The concept of narcissism originated from Greek mythology (Hamilton, 1942) has always been of interest in psychology and psychiatry. Most researchers seem to agree with its dimensional aspect (e.g., Foster and Campbell, 2007), ranging from a normal adaptive mechanism in healthy individuals to pathological narcissism that causes distress and impairment (Ackerman et al., 2011). According to the Diagnostic and Statistical Manual of Mental Disorders [5th edition, (DSM-V); American Psychiatric Association, 2013], individuals with a Narcissistic Personality Disorder (NPD) have significant impairments in (1) personality functioning, either in identity (excessive reference to others for self-definition and self-esteem regulation) or in self-direction (goal-setting based on gaining approval from others; personal standards unreasonably high in order to see oneself as exceptional, or too low based on a sense of entitlement) and in (2) interpersonal functioning, either empathy or intimacy (relationships largely superficial and existing to serve self-esteem regulation). Individuals with NPD have also pathological personality traits such as: (1) grandiosity (feelings of entitlement; self-centeredness; firmly holding to the belief that one is better than others; condescending toward others) and (2) attention seeking (excessive attempts to attract and be the focus of the attention of others; admiration seeking). These days, two expressions of narcissism are considered: grandiosity and vulnerability (Cain et al., 2008). While grandiosity is associated with the tendency to exploit others and with the feeling of entitlement and superiority, vulnerability is related to the feeling of inadequacy and incompetence with negative affect (Maxwell et al., 2011; Miller et al., 2013).

Ideally, narcissism should be studied in its multidimensional perspective (see for example, Pincus and Lukowitsky, 2010). However, the majority of researchers tend to use the Narcissistic Personality Inventory (NPI; Raskin and Hall, 1979, 1981) to only evaluate its maladaptive grandiosity dimension. For example, Cain et al. (2008; cited by Brin, 2011) report that the NPI was used in 77% of the empirical studies dealing with the concept of narcissism. This scale is based on the description of the NPD included in the DSM-III (American Psychiatric Association, 1980), and consists of 40 forced-choice dichotomous items (between narcissistic and nonnarcissistic statements). In this scale, participants are asked to select one statement within every dichotomous item that best describes their personality or opinion. Most researchers choose this dichotomous forced-choice format whilst some others (e.g., Kubarych et al., 2004; Barelds and Dijkstra, 2010) choose to convert the forced-choice response format into a Likert one. According to Barelds and Dijkstra (2010), the correlation between the dichotomous forced-choice format and a 5-point Likert version of the scale (using the dichotomous statements as anchor points) is excellent (r = 0.97; p < 0.01). In Boldero et al. (2015) the use of a 6-point rating scale showed a similar correlation (r = 0.96; p < 0.01).

Some studies aimed to explore the multidimensional structure of the NPI using exploratory factor analysis (EFA), principal component analysis (PCA) and/or confirmatory factor analysis (CFA). Raskin and Terry's model suggests seven subscales including: Authority, Self-Sufficiency, Superiority, Vanity, Exhibitionism, Entitlement, and Exploitativeness (Raskin and Terry, 1988). Other researchers proposed alternative models, such as the 4-factor solution of Emmons (1984, 1987) that includes: Leadership/Authority, Superiority/Arrogance, Self-Absorption/Self-Admiration, and Exploitativeness/Entitlement. Kubarych et al. (2004) proposed a 2-factor or a 3-factor model with (1) Power and (2) Exhibitionism as the two factors of the first one, and (1) Power, (2) Exhibitionism, and (3) Being a Special Person as the three factors of the second one. Inconsistent results in factor structure are generally considered to be associated with methodological differences between studies (see in Boldero et al., 2015). However, even the most recent studies that have taken the current psychometric recommendations into account (performing tetrachoric correlation matrix because of the dichotomous format, using the scree plot and interpreting the items' meanings to obtain a factor solution, …), failed to replicate the 7-factor structure (Corry et al., 2008; Ackerman et al., 2011). For example, Corry et al. (2008) tested several models with confirmatory factor analysis (CFA) and concluded that a 2-factor solution (with Leadership/Authority and Exhibitionism/Entitlement as factors) appeared to be the most parsimonious model, in terms of fit indices and internal consistencies values. For Ackerman et al. (2011), these two main dimensions are useful in distinguishing the part of the adaptive content from the maladaptive content of the narcissism construct. They observed a Leadership/Authority factor similar to Corry et al.'s (2008) but suggested to split the Exhibitionism/Entitlement dimension into two parts: Grandiose Exhibitionism and Entitlement/Expoitativeness. However, Boldero et al. (2015) point toward the fact that, despite psychometrical precautions, some methodological differences remain between Corry et al.'s (2008) and Ackerman et al.'s (2011) studies regarding the criteria for the EFA (type of rotation, choice of estimator and utilization of the internal consistency indices). Moreover, Boldero et al. (2015) emphasize the superiority of the parallel analysis (Horn, 1965) over the traditional criteria (scree plot, eigenvalues) in determining the number of factors to retain, and support the choice of an oblique rotation in this context (because of probable correlated items). The results of their study showed a 2-factor structure with the binary or with the rating response's format (including a general factor and six specific factors). The authors, therefore, assumed that the data were unidimensional (Reise et al., 2010) and considered the NPI's general factor as a measure of a narcissism latent trait. They also pointed out that the rating item general factor assessed more narcissism components than the binary item.

Several NPI adaptations in various languages already exist. For example, Barelds and Dijkstra (2010) proposed a Dutch version that supports neither the 4-factor model of Emmons (1987) nor the 7-factor model of Raskin and Terry (1988). Performing an EFA, they suggested a single-factor solution or a 2-factor model explaining respectively 27.3 and 33.7% of the variance. Other adaptations in Swedish (Kansi, 2003), German (Schutz et al., 2004), and Greek (Coccosis et al., 1998) also failed to replicate the 4-factor model of Emmons (1987) or the 7-factor model of Raskin and Terry (1988).

A French adaptation of the NPI (with a 7-point Likert response format) was proposed by Brin (2011) in order to obtain the grade of Doctor in Psychology (University Laval, Québec). To our knowledge, this study has not yet been published in a peer-reviewed journal. In Brin's study, a CFA testing the 7-factor model of Raskin and Terry (1988) showed no reasonable fit indices. Multiple exploratory factors analyses (EFA) were performed with promax rotation and suggested two interesting models with two or three factors, with similar findings to Corry et al.'s (2008) regarding the first one (2-factor model called Exhibitionism/Looking for attention and Leadership/Arrogance). This study confirmed (see Raskin and Terry, 1988 for other examples) an excellent internal consistency (α = 0.91) and a gender difference where males obtained higher scores than females (as in Corry et al., 2008).

The aim of our study was to propose a French adaptation of the NPI using a sample of Belgian French-speaking students. Because our data were collected before our knowledge of the Brin's study, our own French version was used (Brin, 2011). A comparison with Brin's version showed minor differences that do not affect the meaning of the sentences. As mentioned earlier, there are large inconsistencies between studies in terms of factor structure. We therefore chose PCA as the most adequate method to explore the factor structure of this new adaptation of the scale. Regarding methodological recommendations, we expected to reproduce the results of Boldero et al. (2015) and confirm the use of a unique total score as a measure of the narcissism latent trait. We also expected to obtain similar results as in Brin (2011) in terms of internal consistency and gender differences. Because a number of researchers have highlighted and tested the possible contribution of social desirability in the responses of the NPI, we decided to explore this too and to confirm minor correlations (e.g., Auerbach, 1984; Watson et al., 1984).

In this study, we proposed a French adaptation of the NPI in a French-speaking Belgian subjects sample with the aim to explore its psychometrical aspects. Given the inconsistencies of previous studies (e.g., Corry et al., 2008; Ackerman et al., 2011) and considering methodological recommendations (see on Boldero et al., 2015 for a review), we performed a PCA on an unrotated tetrachoric correlation matrix and used parallel analysis to interpret our results. Just like Boldero et al. (2015), we observed that the first of the 7 factors we obtained explains the largest part of the variance. We therefore agree with them and other researchers such as Barelds and Dijkstra (2010) who consider the total score of the NPI as a good measure of the narcissistic trait at least in its grandiosity aspect.

Some psychometrical methods support the unidimensional hypothesis. Although Reckase (1979) proposes to retain a rate of variance explained by the first factor higher than 20% (which is the case in our study), Carmines and Zeller (1979) suggest four criteria to apply on the unrotated matrix: (1) the first component should explain a large part of the variance (>40%), (2) the next components should explain fairly equal components of the remaining variance, (3) all or most of the items should have substantial loadings on the first factor (>0.30), (4) all or most of the items should have their highest loadings on the first factor. Three of the four criteria were met in our study: (1) the part of the variance explained by the first factor is 26.59%, which corresponds to nearly half (46%) of the total variance explained (57.87%) (2nd criteria), (2) 35 items load on the first factor above 0.30 (3rd criteria), and (3) 28 items that met the 4th criteria (see on Table 2). The 1st criteria—the strictest—is not met. Finally, Lord (1980) envisages a procedure which implies creating an index (1st eigenvalue—2nd eigenvalue/2nd eigenvalue—3rd eigenvalue). In our study, this index was (10.63–3.11)/(3.11–2.43) = 11.05. Unfortunately, the drawback of this method is the absence of consensus for determining a cut-off score to identify unidimensionality.

Because our data collection took place some years ago, we did not convert the forced-choice response format into a Likert format scale, as suggested by Boldero et al. (2015). This could be considered a limitation in our study. A comparison of the two response formats in Boldero et al. (2015) suggested that a rating format enhances the information obtained with the NPI. Moreover, Boldero et al. (2015) express concerns regarding the influence of social desirability in the dichotomous format. They hypothesized that several narcissistic vs. nonnarcissistic statements are not equivalent in terms of social desirability and that this could underestimate the ranking of the subjects on the narcissism latent trait. In our study, we found statistically significant correlations with social desirability but these were inferior to 0.30. Interestingly, the correlation is significant for females but not for males. This might indicate a potential instability of the measure across gender, at least with the binary format. Moreover, we believe that it could be difficult for participants to choose a “right” answer in a forced-choice situation. Sometimes participants find themselves having to choose between two unsatisfactory answers. This is especially problematic (in terms of methodology) when the range between the items is variable. While some pairs are clearly antagonistic (“I have a natural talent for influencing people” vs. “I am not good at influencing people”), we doubt that it is the case for other pairs (“I find it easy to manipulate people” vs. “I don't like it when I find myself manipulating people”).

Regarding gender differences, we found a higher NPI total score for males than for females, thus replicating other studies' findings (Corry et al., 2008; Brin, 2011). However, we did not test metric gender invariance; results should therefore be interpreted with caution. Indeed, testing for gender invariance could ensure that the measure is not influenced by systematic response's bias linked to gender (Meredith, 1993). Significant correlations we observed between the NPI's total score and social desirability in females but not in males might be another argument in favor of testing gender invariance in a future study. If this invariance condition was accepted, the difference on the NPI's total score we found between males and females could be interpreted in terms of stereotypes in gender roles. Considering the environmental aspects of this hypothesis (education, culture …), it would be interesting to develop a version of the NPI specifically aimed at children and test a possible interaction between age and gender.

In conclusion, although Narcissism Personality Disorder (NPD) is considered a multidimensional construct in the DSM-V, results from our research study support the idea that the NPI cannot explore its entire complexity. The lack of consensus in literature regarding the limits between normal and pathological narcissism renders its evaluation particularly complicated (Pincus and Lukowitsky, 2010) and future studies are necessary. Vulnerability aspects of the NPD should be explored further and integrated in new measures such as in the Pathological Narcissism Inventory (PNI; Pincus et al., 2009). Notwithstanding, the general limitations of self-report scales, the use of the NPI and the interpretation of its total score may be useful for clinical purposes. Our French adaptation showed good psychometrical properties, congruent with the literature data which allows us to consider it a satisfactory instrument to measure narcissism in its grandiosity dimension. For future studies, we suggest using a rating scale rather than the dichotomous format and to test the unidimensional hypothesis with CFA. A gender invariance measure should also be performed to confirm the stability of the scale across gender and to clarify the social desirability influence.

The study received the approval from the Ethical Committee of the Erasmus Hospital. The study took place during a scheduled class time.

PL initiated and designed the study and provided the funding, SB and CK collected the data, SB and GL co-wrote the paper.

The present study was supported by the Belgian National Funds for Scientific Research (grants 1.5.123.04, 1.5.175.06, and 3.4.553.01.F).