BACKGROUND

Although personality disorders (PD) have been considered as categories, since the first version of the Diagnostic and Statistical Manual of Mental Disorders (DSM; APA, 1952), evidence exists only for schizotypal personality disorder (Conway et al., 2019; Haslam et al., 2020; Hopwood et al., 2018). According to Krueger et al. (2018), categorical approaches tend to weight disciplinary background and tradition. Conversely, there is accumulating evidence that mental disorders are continuously distributed in the population, without the discontinuity expected of categorical diseases (Krueger et al., 2018). The existing body of research on comorbidity and the variations of symptoms has shown that psychopathology is dimensional rather than categorical (DeYoung et al., 2020). The categorical classification systems, in which each disorder is a discrete entity, do not reflect reality (Haslam et al., 2020). Only 14% of psychopathological findings would be related to categories (Haslam et al., 2012). Among the limitations of categorical systems, it is worth mentioning the excessive comorbidity between personality disorders, the arbitrary distinction between normality and pathology, the lack of clinical utility, the overlapping of symptoms (e.g., symptoms belonging to several diagnoses), and the limited scope of the existing categories (for which the most frequent diagnosis is “not specified”) (First et al., 2002; Kessler et al., 2005; Simonsen et al., 2008).

From the beginning of the development of DSM, it was already recognized that the categorical system might never be able to account for the underlying etiology of mental disorders; therefore, a paradigm shift would be necessary (Kupfer et al., 2002). Although the DSM-5 (APA, 2013) maintained the categorical approach, it presents an Alternative Dimensional Model for Personality Disorders (AMPD) in Section III, “Emerging Measures and Models”. AMPD includes an assessment of impairments in the self and interpersonal functioning (criterion A) and the presence of pathological personality traits (criterion B). Pathological personality traits (criterion B) are organized into five higher-order domains, i.e. negative affectivity, detachment, antagonism, disinhibition, and psychoticism, and 25 specific facets nested within these trait domains. Although the DSM-5 proposal (APA, 2013) has also received harsh criticism from the scientific community (Livesley, 2012; Sánchez, 2019; Shedler et al., 2010; Verheul, 2012; Widiger, 2013), it has established a structure of five domains to understand the pathology of personality.

DIMENSIONAL AND HIERARCHICAL APPROACHES TO PSYCHOPATHOLOGY

Currently, new research lines seek a more appropriate classificatory system for psychopathology (Caspi et al., 2014; Kotov et al., 2020; Krueger & Markon, 2014; Lahey et al., 2012, 2017). In pursuit of this goal, two alternative approaches, hierarchical and dimensional, have favorable evidence: multi-level models, such as the Hierarchical Taxonomy of Psychopathology (HiTOP) (Conway et al., 2019; Kotov et al., 2017; Krueger et al., 2018) and bifactor models (Caspi et al., 2014; Lahey et al., 2012, 2017; Montes & Sánchez, 2019). These proposals seek to provide a more parsimonious structure of psychopathology than the system of numerous categorical disorders, although they differ in form. On one hand, the HiTOP asserts that most psychopathology can be organized hierarchically, with broad spectrums at the top of the hierarchy (e.g., internalizing, disinhibited externalizing, antagonistic externalizing, thought disorder) and a higher-order dimension above these spectrums (a general psychopathology factor) (Kotov et al., 2017, 2020). On the other hand, the bifactor models distinguish between a general factor of severity and specific factors reflecting style (their content), understanding that each element provides unique diagnostic information (Ringwald et al., 2019). While the style would represent the how of the pathology – the way the dysfunction is expressed – the general factor would inform the how much of the pathology – the prognosis of dysfunctional functioning. Both components would be independent, and the bifactor models would be the statistical way of accounting for their existence (e.g., Shields et al., 2021). Both models, bifactor (Caspi et al., 2014) and the HiTOP (Kotov et al., 2020), consider the existence of a general dimension of psychopathology called the “general psychopathology factor” (or “p factor”) (Caspi et al., 2014).

GENERAL PSYCHOPATHOLOGY FACTOR

Lahey et al. (2012) presented the first evidence of a general psychopathology factor, and later it was replicated by Caspi et al. (2014). This common risk factor (“p factor”) summarizes individuals’ propensity to develop any forms of common psychopathologies throughout the life course. In other words, it could measure the risk of a greater propensity for mental disorders in terms of disease recurrence, chronicity, comorbidity, and severity (Caspi et al., 2014). According to Caspi and Moffitt (2018), a higher p would be typical of the most severe disorders and could even lead to psychotic thought processes.

The issue of psychopathology severity is crucial for understanding PDs. Tyrer (2005) considers severity as a predictor of persistent social dysfunction, evolution and chronicity, response to treatment, and quality of life in general. Millon (1969) early warned about a severity dimension for PDs, even within the categorical canons, through the concept of syndromic continuity. According to Millon, the personality is oriented on a continuum, with the normal personality at one extreme and the severe pathology at the other. No line sharply separates the normal from the pathological, since the pathological has the same determinants as normal functioning. At the top of this pathological structure is the decompensated personality, characterized by decreased awareness of reality and cognitive and emotional lack of control. In more current terms, syndromic continuity is equivalent to the p factor, in the sense of a dimension of severity. However, the severity of PDs has been absent in the different versions of the DSM.

THE PRESENT STUDY

The idea of a general factor of personality psychopathology, specifically in the DSM-5 AMPD (APA, 2013), has been little explored. As noted, the available evidence around personality disorders, dimensional approaches, general psychopathology factor, and bifactor models is scattered and fragmented.

The Personality Inventory for DSM-5 (PID-5; Krueger et al., 2012) was designed to assess maladaptive personality traits specified within the DSM-5 AMPD criterion B (APA, 2013). PID-5 has mostly satisfactory psychometric qualities and has subsequently been adapted for use in different cultures and languages (for a review, see Al-Dajani et al., 2016; Sanchez et al., 2020, 2023). However, it also has shown considerable variability across studies, and several facets’ scales show evidence of interstitiality – the cross-loading of facets onto more than one domain (Waters & Bagby, 2018). One possibility is the presence of a general underlying factor that explains the cross-loading of several facets. Nevertheless, few studies have examined the PID-5 using a bifactor modeling approach. Montes and Sánchez (2019) and Gomez et al. (2020) worked with briefs forms of the PID-5 – which assess only the five higher-order personality domains – and found a better fit of the bifactor model against competing models – a robust general factor involving a general factor alongside the five specific factors. Gutierrez et al. (2021) also examined the relationships between the ICD-11 and DSM-5 systems for PD and found that the extraction of a general factor of PD fixes the problems of blurred boundaries between domains.

This work aimed to study the underlying structure of pathological personality traits of the AMPD model (APA, 2013). For this purpose, the internal structure of a version of the PID-5 extensive form (Sanchez et al., 2023) was examined by a confirmatory factor analysis. This version assesses both domains and facets through fewer items (108). Following Caspi et al. (2014), four alternative models were compared: Model 1 (M1), five-factor oblique; Model 2 (M2), higher-order factor (five first-order factors and one second-order factor) (p); Model 3 (M3), bifactor (five specific factors and a general factor – p) and Model 4 (M4), one-factor (p) (unidimensional).

Tested models. M1 is a five-factor model based on PID-5 theoretical structure and empirical evidence (Al-Dajani et al., 2016; Krueger et al., 2012; Sanchez et al., 2023). In this model, each PID-5 facet was allowed to load only on one factor (e.g., anxiousness in negative affectivity), and factors were allowed to be correlated. M2 is a second-order model, where the 25 facets load on their respective factors, and these factors load on a single higher-order factor, labeled p factor, or a general factor of personality psychopathology. M3 is a bifactor model, with all 25 facets loading on the general factor (GF) and on their own pathological personality specific factors (SF, domains, e.g., negative affectivity). GF and SF were modeled orthogonally. Finally, M4 is a one-factor model, where all 25 facets load on a single general factor (p), that arises as a latent factor from the observed variables.

PARTICIPANTS AND PROCEDURE

PARTICIPANTS

We reanalyzed data from the PID-5 Spanish validation study (Sanchez et al., 2023). The sample consisted of 525 subjects drawn from the general population of Argentina (over 18 years). Participant’s age ranged from 18 to 85 (M = 40.10, SD = 14.20). Women accounted for 70.3% of the sample. Most participants had at least completed high school (98.8%).

MEASURES

Personality traits were assessed with the Argentine version of the Personality Inventory for DSM-5 (Krueger et al., 2012; Sanchez et al., 2023). This version assesses the five higher-order pathological personality domains (negative affectivity, detachment, antagonism, disinhibition, and psychoticism) and the 25 lower-order pathological personality facets through a reduced number of items (108). The study of Sanchez et al. (2023) showed that the 25 facets can be evaluated with half of the test items without losing relevant information. An expert judge’s review and a pilot test suggested that the original 4-point format is unusual in our context and difficult to translate. Argentine respondents are more familiar with a 5-point Likert scale ranging from 1 (does not describe me at all) to 5 (describes me completely) and find it easier to use. The Sanchez et al. (2023) version had psychometric properties comparable to the original one, regarding dimensional structure (five factors), internal consistency, construct validity, and relationships with the FFM personality traits.

PROCEDURE

Ethical approval was obtained from the Interdisciplinary Program in Bioethics of the National University of Mar del Plata, Argentina. The research conformed to the provisions of the Declaration of Helsinki. The participants were contacted by email and social networks, and answered the questionnaire online. We used a convenience sample following snowball sampling. Subjects had written informed consent for this study, when they were informed about the purposes and that their responses would be kept anonymous. Participation was voluntary, and no compensation was offered.

DATA ANALYSIS

A series of confirmatory factor analyses models were tested. We used the robust diagonally weighted least squares estimator (DWLS), a recommended estimation method for non-normal and ordinal data (Li, 2016; Mîndrilă, 2010). The following fit indices were computed: a) for absolute fit: chi-square goodness-of-fit statistic (non-significant chi-square tests indicate good model fit); Satorra-Bentler scaled chi-square (S-B χ2; Satorra & Bentler, 1990); root-mean-square error of approximation (RMSEA; cut-off value < .08, better is < .05); standardized root mean square residual (SRMR; cut-off value < .08); b) for comparative fit: comparative fit index (CFI); and c) for parsimonious fit: Akaike information criterion (AIC) and consistent Akaike information criterion (CAIC). Both AIC and CAIC can be used to compare competing models, and lower values indicate a better trade-off between fit and complexity (Van de Schoot et al., 2012).

Given that the bifactor model tends to show a superior fit (see, for example, Forbes et al., 2021; Greene et al., 2019; Reise et al., 2016), it is necessary to include complementary statistical indices to reach more solid conclusions regarding the general factor. For this reason, additional and specific bifactor model indices were calculated based on recommendations by Rodriguez et al. (2016b): coefficient omega (ω), omega hierarchical (ωH; Zinbarg et al., 2006), omega hierarchical subscale (ωHs; Reise, 2012), Index H – construct reliability (Hancock, 2001), explained common variance (ECV; Sijtsma, 2009; Ten-Berge & Socăn, 2004), ECV-I (Stucky et al., 2013), and percentage of uncontaminated correlations (PUC; Bonifay et al., 2015; Reise et al., 2013b). These indices, despite being recommended, are little used in this area.

The ω coefficient (McDonald, 1999) is a model-based estimate of internal reliability. When data are suitably represented by a bifactor structure, coefficient omega hierarchical (omegaH or ωH) is a useful model-based reliability index, which estimates the proportion of variance in total scores that can be attributed to a single general factor (Rodriguez et al., 2016b). ωHs is an index reflecting the reliability of a subscale score (specific factor, SF) after controlling for the variance due to the general factor (Reise et al., 2013a). Regarding ωH, magnitudes ≥ .70 are expected to conclude, at least partially, in favor of unidimensionality, and in the case of ωHs values ≥ .30 could be considered substantial (Smits et al., 2015). Smits et al. (2015) consider this unique variance to be substantial for a value ωHs ≥ 30, moderate for a value > 20, < 30; and low for a value ωHs ≤ 20.

Explained common variance (ECV) is the proportion of all common variance explained by the general factor. This is a degree of unidimensionality index and is directly related to the relative strength of the general factor (Rodriguez et al., 2016b). The appropriate ECV benchmarks are determined by the PUC (Reise, 2012; Reise et al., 2013b). The percent of uncontaminated correlations (PUC) represents the percentage of covariance terms that only reflect variance from the general dimension. Rodriguez et al. (2016a) state that “when ECV is > .70 and PUC > .70 relative bias will be slight and the common variance can be regarded as essentially unidimensional” (p. 232). Reise et al. (2013b) claim that “when PUC values are lower than .80, general ECV values greater than .60 and omegaH > .70 (of the general factor) suggest that the presence of some multidimensionality is not severe enough to disqualify the interpretation of the instrument as primarily unidimensional” (p. 22). Finally, I-ECV can be computed at the item level to identify the percent of item common variance attributable to a general factor (Stucky et al., 2013).

Analyses were carried out on facet total scores following the original study (Krueger et al., 2012) and subsequent validation studies (e.g., Bagby et al., 2022; Coelho et al., 2022; Gutiérrez et al., 2017; Koster et al., 2020; Labancz et al., 2022; Maples et al., 2015; Wright et al., 2012). All CFA analyses were performed using LISREL 8.80 for Windows (Jöreskog & Sörbom, 2006). Bifactor coefficients were computed using the Bifactor Indices Calculator (Dueber, 2017).

RESULTS

DESCRIPTIVE STATISTICS

In a first step, means, standard deviations, skewness, and kurtosis were calculated (see Table 1). Kolmogorov-Smirnov test yielded a p-value of p < .001 for all the variables, which allows the null hypothesis of univariate normal distribution to be rejected. The multivariate normality hypothesis was also rejected since the Mardia (1970) coefficient showed significant kurtosis (p < .001). Cronbach’s α values for the scales negative affectivity, detachment, antagonism, disinhibition, and psychoticism in the current study were .92, .88, .88, .87, and .86 respectively.

Table 1

Descriptive statistics of the PID-5 facets and domains

PID-5MSDSkewKurtK-S1p
Facets
  1. Withdrawal2.121.010.67–0.43.15.001
  2. Anhedonia2.050.900.880.08.14.001
  3. Intimacy avoidance1.860.901.120.68.17.001
  4. Restricted affectivity2.590.930.34–0.50.08.001
  5. Eccentricity2.271.020.66–0.46.11.001
  6. Perceptual dysregulation1.680.811.361.38.20.001
  7. Unusual beliefs and experiences1.760.881.381.44.19.001
  8. Grandiosity1.870.791.000.75.13.001
  9. ANention seeking2.311.030.53–0.54.11.001
  10. Manipulativeness2.160.830.820.39.14.001
  11. Deceitfulness1.740.811.281.54.18.001
  12. Callousness1.440.692.225.53.27.001
  13. Risk taking1.790.731.081.02.14.001
  14. Impulsivity2.060.930.79–0.11.14.001
  15. Irresponsibility1.810.731.010.98.14.001
  16. Distractibility2.621.090.34–0.79.08.001
  17. Rigid perfectionism2.261.340.72–0.72.23.001
  18. Anxiousness2.750.940.29–0.59.08.001
  19. Depressivity2.040.890.990.34.14.001
  20. Submissiveness2.430.970.47–0.34.09.001
  21. Emotional lability2.780.930.33–0.26.09.001
  22. Separation insecurity2.010.940.970.34.14.001
  23. Perseveration2.510.940.21–0.73.08.001
  24. Hostility2.450.950.35–0.57.10.001
  25. Suspiciousness2.430.950.46–0.41.10.001
Domains.29.001
  Detachment2.170.730.60–0.04.09.001
  Psychoticism1.930.750.840.05.10.001
  Antagonism1.900.611.041.45.07.001
  Disinhibition2.110.680.680.20.11.001
  Negative affectivity2.410.650.46–0.20.07.001

[i] Note. 1 Lilliefors significance correction; skew – skewness; kurt – kurtosis; PID-5 – Personality Inventory for the DSM-5.

CONFIRMATORY FACTOR ANALYSIS

The fit values for the tested models are shown in Table 2. All models obtained a good fit as they reached appropriate CFI (> .90) and RMSEA (< .08) values. However, the bifactor model provided the best fit (CFI ≥ .95; RMSEA ≤ .06) and was the only one with adequate SRMR values (< .08; Browne & Cudeck, 1992; Hu & Bentler, 1999).

Table 2

Fit indices for the tested models

Modelχ2dfS-B χ2CFIRMSEA [90% CI]SRMRAICModelCAICModel
1. Five-factor oblique1113.95265506.16.983.041 [.036; .047].083626.16941.96
2. Second order: five first-order factors and one second-order factor1173.94270522.95.982.041 [.036; .047].084632.95922.44
3. Bifactor923.88250381.94.990.032 [.025; .038].075531.94926.70
4. One factor1245.43275886.78.958.065 [.060; .070].105986.781249.95

[i] Note. df – degrees of freedom; S-B χ2 – Satorra-Bentler scaled chi-square; CFI – comparative fit index; RMSEA – root mean square error of approximation; CI – confidence interval; SRMR – standardized root mean square residual; AIC – Akaike information criterion; CAIC – constant AIC.

M1, the five correlated-factor model, fit the data well: χ2 = 1113.94 (265), p < .001; CFI = .983; RMSEA = .041, 90% CI [.036; .047]. The estimated parameters were all significant (p < .01) and ranged from .47 to .84 (see Table S1 in Supplementary materials). The correlations between factors were all positive and ranged from .40 (e.g., between detachment and antagonism) to .74 (e.g., between negative affectivity and disinhibition).

Supplementary materials

M2, second-order factor, also had a good fit, but slightly lower than M1: χ2 = 1173.94 (270), p < .001; CFI = .982; RMSEA = .041, 90% CI [.036; .047]. This model had the lowest CAIC (922.43), although it has far from good SRMR values (.084). It had good factor loadings that were all significant (p < .01) and ranged from .46 to .84 (see Table S2 in Supplementary materials). The correlations between the second-order factor (p) and the first-order factors ranged between .61 and .86. Correlations were also observed between the first order factors (between .38 and .73) and were quite similar to M1.

M3, bifactor showed the best fit to the data: χ2 = 923.88 (250), p < .001; CFI = .99; RMSEA = .032, 90% CI [.025; .038]. S-B χ2 was the lowest of all the models (381.94). It is the only one that reaches a CFI of .99. Furthermore, only this model approaches the appropriate cut-off point for SRMR (< .075), and presents optimal RMSEA values (< .032; Hu & Bentler, 1999). However, it should be noted that the CAIC (926.70) was slightly better in Model 2, although the differences are minimal.

Finally, M4, the one-factor model, showed a good fit, although lower than rival models: χ2 = 1245.43 (275), p < .001; CFI = .958; RMSEA = .065, 90% CI [.060; .070]. This model had an inadequate SRMR (.105). Factor loadings ranged from .31 to .67 (see Table 3 in Supplementary materials). Consequently, M3 was chosen.

Table 3 shows the factor loadings of the best-fit model, M3 bifactor. As can be seen, the items (facets) loaded significantly on the general factor, and also had loadings on their expected specific factor. Item loadings on SF differed from item loadings on GF by the corresponding factor: the contribution of the SF in some cases is above the GF, and in other cases, it is below. Although most of the items had loads above .30, there were three exceptions with low loads in the SF but higher in the GF and vice versa (hostility SF = .08, GF = .56; suspiciousness SF = .13, GF = .70; and risk-taking SF = .15, GF = .69). In general, factor loadings of the SF are lower than in Models 1 and 2. In many cases, they are absorbed by the GF.

Table 3

Confirmatory factor analysis of the Personality Inventory for DSM-5 scales: standardized factor loadings based on best-fitting model (bifactor)

PID-5 scales (Facets)DetachmentPsychoticismAntagonismDisinhibitionNegative affectivityGeneral
factor (p)
ECV-I1
1. Withdrawal.45.48.53
2. Anhedonia.49.55.56
3. Intimacy avoidance.51.28.23
4. Restricted affectivity.66.41.28
5. Eccentricity.41.65.72
6. Perceptual dysregulation.37..63.74
7. Unusual beliefs and experiences.57.37.30
8. Grandiosity.56.29.21
9. ANention seeking.52.28.23
10. Manipulativeness.67.39.25
11. Deceitfulness.52.56.54
12. Callousness.31.41.64
13. Risk taking.15.69.95
14. Impulsivity.33.53.72
15. Irresponsibility.40.51.62
16. Distractibility.38.61.72
17. Rigid perfectionism.30.39.63
18. Anxiousness.71.46.30
19. Depressivity.28.68.85
20. Submissiveness.45.35.38
21. Emotional lability.50.38.37
22. Separation insecurity.44.39.44
23. Perseveration.33.62.78
24. Hostility.08.56.98
25. Suspiciousness.13.70.97

[i] Note. 1I-ECV is the item explained common variance attributable to a general factor. Estimated coefficients were all significant (p < .01).

Regarding the bifactor model-based indices, omega was high for GF (total scores) .93, and omegas (ωs) were acceptable: negative affectivity = .85, detachment = .78, psychoticism = .76, antagonism = .79, disinhibition = .77. ωH was .80, which would allow one to conclude, at least partially, in favor of unidimensionality. A high ωH implies that the scores predominantly capture the general factor and, in turn, are due to a single common source even in a multidimensional model (Rodriguez et al., 2016a). ωHs values were: antagonism = .50, detachment = .47, psychoticism = .31, negative affectivity = .28, and disinhibition = .19. According to Smits et al. (2015), antagonism, detachment and psychoticism reflect a substantial proportion of common variance that is unique from the general factor; negative affectivity reflects a moderate proportion of unique variance (between .30 and .20), and disinhibition shows a low (< 20) proportion of unique variance. The H value for the GF was satisfactory (.91), but for the SF values were low (antagonism = .68; negative affectivity = .67; detachment = .63; psychoticism = .46; disinhibition = .37). The coefficients of the specific and general factors differ, obtaining H < .70 for all the specific factors, thereby providing additional evidence in favor of the general factor.

The ECV was just under .60 (i.e., ECV = .56). This suggests that the PID-5 may not be purely unidimensional but may have a structure that is at least somewhat multidimensional (Rodriguez et al., 2016b). As expected, subscores for the multiple subscales will provide added value over simply reporting a total score. PUC was high (.81), so relative bias still is small (Rodriguez et al., 2016a). These results, taken together (ECV, PUC, and high values of omega, ωH, and H values), can be interpreted as supporting a robust general factor, but the GF alone is not sufficient to describe the data. Specific factors (domains) add information beyond the GF.

Four facets (see Table 3) had substantial I-ECV values (>.80): hostility, suspiciousness, depressivity (negative affectivity), and risk-taking (disinhibition), suggesting that they are relatively pure markers of the general factor and not their corresponding specific factor (see Rodriguez et al., 2016b). On the other hand, some facets are strong markers of the general and specific factors (e.g., .50) with corresponding I-ECV values around .50 (e.g., withdrawal, anhedonia, and deceitfulness).

DISCUSSION

The present work aimed to examine the underlying structure of the pathological personality traits of the AMPD model (APA, 2013). For this purpose, different alternative models were compared using CFA. Overall, the results provided evidence compatible with the idea of a general factor of psychopathology. The bifactor model provided a better fit than the original model (with five correlated factors) and the second-order model (with a higher-order factor, p). The one-dimensional model obtained the lowest fit; therefore, p alone would not be enough to describe the underlying structure. Specific factors (domains) add information beyond p. Thus, the bifactor model could be accepted as the best representation of the data.

These findings, while preliminary, suggest that the observed variables reflect five specific pathological personality traits (which correspond to those of AMPD) but also a general factor (which would reflect a general propensity for psychopathology), in line with Caspi et al. (2014), Caspi and Moffitt (2018), and Lahey et al. (2012). According to studies of personality disorder criteria, the general factor represents features of core personality functioning (their severity), whereas the specific factors denote personality style (their content; e.g., detachment, antagonism) (Ringwald et al., 2019). This differentiation could partly explain the high comorbidity among PD, since the most severe cases would meet more than one disorder criteria. Shields et al. (2021) argues that studying the p factor is crucial to identifying transdiagnostic vulnerability to mental disorders.

Additionally, the general factor observed in this work is consistent with previous studies (Gomez et al., 2020; Gutierrez et al., 2021; Montes & Sánchez, 2019) and could explain the substantive cross-loadings between several facets observed across several studies (DeYoung & Krueger, 2018; Krueger et al., 2012; Lahey et al., 2017; Watters & Bagby, 2018; Wright et al., 2012; Sanchez et al., 2023).

LIMITATIONS AND FUTURE RESEARCH

The present study has several limitations. Probably the most important is related to flaws inherent to this version of the PID-5. We worked with a modified version that needs more evidence of validity, which may reduce the scope of the conclusions derived from this work. Although this version showed good psychometric properties (Sanchez et al., 2023), some facets had modest reliability, which could have affected some domain’s values.

Secondly, because we used a convenience sample (only 23.8% were undergoing psychological and/or psychiatric treatment), the findings here may not be applicable to other samples, including clinical samples. In addition, the sample composition (mainly female) limits the possibility of generalizing the results. Therefore, this research should be replicated with clinical population data (e.g., to study the ability of the PID-5 to differentiate clinical from non-clinical samples).

Third, the bifactor model itself also has some drawbacks that must be considered. Although the statistical results suggest latent variables, this does not necessarily mean the existence of an underlying structure of psychopathology. Conversely, these results could be only a statistical artifice (Bonifay et al., 2017; Forbes et al., 2021; Greene et al., 2019). As Reise et al. (2016) point out, one should be cautious in concluding that the bifactor model is better only based on a superior model fit. For this reason, in this work, we used different criteria, including theoretical considerations (as mentioned above) and statistical indices derived from bifactor models (e.g., ECV, PUC, H). As noted by Caspi and Moffitt (2018), any new proposal must be supported with contributions from other areas of knowledge (e.g., genetic and neuropsychological aspects, developmental psychopathology). These contributions can provide a broader framework for empirical findings. Thus, a new psychopathological model, clinically useful, must articulate the available empirical evidence with theoretical work.

Fourth, the interpretation of the p factor can be confusing. This construct has been understood heterogeneously as a general vulnerability to psychopathology, a general maladaptation, a risk of a greater propensity for mental disorders, as the result of maladaptive consequences shared by all disorders, or even as a response bias (Caspi et al., 2014; Gutierrez et al., 2021; Ruggero et al., 2019). Future research would benefit from further understanding how this general factor, specifically in personality psychopathology, is understood. In this sense, the relationship with criterion A could help to clarify the meaning of the general factor. In fact, previous research has shown that measures of criterion A and B are highly correlated, and that criterion B is informative of criterion A severity (see for example Bastiaens et al., 2021; Zimmermann et al., 2019).

Despite these limitations, this study provides support for the existence of a strong general factor of personality psychopathology. The results can contribute to a better understanding of the underlying structure of the AMPD, assessed with an Argentine version of the PID-5.

Supplementary materials are available on journal’s website.