Exploratory Graph Analysis of the Multidimensional Schizotypy Scale

Abstract

The present study examined the dimensional structure underlying the Multidimensional Schizotypy Scale (MSS) and its brief version (MSS-B). We used Exploratory Graph Analysis (EGA) to evaluate their dimensional structure in two large, independent samples (n = 6265 and n = 1000). We then used Confirmatory Factor Analysis (CFA) to compare the fit of the theoretical dimensions with the EGA dimensions. For the MSS, EGA identified four dimensions: positive schizotypy, two dimensions of negative schizotypy (affective and social anhedonia), and disorganized schizotypy. For the MSS-B, EGA identified three dimensions, which corresponded to the theorized positive, negative, and disorganized dimensions. Based on the MSS's EGA dimensions, we also estimated a four-factor model for the MSS-B. The CFA comparison found that the four-factor model fit significantly better than the theoretical three-factor model for both the MSS and MSS-B, providing support for the theoretical model and offering a more nuanced interpretation of the negative schizotypy factor. In addition, EGA also revealed that the positive and negative schizotypy dimensions of the MSS and MSS-B might be mediated by the disorganized dimension. Our findings offer new implications for future research on the MSS and MSS-B dimensions that may provide differential associations with interview and questionnaire measures.

Introduction

Schizotypy is a multidimensional construct that offers a promising framework for understanding the development and etiology of schizophrenia-spectrum psychopathology (Lenzenweger, 2010). Despite extensive evidence demonstrating the validity and utility of the construct of schizotypy, the proposed underlying structure has varied considerably across studies and measures (Kwapil and Barrantes-Vidal, 2015). Specifically, there has been disagreement regarding the heterogeneous nature of the construct, as well as variability regarding the content and structure of the questionnaires developed to measure schizotypy (Gross et al., 2014). Previous factor analytic studies typically identified two to five dimensions of schizotypy; however, current conceptual models indicate that positive, negative, and disorganized schizotypy are the strongest supported dimensions (e.g., Cohen and Fonseca-Pedrero, 2017; Gross et al., 2014; Kwapil and Barrantes-Vidal, 2015). The dimensional structure of schizotypy is usually examined using confirmatory factor analysis (CFA).

More recently, psychometric network models, such as Exploratory Graph Analysis (EGA; Golino and Epskamp, 2017), have been used as an alternative method to identify the dimensional structure of constructs. When defining dimensions, psychometric network models do not rely on a priori assumptions but instead develop an emergent structure based on the data. EGA is exploratory in nature and allows constructs to be vetted by using a model that does not conform to a researcher's a priori beliefs; therefore, it's an ideal method to test or re-evaluate the theoretical structure of a construct. Evidence suggests that EGA has comparable or better accuracy identifying dimensions than traditional factor analytic methods (Golino and Demetriou, 2017; Golino and Epskamp, 2017; Golino et al., 2018). In the present study, we sought to evaluate the dimensional structure of the Multidimensional Schizotypy Scale (MSS; Kwapil et al., 2018b) and its brief version, Multidimensional Schizotypy Scale-Brief (MSS-B; Gross et al., 2018b), using EGA in two large samples.

Questionnaire measures are widely used to assess schizotypy (see review by Kwapil and Chun, 2015). However, extant measures suffer from conceptual and empirical limitations, including not mapping on to current models of schizotypy, psychometric shortcomings, and outdated items. The MSS and MSS-B were recently designed to improve upon these limitations and provide theoretically-based and psychometrically sound assessments of positive, negative, and disorganized schizotypy—the three most commonly identified dimensions in the literature (American Psychiatric Association, 2013; Kwapil and Barrantes-Vidal, 2015; Tandon et al., 2009). These three dimensions have also been shown to be invariant across cultures (Fonseca-Pedrero et al., 2018c) and are representative of schizophrenia-spectrum disorders (Lenzenweger and Dworkin, 1996).

The positive dimension involves disruptions in the content of thought (e.g., magical ideation and delusions), perceptual oddities (e.g., illusions and hallucinations), and paranoia/suspiciousness. The negative dimension is characterized by diminished experiences and expression (e.g., alogia, anergia, avolition, anhedonia, and flattened affect). The disorganized dimension involves cognitive-behavioral disturbances in the organization and expression of thoughts and behavior. Before the MSS and MSS-B, most schizotypy scales captured components of one or more of these three dimensions, but none comprehensively measured this theoretical three-factor structure (e.g., Gross et al., 2014).

The development of the MSS and MSS-B adhered to comprehensive scale development guidelines (DeVellis, 2012), including the development of detailed trait specifications for each dimension that guided item development. Items were selected based on content validity, item response theory (IRT), classical test theory (CTT), and differential item functioning (DIF). These methods were employed to overcome limitations of previous schizotypy scales and to produce new schizotypy scales based on a strong theoretical foundation that possessed robust psychometric properties. To date, the MSS and MSS-B have shown good reliability, high item discrimination, and negligible item bias for sex and ethnicity (Gross et al., 2018b; Kwapil et al., 2018b). Furthermore, initial studies support the construct validity of both the MSS (Kwapil et al., 2018a) and the MSS-B (Gross et al., 2018a). The utility of schizotypy, and more specifically the scales that measure it, however, depends on the clear articulation of its multidimensional structure (Kwapil and Barrantes-Vidal, 2015). Thus far, the three-dimensional structure of the MSS and MSS-B has yet to be rigorously investigated. To investigate this structure, we applied the network approach.

Network psychometrics is a rapidly developing field that has been applied to many psychopathological constructs, including schizotypy (Christensen et al., 2018b; Fonseca-Pedrero et al., 2018b). The psychometric network approach defines constructs (e.g., schizotypy) as complex systems, which arise from mutually reinforcing interactions between the construct's constituent elements (e.g., schizotypy items; Borsboom and Cramer, 2013; Schmittmann et al., 2013). This definition forms the foundation of the network theory of psychopathology, which suggests that symptoms can reinforce one another, be influenced by other factors (e.g., biological, environmental, and social mechanisms), and lead to self-sustaining states that persist at the level of disorder (Borsboom, 2017). This theory aligns with current assessments of schizotypy as the latent liability of schizophrenia spectrum disorders, where interactions with biological and environmental influences may facilitate transition into disorder (Isvoranu et al., 2016; Lenzenweger, 2018).

Psychometric network models consist of nodes which represent variables (e.g., MSS items) and edges or connections which represent relations between the nodes (e.g., partial correlations given all other nodes in the network). Partial correlations are the unique shared variance between nodes in the network, which typically shrink many relations near or to zero. Often larger relations that remain form communities (or sets of many connected nodes) in the network. These communities are shown to be mathematically equivalent to factors (Golino and Epskamp, 2017).

One network method, EGA, was recently developed to detect and discover these communities (Golino and Epskamp, 2017). EGA applies a Gaussian Graphical Model (Lauritzen, 1996), which is computed using the graphical least absolute shrinkage and selection operator (glasso; Friedman et al., 2008). Then, the walktrap community detection algorithm is applied to identify the dimensions of the network (Pons and Latapy, 2006). The walktrap algorithm uses “random walks” to identify the content and number of dimensions in the network. Random walks are steps or jumps from one node to another in the network. Each node is repeatedly used as a starting point, traversing over neighboring edges, with larger edge weights (i.e., partial correlation values) being more likely to be traversed. In this process, communities form based on a node's proportion of many, densely connected edges and few, sparsely connected edges.

The dimensions discovered by EGA are deterministic and require no direction from the researcher. Thus, EGA offers a potential advantage over other exploratory dimension reduction methods because the content and number of dimensions are immediately interpretable, without the need to interpret component loadings of individual items. Despite the deterministic allocation of items, researchers should still verify the theoretical consistency of item placement. In both simulation and real-world datasets, EGA has produced comparable or better accuracy in identifying dimensions than other more common dimension reduction methods (e.g., principal component analysis, factor analysis, parallel analysis; Golino and Demetriou, 2017; Golino and Epskamp, 2017; Golino et al., 2018). Moreover, EGA has been effective at replicating factor analytic findings (Bell and O'Driscoll, 2018) as well as discovering new dimensions of constructs (Christensen et al., 2018a).

An advantage of psychometric network models more generally is that they allow a representation of item-level relations that afford interpretations across hierarchical resolutions—that is, the influence of item-level relations can be understood between items, within and between dimensions, and at the level of the construct itself (Blanken et al., 2018; Letina et al., 2018). Latent variable approaches assume local independence, suggesting that items are independent given a latent variable (Edwards and Bagozzi, 2000). These latent variables may then be correlated amongst themselves or independently related given some superordinate latent variable (e.g., the construct). From this perspective, the hierarchical resolution of relations is discrete and lateral, meaning relations are only allowed at the level of latent variables and these relations only occur across this level but not above (i.e., construct-level) or below (i.e., item-level). By contrast, psychometric network models permit a continuous resolution of each variable's relations, occurring simultaneously rather than independently.

The goal of the present research was to validate the theoretical dimensional structure of the MSS and MSS-B in two large, independent samples. To do so, we implemented EGA to discover the dimensional structure of the MSS and MSS-B. Because EGA does not impose a priori assumptions about the dimensional structure of the scales, it stands as an exploratory test for whether the theoretical dimensions intended by the scales' developers are measured as intended. Then, we used CFA to compare the dimensions identified by EGA to the theoretical MSS and MSS-B dimensions. For all analyses, we expected to find three factors, corresponding to positive, negative, and disorganized schizotypy.