other relevant traits (e.g., Risk Taking). With particular regard to Manipulativeness and Grandiosity, once accounting for the shared variance between PNI-G and V, PNI-G uniquely predicted these two features and PNI-V has a nonsignificant association. Contrary to Miller et al.’s (2014) findings, this demonstrates that PNI-G tends to uniquely predict antagonism-associated features. Also demonstrating suppression effects, negative associations emerged with some scales from the negative affectivity (e.g., Emotional Lability, Anxiousness, and Separation Insecurity) and detachment domains (e.g., anhedonia), which are antithetical to grandiose themes. Thus, once accounting for the shared variance in PNI-G and PNI-V, PNI-G no longer has positive associations with negative emotionality/fragility.

Thus, our results suggest that although the PNI-G and PNI-V scales share significant amounts of variance, both also contain unique elements that closely align current conceptualizations of grandiosity and vulnerability. By virtue of their shared focus on maladaptive (i.e., pathological) manifestations of narcissism, PNI-G and PNI-V both exhibit significant associations with measures of distress and dysfunction. When controlling for this shared variance, the unique aspects of each scale are revealed. Indeed, taking only a univariate approach (or zero-order correlational approach) does not account for what appear to be a large number of suppression effects in the PNI’s nomological network. Specifically, the emphasis on pathological manifestations of grandiosity and maladaptivity in the PNI leads to the previously observed associations with distress and dysfunction writ large, thereby suppressing associations with scales that tap in to conceptualizations of grandiosity uncomplicated by clinical distress. The implication of these findings is that future research seeking to understand prototypical manifestations of narcissistic grandiosity and vulnerability should use both scales together in multivariate models, in addition to the customary univariate analyses.

From a clinical perspective, these findings also have important implications. These results suggest that criteria for clinical cutoffs should perhaps be established for the PNI such that PNI-G scores are adjusted for total PNI elevation. Other clinical tools have followed a similar procedure. For example, the Minnesota Multiphasic Personality Inventory–2–Restructured Form (MMPI-2-RF; Ben-Porath & Tellegen, 2011) uses the Restructured Clinical Demoralization scale as a general measure of distress and negative affect. The items identified for Restructured Clinical Demoralization scale were used with factor analysis to differentiate those items in the remaining clinical scales that were distinct from the general level of distress. Thus, the remaining revised clinical scales essentially partial out the general factor of distress by retaining only those items that mark the unique variance in each factor. Another similar example is the MMPI-2 K scale and K corrected clinical scales. The K scale was designed to correct for ever-present biases in reporting, and used to generate corrected values of the clinical scales such that the corrected scale scores would provide estimates that would not be influenced by these biases (e.g., Greene, 1991/2011). Finally, the Inventory of Interpersonal Problems (Alden, Wiggins, & Pincus, 1990) distress factor is another relevant example. Specifically, elevation (or the average across all items) on the Inventory of Interpersonal Problems is thought to reflect general interpersonal distress, but calculation of the dominance and affiliation domains effectively remove this variance because they are weighted combinations of truly bipolar scales (Conroy, Elliot, & Pincus, 2009; Tracey, Rounds, & Gurtman, 1996). Similar to these examples, we believe the clinical use of the PNI would likely benefit from corrected scales such that the general level of narcissistic pathology is accounted for when interpreting the PNI-G. Nevertheless, establishing the specific methods and formal criteria for such a correction go beyond the scope of this article because they should ideally involve additional normative and clinical samples.

Results of the present study also have implications for how we think about narcissism. Despite the growing body of literature surrounding narcissism, many unanswered questions remain (Miller et al., 2017). Researchers typically agree that narcissism contains both grandiose and vulnerable components. How best to define these features and, consequently, how best to measure them has often been the subject of contentious debate. Some researchers (e.g., Miller et al., 2014) argue that narcissistic grandiosity and vulnerability should be largely unrelated to each other, and, thus, should uniquely predict different aspects of narcissism as well as other pathologies. We argue, however, that some degree of overlap is necessary if the two are to define a synthetic and unified construct such as pathological narcissism. If the two were completely unrelated, previous literature, suggesting that individuals can vacillate between the two (e.g., Gore & Widiger, 2016; Hyatt et al., 2017; Pincus, Cain, & Wright, 2014; Ronningstam, 2009; Wright, 2014) would be hard to conceptualize. These findings underscore the need to use grandiose and vulnerable scales in concert with each other.

This study sheds light on the types of analyses that should be considered to address the questions the field is asking and has important implications for how we use these measures in research and in clinical practice. We cannot accurately capture what the scales are intended to measure without accounting for the shared variance in each. As a result, we suggest that only when investigators move away from an overreliance on profiles of correlations and move toward regression-based approaches will we more fully understand the structure of complex constructs such as narcissism.

Limitations and Future Directions

There are several potential limitations of the present study. First, as previously mentioned, the psychiatric data set had a large amount of missing data associated with the PNI. Given that this was the primary measure of interest, we were forced to drop a large number of participants from the full clinical sample. Although there was a difference in age between those included and those not, we do not believe that our results would have been different had we been able to include the full data set (i.e., no differences were found in the results when using model-based missing data handling).

Furthermore, although a strength of our study was that we were able to examine PNI-G and PNI-V with a clinical sample and undergraduate sample, we did not have all three personality measures used in both data sets. Future research should examine each of the 10 DSM-5 Section II PDs, each of the FFM domain scores, and each of the 25 DSM-Section III PD traits, as well as other personality variables across a wide range of populations.

Although not a unique limitation to this study, but rather the field overall, narcissism’s definition remains relatively unclear and debated. As a result, there is a lack of convergence among measures which leads to the need for studies such as this one, evaluating the composition of different narcissism measures. Future studies should work to further understand the various mechanisms at play in the construct. Furthermore, given it appears different narcissism measures do not capture narcissistic features in the exact same way, researchers should select measures that align with the aims of their study (e.g., pathological narcissism vs. adaptive, vulnerable vs. grandiose narcissism).

Conclusion

Previous literature (e.g., Miller et al., 2014; Miller et al., 2016; Thomas et al., 2012, 2016) has found that the PNI-G scale does not align with contemporary conceptualization of narcissistic grandiosity. Both in previous studies and the present study, it is clear that PNI-G does contain a large amount of maladaptivity and pathology that is typically conceptualized as part of narcissistic vulnerability. We do not argue otherwise. However, our results suggest that PNI-G also contains the core grandiose themes. Examining zero-order correlations of the PNI-G and PNI-V yield a wide range of associations to different personality variables. Taking a multivariate approach results in more well-differentiated associations, at least with regard to PNI-G. We argue that to understand the depth of any measure’s associations, it is important to examine the full extent of associations, which can only be done from taking a multivariate approach in addition to a univariate approach. These results have important implications for how to conduct future research with the PNI and other multiscale measures of narcissism, and how a measure’s clinical validity and utility is evaluated.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by grants from the National Institute of Mental Health (L30 MH101760, Wright; R01 MH080086, Simms). The opinions expressed are solely those of the authors and not those of the funding source.

Notes

References

Alden

L. E.

Wiggins

J. S.

Pincus

A. L.

(1990). Construction of circumplex scales for the Inventory of Interpersonal Problems. Journal of Personality Assessment, 55, 521-536.

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). Washington, DC: Author.

Back

M. D.

Kufner

A. C. P.

Dufner

Gerlach

T. M.

Rauthmann

J. F.

Denissen

J. J. A.

(2013). Narcissistic admiration and rivalry: Disentangling the bright and dark sides of narcissism. Journal of Personality and Social Psychology, 105, 1013-1037. doi:10.1037/a0034431

Ben-Porath

Y. S.

Tellegen

(2011). MMPI-2-RF (Minnesota Multiphasic Personality Inventory-2-Restructured Form): Manual for administration, scoring, and interpretation. Minneapolis: University of Minnesota Press.

Cain

N. M.

Pincus

A. L.

Ansell

E. B.

(2008). Narcissism at the crossroads: Phenotypic description of pathological narcissism across clinical theory, social/personality psychology, and psychiatric diagnosis. Clinical Psychology Review, 28, 638-656.

Conroy

D. E.

Elliot

A. J.

Pincus

A. L.

(2009). The expression of achievement motives in interpersonal problems. Journal of Personality, 77, 495-526.

Costa

P. T.

McCrae

R. R.

(1992). Revised NEO Personality Inventory (NEO-PI-R) and the NEO Five-Factor Inventory (NEO-FFI) professional manual. Odessa, FL: Psychological Assessment Resources.

First

M. B.

Spitzer

R. L.

Gibbon

Williams

J. B. W.

(2002). Structured Clinical Interview for DSM-IV-TR Axis I Disorders, Research Version, Patient Edition (SCID-I/P). New York: Biometrics Research, New York State Psychiatric Institute.

Glover

Miller

J. D.

Lynam

D. R.

Crego

Widiger

T. A.

(2012). The Five-Factor Narcissism Inventory: A five-factor measure of narcissistic personality traits. Journal of Personality Assessment, 94, 500-512.

10.

Gore

W. L.

Widiger

T. A.

(2016). Fluctuation between grandiose and vulnerable narcissism. Personality Disorders: Theory, Research, and Treatment, 7, 363-371.

11.

Greene

R. L.

(2011). The MMPI-2/MMPI-2-RF: An interpretive manual. Boston, MA: Pearson Education. (Original work published 1991)

12.

Hyatt

C. S.

Sleep

C. E.

Lynam

D. R.

Widiger

T. A.

Campbell

W. K.

Miller

J. D.

(2017). Ratings of affective and interpersonal tendencies differ for grandiose and vulnerable narcissism: A replication and extension of Gore and Widiger (2016). Journal of Personality. Advance online publication. doi:10.1111/jopy.12325

13.

Krueger

R. F.

Derringer

Markon

K. E.

Watson

Skodol

A. E.

(2012). Initial construction of a maladaptive personality trait model and inventory for DSM-5. Psychological Medicine, 42, 1879-1890.

14.

McCrae

R. R.

Costa

P. T.

Jr. (2007). Brief versions of the NEO-PI-3. Journal of Individual Differences, 28, 116-128.

15.

McCrae

R. R.

Costa

P. T.

Jr. Martin

T. A.

(2005). The NEO-PI-3: A more readable revised NEO Personality Inventory. Journal of Personality Assessment, 84, 261-270.

16.

Miller

J. D.

Lynam

D. R.

Campbell

W. K.

(2016). Measures of narcissism and their relations to DSM-5 pathological traits: A critical reappraisal. Psychological Assessment, 23, 3-9.

17.

Miller

J. D.

Lynam

D. R.

Hyatt

C. S.

Campbell

W. K.

(2017). Controversies in narcissism. Annual Review of Clinical Psychology, 13, 1-54.

18.

Miller

J. D.

McCain

Lynam

D. R.

Few

L. R.

Gentile

MacKillop

Campbell

W. K.

(2014). A comparison of criterion validity of popular measures of narcissism and Narcissistic Personality Disorder via the use of expert ratings. Psychological Assessment, 26, 958-969.

19.

Morey

L. C.

(1991). Personality Assessment Inventory professional manual. Odessa, FL: Psychological Assessment Resources.

20.

Morf

C. C.

(2006). Personality reflected in a coherent idiosyncratic interplay of intra- and interpersonal self-regulatory processes. Journal of Personality, 76, 1527-1556.

21.

Pincus

A. L.

Ansell

E. B.

Pimentel

C. A.

Cain

N. M.

Wright

A. G. B.

Levy

K. N.

(2009). Initial construction and validation of the Pathological Narcissism Inventory. Psychological Assessment, 21, 365-379.

22.

Pincus

A. L.

Cain

N. M.

Wright

A. G. C.

(2014). Narcissistic grandiosity and narcissistic vulnerability in psychotherapy. Personality Disorders: Theory, Research, and Treatment, 5, 439-443.

23.

Ronningstam

(2009). Narcissistic personality disorder: Facing DSM-V. Psychiatric Annals, 39, 111-121.

24.

Samuel

D. B.

Lynam

D. R.

Widiger

T. A.

Ball

S. A.

(2012). An expert consensus approach to relating the proposed DSM-5 types and traits. Personality Disorders: Theory, Research, and Treatment, 3, 1-16.

25.

Thomas

K. M.

Wright

A. G. C.

Lukowitsky

M. R.

Donnellan

M. B.

Hopwood

C. J.

(2012). Evidence for the criterion validity and clinical utility of the Pathological Narcissism Inventory. Psychological Assessment, 19, 135-145.

26.

Thomas

K. M.

Wright

A. G. C.

Lukowitsky

M. R.

Donnellan

M. B.

Hopwood

C. J.

(2016). Correction to “evidence for the criterion validity and clinical utility of the Pathological Narcissism Inventory.” Psychological Assessment, 23, 262-263.

27.

Tracey

T. J. G.

Rounds

Gurtman

(1996). Examination of the general factor with the interpersonal circumplex structure: Application to the Inventory of Interpersonal Problems. Multivariate Behavioral Research, 31, 441-466.

28.

Wright

A. G. C.

(2014). Integrating trait and process based conceptualizations of pathological narcissism in the DSM-5 era. In Besser

(Ed.), Handbook of psychology of narcissism: Diverse perspectives (pp. 153-174). Hauppauge, NY: Nova Science.

29.

Wright

A. G. C.

Edershile

E. A.

(2018). Issues resolved and unresolved in pathological narcissism. Current Opinion in Psychology, 21, 74-79.

30.

Wright

A. G. C.

Lukowitsky

M. R.

Pincus

A. L.

Conroy

D. E.

(2010). The higher order factor structure and gender invariance of the Pathological Narcissism Inventory. Psychological Assessment, 17, 467-483.

31.

Wright

A. G. C.

Pincus

A. L.

Thomas

K. M.

Hopwood

C. J.

Markon

K. E.

Krueger

R. F.

(2013). Conceptions of Narcissism and the DSM-5 Pathological Personality traits. Psychological Assessment, 20, 339-352.

32.

Wright

A. G. C.

Simms

L. J.

(2014). On the structure of Personality Disorder traits: Conjoint analyses of the CAT-PD, PID-5, and NEO-PI-3 trait models. Personality Disorders: Theory, Research, and Treatment, 5, 43-54.

33.

Wright

A. G. C.

Simms

L. J.

(2015). A metastructural model of mental disorders and pathological personality traits. Psychological Medicine, 45, 2309-2319.