Psychometric Properties of the Chinese Version of the Personality Inventory for DSM-5 Brief Form in Undergraduate Students and Clinical Patients

Background: The Personality Inventory for DSM-5 Brief Form (PID-5-BF) is a 25-item measuring tool evaluating maladaptive personality traits for diagnosis of personality disorders(PDs). As a promising scale, its impressive psychometric properties has been veried in some countries, however, there has no studies about the utility of PID-5-BF in Chinese settings. The current study aimed to examine cultural applicability of the Chinese version of PID-5-BF among undergraduate students and clinical patients. Methods: 7155 undergraduate students and 302 clinical patients completed the Chinese version of PID-5-BF. 228 students were chosen randomly for test-retest reliability at a 4-week interval. Exploratory factor analysis (EFA) and conrmatory factor analysis (CFA) were conducted to discover the most suitable construct in Chinese, measurement invariance(MI), internal consistency and external validity were also calculated. Results: An exploratory six-factor model was supported more suitable in both samples(Undergraduate sample: CFI = 0.905, TLI = 0.888, RMSEA = 0.044, SRMR = 0.039; Clinical sample: CFI = 0.904, TLI = 0.886, RMSEA = 0.044, SRMR = 0.063), adding a new factor“Interpersonal Relationships”. Measurement invariance across nonclinical and clinical sample was established(congural, weak, strong MI, and partial strict MI). Aside from acceptable internal consistency(Undergraduate sample: alpha=0.84, MIC=0.21; Clinical sample: alpha=0.82, MIC=0.16) and test-retest reliability(0.73), the association with 220-item PID-5 was signicant(r = 0.93, p < 0.01), and six PDs measured by Personality diagnostic questionnaire-4+ (PDQ-4+) was correlated with expected domains of PID-5-BF. Conclusions: The Chinese version of the PID-5-BF showed satisfactory psychometric properties, which is a convenient and useful screening tool for personality disorders. standard root mean square residual; RMSEA, root-mean-square error of approximation; LO90/HI90, lower/upper 90% condence interval


Background
Personality disorders (PDs) are common psychiatric conditions that are disruptive to everyday functioning. Because PDs are often misdiagnosed or missed entirely, reliable and valid clinical diagnostic tools for PDs are needed [1]. In light of the numerous weaknesses of the traditional PD taxonomic diagnostic system, including a high comorbidity rate, arbitrary cutoff scores, and substantial heterogeneity within PDs in Section III of the DSM-5 [2], the APA (2013) proposed an Alternative Model of Personality Disorder (AMPD) diagnosis, wherein 25 maladaptive personality trait facets are organized into ve domains (Criterion B) after measuring personality functioning (Criterion A) [3]. The AMPD provides theoretical trait domains across six speci ed PDs, thus converting PD diagnosis from a categorical to a dimensional scheme.
The Personality Inventory for DSM-5 (PID-5) is a 220-item self-report measurement that was developed speci cally to evaluate hierarchically organized personality traits in accordance with the AMPD. The reliability and validity of the PID-5 have been con rmed in multiple studies, which have yielded internal consistency values above 0.8 for most domains [4][5][6][7]. The ve broad domains a rmed in prior studies [8,9] have been described as comparable to maladaptive variants of the Big-Five Model [10], which would be expected given that the development of the associated dimensional model of personality pathology was informed by the normal personality taxonomy [11]. Although the PID-5 has satisfactory psychometric properties, its utility is limited due to its having a large number of items, and thus its taking substantial time to complete [12]. With the aim of screening for PDs quickly and accurately, the PID-5 Brief Form (PID-5-BF) was developed by extracting core items from the ve domains (Negative Affect, Detachment, Antagonism, Disinhibition, and Psychoticism) of the PID-5 [13]. Correlation coe cients between the ve PID-5-BF domains and the original PID-5 domains have been reported to be very good, with Bach et al. (2016) reporting a mean correlation coe cient value of 0.90 [14], and Debast et al. (2017) reporting correlation coe cients in the range of 0.81-0.87 [15].
PID-5-BF results provide an overall assessment of degree of personality maladjustment and point to potential PDs [16]. The PID-5-BF has been shown to be reliable and valid in a number of countries in Europe [14][15][16][17][18], North America [19][20][21], and Asia [22]. In all but two cases, internal consistency coe cients above 0.8 were obtained; internal consistency coe cients for Belgium [15] and Denmark [14] were between 0.66 and 0.87. The PID-5-BF has been shown to have satisfactory discriminant validity (effect size of mean domain level = 0.46, p < .05) between individuals with and without PDs [14]. In terms of criterion validity, each PD measured by Personality Diagnostic Questionnaire-4(PDQ-4) was associated with and predicted by theoretical PID-5-BF domains. Regarding internalizing and externalizing criteria, the PID-5-BF Negative Affect domain subscore is a robust predictor of Inventory for Depression and Anxiety Symptoms-2 scores. Meanwhile, the PID-5-BF Disinhibition and Antagonism domain subscores are speci cally predictive of Externalizing Spectrum Inventory scale scores. Somewhat unexpectedly, PID-5-BF Psychoticism domain subscores were found to be signi cantly predictive of scores for the aforementioned three scales [20]. PID-5-BF domain subscores and total scores correlated signi cantly with Personality Assessment Screener scores [21].
Although the PID-5 domains have been understood as putative maladaptive variants of Big-Five Model dimensions, the appropriateness of the Big-Five Model differs across cultures. For example, a six-factor model has been shown to be more suitable than a ve-factor model in Chinese populations, while still tting data from multiracial samples across Asia, Europe, and North America [23][24][25]. The sixth factor has been suggested to re ect "Interpersonal Relatedness", which encompasses the traditional Chinese concepts of relationship orientation (Ren Qing; i.e. reciprocity of favors, affections, etc.), Harmony (e.g. lack of con ict, balance), and Face (as in "saving face" or reputation). A seven-factor model was also proposed [26] and con rmed in a sample of Chinese undergraduate students [27]. The Big-Five Model was found to have a poor t with an Indian sample [28]. Similarly, the ve-factor model of the PID-5-BF did not achieve an adequate t in a study of Filipino college students [22]. In consideration of cultural dissociations of personality constructs, it is also of interest to note that the replicability of the Openness domain of the NEO Personality Inventory was also found to be poor in Asian countries in a cross-cultural study that included 24 cultures [29].
Given that ve-factor models of the PID-5-BF have been supported principally in studies with Western samples, it is likely that cultural differences may underlie differing factor structure ndings in the literature. Accordingly, the most suitable factor structure of the PID-5-BF in Chinese respondents remains to be clari ed. Measurement invariance (MI) studies of the PID-5-BF should be conducted to determine whether factor structure differs between populations constituted by individuals of different cultures [30,31]. Although MI of the PID-5 has been examined across cultures, sexes, and across clinical and nonclinical samples [32][33][34], the MI of the PID-5-BF is unknown.
The current study was the rst to examine the psychometric properties of the Chinese version of PID-5-BF. The aims of this study were threefold. First, we set out to determine the most suitable factor structure of the PID-5-BF in a Chinese population sample. Second, we assessed MI of the Chinese PID-5-BF across normal students and clinical patient samples, which is important for generalization of the PID-5-BF in both research and clinical settings. Third, to investigate criterion-related validity, we analyzed how well PID-5-BF scores correlate with scores obtained on the 220-item PID-5 and PDQ-4+.

Participants
The normal sample consisted of 7,985 university students recruited from two Chinese universities in Hunan Province. After omitting subjects with missing data values, we retained a nal normal sample of 7,155 subjects (3,   years, who had been referred to the psychological clinic in hospital for assessment and treatment. Each patient was diagnosed based on the Structured Clinical Interview for DSM-IV system by two experienced psychiatrists (WX, LXW). The distribution of diagnoses in the clinical sample were as follows: PD, 61.9%; major depressive disorder, 16.6%; anxiety disorder, 6.3%; bipolar disorder, 6.0%; obsessivecompulsive disorder, 4.6%; schizophrenia, 3.3%; and other mental disorders, 1.3%.
All study procedures were approved by the Ethics Committee. All participants signed an informed consent.

Instruments
Chinese PID-5 and PID-5-BF The full-length PID-5 [35] is a 220-item self-report scale developed in the USA to index 25 lower-order trait facets (Cronbach's α, 0.72-0.96) organized into ve higher-order trait domains of personality pathology (Cronbach's α, 0.84-0.96) [35]. We invited two psychologists to translate the PID-5 scale from American English into Chinese; and then it was translated back into English by a bilingual teacher, with repeated revisions to ensure translation accuracy. The PID-5-BF [36] was developed by extracting 25 items from the original PID-5, representing 21 of the 25 trait facets (facets not included: Restricted Affectivity, Rigid Perfectionism, Submissiveness, and Suspiciousness). Items are rated on a 0-3 Likert-type scale, with higher scores representing greater dysfunction. Each of the ve higher-order domains is represented by ve items (Negative Affect: Items 8,9,10,11

Model testing procedures
In the normal sample, 3,985 students nished the 220-item PID-5, and 7,155 students completed the PID-5-BF and PDQ-4+. For the clinical sample, we obtained 302 valid PID-5-BF, 224 valid PID-5, and 231 valid PDQ-4 + questionnaires. For evaluation of test-retest reliability, 228 normal sample participants (93 male, 135 female) were chosen randomly for a PID-5-BF re-test taken 4 weeks after the initial test. Construct validity of the Chinese PID-5-BF was assessed with exploratory factor analysis (EFA) and con rmatory factor analysis (CFA) after randomly (roughly) halving the normal sample randomly into an EFA subsample (N = 3,633) and a CFA subsample (N = 3,522). CFA was conducted to test the theoretical model suggested to be the best tting model in the EFA.
MI tests across population included four nested models. Model 1 (con gural invariance) tested the factor structure of latent variables across our two population samples with all parameters freely estimated. Model 2 (weak invariance) was based on the con gural results with factor loadings equalized across groups. Next, Model 3 (strong invariance) had consistency of variable intercepts; achieving this model indicates that latent factor scores have the same meaning across groups, and thus that group comparisons are tenable. Model 4 (strict invariance) requires equalized error variance on the basis of the previous three models [41], which is rarely achieved [42].

Data analysis
Data analysis was performed in IBM SPSS Statistics 23.0 and Mplus 7.4. To examine construct validity, we rst conducted EFA to identify the most suitable factor model of the PID-5-BF. Oblique rotation was used to allow correlation among factors. Each item with a factor loading ≥ 0.3 was accepted as a factor component [43]. For EFA and CFA, model t indices applied included the comparative t index (CFI), the Tucker-Lewis index (TLI), the standard root mean square residual (SRMR), and the root mean square error of approximation (RMSEA) with a 90% con dence interval (CI). Acceptable t values were: CFI ≥ 0.90, TLI ≥ 0.90, SRMR ≤ 0.08, and RMSEA ≤ 0.08 [44]. Model modi cations were made on the basis of item correlations and MI index values.
MI was estimated based on three indices, namely △CFI, △TLI, and △RMSEA, wherein △ represents the difference between two adjacent models. Invariance was veri ed when △CFI and △TLI were ≤0.01 and △RMSEA was < 0.015 [45]. In cases where these criteria were not met, the largest-value modi cation indices were selected to determine the parameters of which item(s) should be released to be free across groups iteratively until the △CFI was ≤0.01, demonstrating potential partial invariance [32].
Internal consistency was determined by calculating Cronbach's α and mean inter-item correlation (MIC) values. Cronbach's α > 0.8 and > 0.9 signi ed acceptable and good reliability, respectively; a MIC > 0.15 was considered acceptable [46]. Test-retest reliability was estimated by calculating a Pearson's correlation coe cient (r). To examine criterion validity, Pearson's r index values were also calculated between the PID-5-BF and the original PID-5, as well as between the PID-5-BF and the PDQ-4+ (representing the six DSM-4 PDs retained in DSM-5, Section III). Pearson r values > 0.30 and > 0.50 indicated medium and large effect sizes, respectively [47].

EFA and CFA
EFA supported an exploratory ve-factor model and an exploratory six-factor model. The factor designations of the items are compared across these two exploratory models and Krueger et al.'s (2013) previously reported vefactor model [36] in Table 1. In our exploratory ve-factor model, ve items (8,9,11,15, and 20) did not load into any factor; item 10 was the only item to reach a 0.3 loading weight in the Negative Affect domain, and it loaded with item 19, which had belonged to the Antagonism factor in the previously published model. Our exploratory six-factor model had higher t indices (CFI = 0.969, TLI = 0.944) than the exploratory ve-factor model (CFI = 0.952, TLI = 0.922), with fewer items failing to load on any factor (items 8 and 11). A new factor named Interpersonal Relationships was added to the original ve factors. The factor loadings of each item in the exploratory six-factor model are reported in Table 2.  Bold represents the largest factor loading in each item as well as > 0.30.
We chose to pursue analysis of our exploratory six-factor model because it had better model t indices and fewer items that failed to load than our exploratory ve-factor model and because of the signi cant differences in the Negative Affect domain between our exploratory ve-factor model and the theoretical ve-factor model. As shown in Table 3, we obtained signi cantly greater t indices for our exploratory six-factor model than for the theoretical ve-factor model in both our undergraduate student sample and clinical sample.

MI across populations
As shown in Table 4, we established con gural, weak, and strong MI across the normal and clinical samples.
However, the acceptable index criteria were not met for strict MI. We allowed the residual variances of items with the largest modi cation indices to be freely estimated until the △CFI of the last model was ≤ 0.01. Parameter constraints of items 14, 4, 12, 20, 7, 15, 5, and 17 were released in this process. Subsequently, partial strict MI of our modi ed six-factor model was supported. Hence, ultimately, our modi ed six-factor PID-5-BF model achieved con gural MI, weak MI, strong MI, and partial strict MI across our normal and clinical samples.   Table 5, the Interpersonal Relationships domain showed the greatest correlation coe cient with the Negative Affect domain in both samples. Finally, as shown in Table 6, PID-5-BF domain scores correlated signi cantly with the six PD dimensions in Section III of the DSM-5 (schizotypal, antisocial, borderline, narcissistic, avoidant, and obsessive-compulsive).

Discussion
Contrary to most previous studies of the PID-5-BF in other countries, in this study, we found that a six-factor model was more suitable than the theoretical ve-factor model for the PID-5-BF in our Chinese sample. Following modi cations, our six-factor PID-5-BF model achieved con gural MI, weak MI, strong MI, and partial strict MI across our normal and clinical samples. In agreement with prior studies [14,20], we found that the PID-5-BF was highly correlated with the original 220-item PID-5 and the domains generally correlated with the six PDs retained in Section III of the DSM-5. We obtained acceptable Cronbach's α and MIC values for the PID-5-BF [46], revealing a good internal consistency of PID-5-BF, similar to prior studies [16,17,20]. Furthermore, our nding of good test-retest reliability over a 4-week interval is in agreement with prior work showing similarly good test-retest reliability of the PID-5-BF over a 2-week interval in a sample of high school students [16].
Factor structure Although a number of prior studies conducted in Western-culture populations have supported a ve-factor model for the PID-5-BF [14,15,16,19,20], our results supported a six-factor model better than the theoretical ve-factor model derived from the literature. Interestingly, a prior study conducted with Filipino college students also showed a relatively poor t of the theoretical ve-factor model for the PID-5-BF [22]. Thus, we speculate that the discrepancy may re ect differences in the way people from Western versus Eastern cultures understand personality constructs and thus interpret items of the PID-5-BF.
Four of the factors in our exploratory six-factor model were consistent with the theoretical ve-factor structure. Only the Negative Affect domain failed to align, and items 10 (fear being alone) and 19 (I crave attention) were placed in the newly added factor called Interpersonal Relationships. According to traditional personality theories in Western cultures, which focus on internal characteristics of the individual [48], loneliness is regarded as a source of distress related to experiencing a lack of empathy. On the contrary, in Eastern cultures, which are generally more collectivist, individuals are often considered to be inherently closely connected with others [49].
Accordingly, loneliness may be viewed as an isolated state due to poor interpersonal skills. Item 19, which is associated with the Antagonism domain of the original PID-5-BF, refers to behaviors that put the individual at odds with others, including an exaggerated sense of self-importance and an expectation of special treatment [10]. However, in a collectivist society, one's sense of belonging is an important aspect of his or her personality constitution [49]. Consequently, item 19 is likely to be understood as one's desire to t into a certain group and to communicate with others. Prior studies have explored the in uences of collectivist versus individualist cultures on personality [50][51][52]. In summary, cultural differences in how one understands and interprets Items 10 and 19 may affect the factor loading of these two items.
The unique structure of the DSM-5 personality trait model in Chinese respondents, compared to respondents from most other examined countries, may be related to cultural differences in general personality models [35,53]. Although the ve-factor model has been widely used globally, it may not be fully applicable in a variety of cultural contexts due to its Western-centric derivation. Consistent with this supposition, the ve-factor model was not well-tted when the NEO Personality Inventory was examined in the Philippines [54], Korea [55], and Japan [56]. Hence, it appears that the Big-Five Model does not fully explain personality traits in collectivist society contexts [23]. Prior studies have proposed a six-factor hypothesis of Chinese personality traits, with the addition of Interpersonal Relationships [25]. The importance of this sixth dimension for Chinese personality analysis has been a rmed in Chinese Personality Assessment Inventory standardization studies [24]. Therefore, although many western personality tests are reasonably reliable and valid when applied to Chinese samples, there are some cultural deviations to be considered [25].

MI
To the best of our knowledge, this study is the rst to explore MI of the PID-5-BF. Establishment of MI provides evidence of a consistent underlying structure across groups and thus enables group means to be compared [30].
When performing nested MI modeling, as was done here with con gural, weak, strong, and strict MI, MI must be established sequentially from lower-to higher-level MI analyses [57]. We were able to achieve MI fully with respect to factor structure (con gural MI), metric (weak MI), and intercept (strong MI) equivalences for the PID-5-BF in both samples. Strict invariance was partially satis ed.
Because our modi cation index analyses led us to release constraints on Items 14,4,12,20,7,15,5, and 17 to better achieve strict invariance, it can be deduced that the residual variances of these items were not equivalent across our two sample groups. Notwithstanding, upon achieving strong MI, we were able to conclude that our nding of higher PID-5-BF scores in our clinical sample, compared to our normal sample, could be considered a reliable nding. Moreover, these data a rm a satisfactory discriminant validity of the PID-5-BF for differentiating between nonclinical and clinical individuals.

External validity and clinical value
Although the 220-item PID-5 has many merits for personality diagnosis-such as close relations with clinical symptoms, the ability to be combined with various psychotherapy methods, and good stability over time [12]its length hinders its clinical utility. Pires et al. (2018) examined the psychometric properties of the 220-item (original), 100-item (short form), and 25-item (brief form) PID-5 versions in a sample of Portuguese university students and concluded that any of the three could be used to assess maladaptive personality traits reliably and validly [18]. Bach et al. (2016) compared the three forms in a Danish population and showed that the three scales were highly similar with respect to internal consistency, factor structure, discriminant validity, and correlation with DSM-4 PD dimensions [14]. The present ndings of very strong correlation coe cients between the PID-5-BF and the 220-item PID-5 in both of our samples indicate that in addition to saving time, reducing the burden upon participants, and being generally more clinic friendly, the PID-5-BF maintains the validity of the original instrument to a remarkable degree.
Moreover, the six factors of the PID-5-BF in our six-factor model showed good alignment with the six PDs in Section III of the DSM-5. Each PD correlated directly and speci cally with its expected domain, with the exception of obsessive-compulsive personality disorder, which demonstrated good continuity from the DSM-4 to the DSM-5. Our unexpected nding of the Psychoticism domain showing strong correlations with most of the PDs in our normal sample may due to college students being sensitive to abnormal behaviors, thereby limiting the speci city of Psychoticism. In summary, the PID-5-BF retained satisfactory psychometric properties, despite its extensive omission of items relative to the 220-item PID-5, a rming its suitability as a preliminary clinical PD screening tool.
The PID-5-BF can be used to differentiate between psychologically healthy and troubled respondents, at least preliminarily. Bach et al. (2016) reported that the PID-5-BF has very good discriminant validity between psychiatric outpatients and community-dwelling individuals [14], consistent with our ndings of signi cantly higher PID-5-BF scores in our clinical patients than in our normal sample of undergraduate students. Clinicians can administer the PID-5-BF to acquire a rough estimation of one's personality functioning, laying the foundation of further treatment planning, and then judge the need for additional assessments. Although the PID-5-BF may not provide unique clinical information regarding speci c symptoms, it can describe personality traits through the assessment of dimensions, embodying differences in degree rather than in category, contributing to individualized therapy development.

Limitations and future directions
The present study had three noteworthy limitations. First, the retested sample and clinical population were relatively small due to practical limitations. Second, the clinical sample was heterogenous, including patients diagnosed with various psychological disorders. Third, the current study was cross-sectional, and crosssectional studies cannot demonstrate predictive validity with the robustness of longitudinal studies. Hence, there is a need for larger longitudinal and clinical-sample studies of the PID-5-BF, particularly with samples constituted by patients with PDs.
Regarding future directions of research, because dimensions represent continua from normal to abnormal, actionable score ranges need to be established based on ample empirical data collected in clinical practice rather than developed from theoretical hypotheses. Further correlational analyses between the PID-5-BF and other psychological scales are also needed to clarify dimensional distinctions among different psychiatric diagnoses. MI should also be further examined across genders, age bands, and cultures, particularly in Asia and the Paci c Islands.

Conclusion
The Chinese version of the PID-5-BF had satisfactory internal consistency and criterion validity. Regarding factor structure, it was well-tted to our exploratory six-factor model, which included the additional Interpersonal Relationships domain, compared to the theoretical ve-factor model, in both our undergraduate student and clinical samples. MI across normal and clinical Chinese samples was established. The PID-5-BF is suitable for assessing personality traits and clinical screening for PDs quickly.

Declarations
Ethics approval and consent to participate This study was approved by the ethics committee of Second Xiangya Hospital, Central South University. All participants were over 16 years old and had written informed consent.

Consent for publication
Not applicable.
Availability of data and materials