Abstract
The introduction of the Alternative Model for Personality Disorders (AMPD) in the
DSM-5 indicated a formal shift towards a more dimensional model for personality
disorders. Despite being deemed as complex, high-volume, high-cost disorders, with
detrimental impacts on all aspects of a person’s life, personality disorders are
frequently overlooked and left untreated. A key contributing factor is the lack of
reliable and valid assessment measures for the presence of personality pathology.
To this extent, the introduction of the Personality Inventory for the DSM-5 (PID-5)
brought with great promise a more dimensional approach in the assessment of
personality pathology. However, despite having a much stronger empirical
foundation, the 220-item length of the PID-5 poses a significant barrier for its clinical
utility. The Personality Inventory for the DSM-5 Short Form (PID-5-SF) addresses
this limitation by capturing both the breadth and depth of the original PID-5 using
only 100 items. However, despite being considered the most effective version of the
PID-5, research on the PID-5-SF remains scarce, particularly research related to its
underlying structure. It is further noted that obtaining evidence on the structure of
any measure is essential as the structure ultimately dictates how scales are scored
and interpreted.
Moreover, within the field of psychometry, latent variable models tend to serve
as the standard theoretical framework from which to study psychological
phenomena. In fact, most of our knowledge on human behaviour to date - including
personality disorders - are based on data derived from measures rooted in a latent
variable model framework. Arguably, a key limitation of latent variable models is its
assumption of local independence which posits that observable indicators (i.e.,
symptoms) only co-occur due to the presence of some underlying latent variable
(i.e., disorder) but once the latent variable has been conditioned for, observable
indicators are rendered independent of one another. Consequently, this discounts
any influential relationships that may exist between observable indicators. A network
model approach, however, has shown to capture these interrelationships, providing
an alternative conceptual framework from which to conceptualise psychopathology.
At the core of network model theory is the assumption that psychological disorders,
including personality disorders, stem from the dynamic, reciprocal interactions
4
between observable indicators. Thereby, capturing exactly that which latent variable
models have failed or are limited to acknowledge.
To this end, the present study aimed to evaluate the structure of the PID-5-SF
using methods from both latent variable- and network models. More specifically, this
study endeavoured to evaluate the dimensionality, reliability, hierarchical factor
structure, and sum scoring model of the PID-5-SF using a general adult sample from
a multicultural South African population (n=1465). Dimensionality at the facet and
domain level were evaluated using parallel analysis and exploratory graph analysis
(EGA). Reliability was determined using both Cronbach’s alpha and McDonald’s
omega. The postulated hierarchical structure of the PID-5-SF was evaluated by
fitting a higher order factor model to the data and inspecting general fit indices. To
obtain evidence for the parallel or sum scoring model used by the PID-5-SF, the fit of
three unidimensional models were evaluated, namely, parallel, tau-equivalent, and
congeneric. Finally, the present study aimed to explore the dynamic
interrelationships between the various maladaptive facet scales of the PID-5-SF by
conducting a network analysis on the 25 facet scales.
Results supported the reliability and hierarchical factor structure of the PID-5-
SF. Although the present study provided evidence for unidimensionality of the 25
facet scales, the expected and postulated five-dimensional structure for the overall
PID-5-SF was not replicated in the present sample. Instead, results from the EGA
delivered a three-dimensional structure. The unexpected dimensional structure of the
PID-5-SF seemed to align strongly with the theorised descending hierarchical model
proposed for the original PID-5 by Wright et al. (2012). Further, results on the sum
scoring model of the PID-5-SF supported its use when applied for general
assessment purposes; however, the same scoring method may not suffice when the
PID-5-SF is used to conduct more rigorous research, or to make formal diagnoses.
Finally, the PID-5-SF network structure revealed a substantial amount of
interconnectedness between the 25 maladaptive facets of the PID-5-SF, possibly
explaining the high comorbidity observed between different personality disorders.
Intimacy Avoidance, Emotional Lability, Anxiousness, and Submissiveness appeared
to be key role players in the PID-5-SF network. Results from the present study
highlighted the need for additional research on the suitability of the PID-5-SF to the
multicultural South African context.
5
Keywords: Alternative Model for Personality Disorders (AMPD); Personality
Inventory for the DSM-5 Short Form (PID-5-SF); Personality Inventory for the DSM-5
(PID-5), latent variable models, network analysis, network psychometrics,
psychological network models, personality assessment, personality disorders