Confirmatory Factor Analysis of SPQ & Chapman Schizotypy Scales

Confirmatory Factor Analysis of the 3 Factor Structure of the Schizotypal Personality Questionnaire and Chapman Schizotypy Scales

Authors:

Viviana Wuthrich[1] & Timothy C. Bates

Macquarie Centre for Cognitive Science,

Macquarie University,

Sydney, Australia

Abstract

The structure of schizotypal personality disorder (SPD) was examined using CFA of the Schizotypal Personality Questionnaire (SPQ: Raine, 1991), the Chapman Magical Ideation (Mag: Eckblad & Chapman, 1983), Perceptual Aberration (Per: Chapman, Edell, & Chapman, 1980), and Revised Social Anhedonia scales (Soc: Mishlove & Chapman, 1985). We tested the fit of competing two and three factor models in order to determine whether slight modifications would result in a better fit in normal samples, and also to assess the fit of these models according to stringent fit criteria (CFI>.095; RSMEA=.05-.08) across three experiments. In experiment 1 we found that the Raine et al. (1994) three factor: interpersonal, cognitive/perceptual, and disorganisation model fit the data best, however it did not provide an adequate fit (CFI=.936, RSMEA=.108). Instead, we found that a similar three-factor model with the additional loading of social anxiety onto the Cognitive/Perceptual factor, magical thinking onto the Interpersonal factor, and the correlation of ideas of reference with paranoia, produced a well fitting model (CFI=.978, RSEMA=.069). In experiment 2 the good fit of this modified three-factor model to SPQ scores was replicated in an independent sample, and further it was shown to fit better than a competing four-factor model. In experiment 3, this modified model was successfully extended to include the three Chapman scales. Together the three experiments indicate that the three-factor model of the SPQ, albeit with some slight modifications, is a good model for schizotypy structure that is not restricted to one measure of schizotypal personality.

Key words: SPQ, schizotypal personality disorder, confirmatory factor analysis, factor structure, perceptual aberration, magical ideation
Research suggests that Schizotypal Personality Disorder (SPD) and schizophrenia have a three-factor structure; a positive dimension (hallucinations, delusions), a negative dimension (restricted affect, social anhedonia, anxiety), and a third dimension of disorganisation (odd behaviour and speech) (Andreasen, Arndt, Alliger, Miller, & Flaum, 1995; Bergman, Silverman, Harvey, Smith, & Siever, 2000; Liddle, 1987; Raine et al., 1994; Reynolds, Raine, Mellingen, Venables, & Mednick, 2000; Vollema & Hoijtink, 2000). Although there is some support for a two-factor model of schizotypy (Kendler et al., 1991; Siever & Gunderson, 1983), and for alternate three-factor models which replace the disorganisation factor with a factor of disordered relationships (Strauss, Carpenter, & Bartko, 1974), or paranoia (Bergman et al., 1996), or oddness (Battaglia, Cavallini, Macciardi, & Bellodi, 1997), and recently for a four factor model (Stefanis et al., 2004), the Raine three-factor model has been the most consistently validated.

Support for the Raine three-factor structure has come both from exploratory factor analytic techniques and from confirmatory factor analyses (CFA). In CFA, models are specified a priori and competing models are compared according to their relative fit. As such previously proposed models can be precisely tested for generalisability to different samples and therefore are superior to exploratory techniques. Using CFA, support for the Raine three-factor structure of schizotypy has been found using scores from the Schizotypal Personality Questionnaire (SPQ: Raine, 1991), Chapman schizotypy scales (Chapman, Chapman, & Raulin, 1978; Eckblad & Chapman, 1983) and clinical interviews. Specifically, the Raine three-factor model was supported in samples of university students using SPQ scores (Raine et al., 1994), and using the Chapman scales scores (Suhr & Spitznagel, 2001), in high school students using SPQ scores (Fossati, Raine, Carretta, Leonardi, & Maffei, 2003), in community members using SPQ scores (Raine et al., 1994; Reynolds et al., 2000), in relatives of patients with schizophrenia using a structured clinical interview (Bergman et al., 2000) and in clinical patient groups using structured clinical interviews (Bergman et al., 1996) and also with SPQ scores (Rossi & Daneluzzo, 2002; Vollema & Hoijtink, 2000).

In CFA model fit is determined by applying a specific criterion to a nominated fit index. Problems with interpreting chi-square difference have resulted in a shift to reliance on fit indices to determine model acceptability. However, recent reviews of CFA techniques indicate that many of the fit indices have severe limitations and “rule of thumb” cut-offs previously used to determine model adequacy may be inappropriate (Hu & Bentler, 1999). Specifically, Hu and Bentler (1999) report that the commonly used Goodness of Fit index (GFI) and Adjusted GFI (AGFI) are inappropriate measures of model adequacy which inflate statistical error and that the Tucker-Lewis Index (TLI) is unduly affected by sample size. Instead, they recommend the application of the Comparative Fit Index (CFI) and the Root-Mean-Square Error of Approximation (RMSEA) with a 90% confidence interval to interpret CFA results. Hu and Bentler (1999) also report that the commonly applied cut-off values of .90 to fit indices often leads to inappropriate decisions regarding model acceptance. Therefore they recommend that a .95 cut-off value should be applied to CFI values, and an upper limit of .06 should be applied to RSMEA values to determine model acceptance. However, Browne and Cudeck (1993) suggest that for the RSMEA, values in the range .05-.08 indicate a fair fit, with values greater than .10 indicating poor fit. Finally the use of y suggest that the Akaike information criterion (AIC: Akaike, 1987) is recommended for comparisons between models with different numbers of factors. The AIC should be used to compare models with different numbers of factors as this criterion assesses parsimony with statistical goodness of fit and the number of estimated parameters in the model taken into account (Hu & Bentler, 1999). Smaller values represent better fit of the hypothesised model.

Given the recency of these recommendations, much of the CFA past research that provided support for the Raine three-factor model of schizotypy was based on fit indices (GFI, AGFI and TLI) and cut-off values (<.90) which are too lenient. As a result, the adequacy of the current three-factor structure needs to be confirmed against these new fit criteria.

Using these new guidelines, one previous report shows support for the Raine three-factor model of schizotypy. Bergman et al. (2000) reported the Raine three-factor model fitted schizotypal symptoms in the relatives of patients with schizophrenia adequately (CFI=.98, RMSEA=.048). These CFI and RMSEA values meet the new .95 and .06 cut-offs respectively and indicate a good fit. However, most research previously interpreted as showing support for the Raine three-factor model of schizotypy no longer meets the adequacy criteria. For example, Raine et al. (1994) reported that the three-factor model was superior to the paranoid three-factor model of Bergman et al. (1996), and to two-factor models (Kendler et al., 1991; Siever & Gunderson, 1983) in both an undergraduate (GFI=.96, TLI=.90) and community sample (GFI=.92, TLI=.94) based on SPQ scores. However, rigid application of the above recommendations indicates that the fit indices (GFI, TLI) and cut-off values used (>.90) in this study may now be considered inappropriate. This is also true for the Rossi and Daneluzzo (2002) study based on SPQ scores in a clinical sample, in which they tested the same models as Raine et al. (1994). Although they reported that the Raine three-factor model fit the data best (CFI=.93), it would no longer be considered to be a well fitting model. Finally, Reynolds et al. (2000) recently reported that the three-factor model explained SPQ scores regardless of culture, age, gender, religious affiliation, family adversity and psychopathology. However, application of a .95 cut-off value to the fit indices reported by these authors suggests the Raine three-factor model did not fit the data adequately for samples segregated for ethnicity, religion, and psychopathology. Therefore more research is needed to understand the three-factor structure of schizotypal personality using these more specific and demanding fit index criteria in CFA. An additional motivation of this paper was to examine whether alternative potential structures may provide better descriptions of SPQ variance within normal populations than are provided by existing models.

In this paper we report three experiments using CFA with a .95 cut-off criteria to CFI values and a .06 cut-off criteria to RMSEA values, as recommended by Hu and Bentler (1999), to interpret the goodness of fit of five competing models of schizotypy structure. The fit of two- and three-factor these models to was based on undergraduate student SPQ scores on the SPQ (Raine, 1991) scores and was tested in Experiment 1. two independent student samples (Experiments 1 and 2). The five specific models tested were the Siever and Gunderson (1983) and the Kendler et al. (1991) two-factor models, and the Raine et al. (1994) disorganisation, Bergman et al. (1996) paranoid, and the Battaglia et al. (1997) oddness three-factor models. These models are depicted in Figure 1 and described in Experiment 1. In thisese two experiments support was found for a slightly modified three-factor model. The validity of this model, and its superiority over a four-factor model was confirmed in Experiment 2 using an independent sample. In Experiment 3 the Magical Ideation (Mag: Eckblad & Chapman, 1983), Perceptual Aberration (Per: Chapman, Chapman, & Raulin, 1978), and Revised Social Anhedonia (Soc: Eckblad, Chapman, Chapman, & Mishlove, 1982) scales were successfully incorporated into this three-factor structure.

The two measures of SPD used in this paper are two of the most commonly used self-report scales. Most of the recent research on the factor structure of schizotypy has been based on the SPQ. Developed from the DSM-III-R (American Psychiatric Association, 1987) criteria for SPD, the SPQ directly assesses the presence of all nine symptoms of SPD: social anxiety, no close friends, constricted affect, paranoia, ideas of reference, magical thinking, unusual perceptual experiences, odd speech and odd behaviour. This scale has been shown to be a reliable and valid self-report measure of SPD, with 55% of individuals scoring in the top 10% of the normative sample meeting criteria for SPD (Raine, 1991). Research does support the validity of SPQ as a measure of schizotypal personality and vulnerability for schizophrenia, with higher schizotypy being associated with increased cognitive and biological dysfunction (Bates, 2005; Chen, Hsiao, & Lin, 1997; Kendler et al., 1991; Maier, Falkai, & Wagner, 2003).

The Chapman schizotypy scales are alternatives to the SPQ and are also frequently used in schizotypy research. These scales were developed prior to DSM-III-R and were based on the premise that schizophrenia is more than one disorder and that different symptom- features may indicate vulnerability for different types of psychoses (Chapman, Edell, & Chapman, 1980). As a result they developed a number of scales to measure psychosis proneness symptoms. However, research now suggests that the Mag (Eckblad & Chapman, 1983), Per (Chapman et al., 1978), and Soc (Eckblad et al., 1982) scales are the most relevant of their scales for the assessment of schizotypal personality as longitudinal research indicates that these three scales may predict development of psychosis (Chapman, Chapman, Kwapil, Eckblad, & Zinser, 1994; Kwapil, 1998). The Physical Anhedonia scale has also been used in schizotypy research as a measure of social adjustment (Freedman, Rock, Roberts, Cornblatt, & Erlenmeyer-Kimling, 1998), but scores are not elevated amongst offspring of schizophrenic individuals (Erlenmeyer-Kimling et al., 1993), further there is little evidence that Physical Anhedonia scores predict development of psychosis or schizophrenia (Erlenmeyer-Kimling et al., 1993).

While several factor analytic studies have been reported for the SPQ, few factor studies have been conducted on the Chapman scales. The results of a factor analysis by Kendler et al. (1991) indicated that the Per and Mag scales loaded onto a positive factor, while the Soc and Physical Anhedonia scale loaded onto a trait anhedonia/negative factor. Similarly, Suhr and Spitznagel (2001) conducted a factor analysis of the SPQ and Chapman Per and Mag scales together and found that that both the Per and Mag scales loaded onto the Cognitive/Perceptual factor, with the Per scale also loading onto the Disorganisation factor. Social Anhedonia was not studied in their paper. Therefore a third aim of this paper was to examine more clearly the relationship between the SPQ and the Per, Mag and Soc scales.

Next in experiment 1, we first compare existing models of the SPQ to current fit criteria and suggest a well-fitting three-factor model. This is replicated in experiment 2, and in experiment 3, the generality and comprehensiveness of this modified three-factor model is tested by extending it to the Chapman scales.

Experiment 1

Introduction

We examined the fit of five competing two and three-factor models of schizotypy factor structure, measured using CFA of SPQ subscales in a large undergraduate sample. These models are depicted in Figure 1. The first model we tested was the simple two latent variable model, proposed by Siever and Gunderson (1983), in which schizotypy consists of cognitive/perceptual and interpersonal factors. The second model was developed by Kendler et al. (1991) who assessed schizotypal features in a normal twin sample using a structured interview for SPD and found evidence for a similar two-factor solution. This model included factors of cognitive/perceptual and interpersonal behaviour but differed from the Siever and Gunderson model with odd behaviour loading onto the cognitive/perceptual factor, and paranoia and social anxiety loading onto both factors rather than just the interpersonal factor.

Three factor solutions of SPD and schizotypy have also been proposed and mirror the three factor solutions found for patients with schizophrenia (Liddle, 1987). These consist of a “positive” (cognitive/perceptual) dimension, a “negative” (interpersonal) dimension and a third dimension that represents either disorganisation (Raine et al., 1994) or paranoia (Bergman et al., 1996) or oddness (Battaglia et al., 1997). In this experiment we tested the ability of each of these two and three factor models to explain the variance in SPQ scores in our normal population.

(Insert Figure 1 about here)

Method

Participants

Participants were 558 (416 female, 142 male) first-year psychology students at Macquarie University, Sydney, Australia. Participants completed the SPQ (Raine, 1991) for course credit. Age range was 17 to 60 years (mean = 22.75, SD= 6.36).

Materials

The SPQ (Raine, 1991) is a 74 item scale which measures the nine diagnostic criteria stipulated by the DSM-III-R (American Psychiatric Association, 1987) for Schizotypal Personality Disorder (SPD): social anxiety, no close friends, constricted affect, paranoia, magical thinking, unusual perceptual experiences, ideas of reference, odd speech and odd behaviour. It has been shown to be a reliable and valid self-report measure of SPD at the high end of scores (Raine, 1991). However, based on dimensional perspectives of schizotypal vulnerability (Claridge & Beech, 1995), scores on the SPQ are also frequently used within normal populations to indicate increasing schizophrenia vulnerability or subclinical SPD features.