Bifactor structure of the Schizotypal Personality Questionnaire (SPQ)

Psychiatry Research 230 (2015) 940–950

Contents lists available at ScienceDirect

Psychiatry Research journal homepage: www.elsevier.com/locate/psychres

Bifactor structure of the Schizotypal Personality Questionnaire (SPQ) Antonio Preti a,b,c,n, Sara Siddi b,d, Marcello Vellante a, Rosanna Scanu b, Tamara Muratore b, Mersia Gabrielli b, Debora Tronci b, Carmelo Masala b, Donatella Rita Petretto b a

Center of Liaison Psychiatry and Psychosomatics, University Hospital, University of Cagliari, Cagliari, Italy Section on Clinical Psychology, Department of Education, Psychology, Philosophy, University of Cagliari, Cagliari, Italy c Genneruxi Medical Center, Cagliari, Italy d Unit of Research and development, Parc Sanitari Sant Joan de Déu, Sant Boi de Llobregat, Spain b

art ic l e i nf o

a b s t r a c t

Article history: Received 19 April 2015 Received in revised form 21 September 2015 Accepted 9 November 2015 Available online 11 November 2015

The schizotypal personality questionnaire (SPQ) is used to characterize schizotypy, a complex construct helpful for the investigation of schizophrenia-related psychopathology and putative endophenotypes. The SPQ factor structure at item level has been rarely replicated and no study had tested a bifactor model of the SPQ so far. The unidimensional, the correlated, the second-order and the bifactor models of the SPQ were tested to evaluate whether the items converge into a major single factor defining the schizotypy-proneness of the participants, to be used for grouping purpose. Parallel principal component analysis (PCA) and confirmatory factor analysis (CFA) were used to determine the optimal number of factors and components in a cross-sectional, survey design involving 649 college students (males: 47%). The first-order, nine-subscale model was confirmed by CFA in the whole sample. The best evidence from parallel PCA in the training set was in favor of a two-factor model; the bifactor implementation of this model showed good fit in the subsequent CFA. Two main dimensions of positive and negative symptoms underlie schizotypy in non-clinical samples, entailing specific risk of psychosis. On a measurement level, the study provided support for the use of the total scores of the SPQ to characterize schizotypy. & 2015 Elsevier Ireland Ltd. All rights reserved.

Keywords: Schizotypal Personality Questionnaire Confirmatory factor analysis Early Intervention Schizotypy Psychosis Schizophrenia

1. Introduction This study set out to investigate the factor structure of the Schizotypal Personality Questionnaire (SPQ) (Raine, 1991), a widely used measure of schizotypy. Schizotypy is conceived as a risk factor for schizophrenia, laying into a dynamic continuum from personality to psychosis (Ettinger et al., 2014; Barrantes-Vidal et al., 2015). The investigation of schizotypy is an important strategy to identify the genes potentially related to psychosis, and to study correlates of psychosis-proneness in the general population without the interference of medications and other confounding factors (e.g., the negative impact of institutionalization on cognition), which may bias the identification of the psychosis correlates. Moreover, the investigation of the schizotypy continuum may favor the identification of the mechanisms operating across different levels of severity along this continuum (BarrantesVidal et al., 2015). Better understanding of the factor structure of the tools that are used to identify people with schizotypy is mandatory for measurement purposes. n Corresponding author at: Centro Medico Genneruxi, Via Costantinopoli 42, 09129 Cagliari, Italy. E-mail address: [email protected] (A. Preti).

http://dx.doi.org/10.1016/j.psychres.2015.11.010 0165-1781/& 2015 Elsevier Ireland Ltd. All rights reserved.

Confirmatory factor analysis (CFA) was used to reproduce the factor structure of the SPQ, but models were tested at the subscale level rather than at the item level. The factor structure of the SPQ at the item level has been rarely considered, and no study had tested a bifactor model of the SPQ so far. Indeed, most studies were based on a “correlated traits” model, in which a construct domain is decomposed into separate, correlated elements. However, this is not a measurement model per se, since “there is no one common target dimension to be measured or that directly affects item variance” (Reise et al., 2010, p. 546). If the SPQ is intended to measure schizotypy as a general construct, a secondorder or a bifactor model is more appropriate than a simply “correlated traits” model. In a standard “correlated traits model”, the common variance on an item is “partitioned into a weighted function of variation on two or more correlated primary traits” (e.g. Siever and Gunderson, 1983, p. 547). Alternatively, a second-order model can be constructed, whereby independent dimensions are correlated because they share a common cause they converge to. This second-order model explains why two or more primary dimensions are correlated (Reise et al., 2010). In a bifactor model, the general and group factors are constrained to be orthogonal (Holzinger and Swineford, 1937; Schmid J, 1957). The general factor reflects the common elements among the items, and represents the individual

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

941

Fig. 1. Schematic representation of the four models — Model A, a strictly unidimensional model; Model B: a standard correlated traits model; Model C, a second-order model; Model D, a bifactor model. Grouping factors: Ud¼ unidimensional; 2nd Ord¼2nd order factor; Gen¼ general factor. Primary factors: Pos ¼ SPQ-Positive symptoms; Neg¼SPQ-Negative symptoms. Sub-domain: IF ¼ Ideas of reference; ESA¼ Excessive social anxiety; MT¼Odd beliefs or magical thinking: UPE¼ Unusual perceptual experiences; OB ¼Odd or eccentric behavior; NCF¼No close friend; OS ¼Odd speech; CA¼ Constricted affect; S¼ Suspiciousness.

differences on the target dimension (Reise et al., 2010; Reise, 2012). The group factors are a reflection of item response variance that is not accounted for by the general factor, and represent additional variance that is explained by sub-dimensions within the items. A bifactor model is alternative to the strictly unidimensional factor, in which all variation on the components is thought to be affected by variation in the target latent trait, and there is only one common source of variance (see Fig. 1). The relevance of these alternative models to the investigation of schizotypy as a risk factor for schizophrenia is a reflection of the conceptualization of schizotypy. 1.1. The concept of schizotypy The psychoanalyst Rado (Rado, 1953) used the term “schizotype” (from “schizophrenic genotype”) to describe individuals who, despite having no psychosis, displayed attenuated symptoms that were phenotypically similar to those observed in schizophrenia. Over time, the term has spread to indicate a schizophrenia-like pattern of beliefs and perceptual experiences observed in first-degree relatives of patients diagnosed with psychosis, and in people from the general population in the absence of psychosis (Tarbox and Pogue-Geile, 2011). The spectrum of traits related to schizotypy includes attenuated psychotic symptoms in the form of unusual subjective experiences and odd beliefs or magical thinking (Chapman, 1978; Chapman et al., 1984; Mason and Claridge, 2006); attenuated negative symptoms such as anhedonia, apathy and social withdrawal (Chapman et al., 1976); and more bizarre or disorganized behaviors expressed through

eccentricity, lack of spontaneity or impulsive nonconformity (Chapman et al., 1984; Mason and Claridge, 2006). Schizotypal traits encompass the recently emphasized ultra high-risk criteria for the detection of people at high risk of psychosis (Fusar-Poli et al., 2013; Yung and Nelson, 2013), but the two do not overlap completely. Emphasis on the early detection and intervention in psychosis in recent years has renewed interest in the assessment of vulnerability traits for psychosis, hence in the investigation of the factor structure and the correlates of schizotypy (Fonseca-Pedrero et al., 2008; Kwapil et al., 2008; Barrantes-Vidal et al., 2009, 2010). Indeed, schizotypy is a complex psychopathology construct, which can be helpful as an overarching framework for the investigation of schizophrenia-related psychopathology and putative endophenotypes (Cohen et al., 2015; Lenzenweger, 2015). 1.2. Framework of the current study The early intervention paradigm refocused the research into risk factors for schizophrenia on the stress-vulnerability model (Zubin and Spring, 1977; Birchwood and Macmillan, 1993; Debbane and Barrantes-Vidal, 2014). According to this model, an underlying genetic vulnerability to psychosis coupled with the impact of environmental stressors may trigger psychotic symptoms in at-risk people (Zubin and Spring, 1977; Birchwood and Macmillan, 1993). This vulnerability has not been identified precisely so far, but a family history of psychosis and schizotypy may represent specific risk factors for the triggering of psychosis (FusarPoli et al., 2013; Yung and Nelson, 2013). The measurement of

942

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

schizotypy is therefore important to identify people at risk of psychosis, and “early” enough to prevent conversion into psychosis of those presenting additional risk factors, such as: being in adolescence or young adulthood, using psychotizing substances (e.g., cannabis, amphetamine, hallucinogens), being exposed to social adversity during childhood or to social stress associated to immigration (Morgan et al., 2010; Large et al., 2011; Varese et al., 2012; Linscott and van Os, 2013; Sara et al., 2014). There is some evidence that early intervention in psychosis can reduce inpatient care and treatment dropout (Craig et al., 2004), and cut global costs of treatment (Valmaggia et al., 2012). Early detection and intervention in people at risk of developing psychosis were found to prevent or delay a first psychotic episode (van der Gaag et al., 2013). Schizotypy is conceived as multidimensional, but there is uncertainty about its exact factor structure. As a matter of fact, the determination of the factor structure of schizotypy has been defined so far on the basis of the measure of schizotypy used in the study. The SPQ is one of the most used measures of schizotypy (Raine, 1991; Fonseca-Pedrero et al., 2012). A large body of studies supported the validity of the SPQ and its association with clinical, functional and cognitive correlates of psychosis (Cohen et al., 2012). In the European Prediction of Psychosis Study (EPOS), SPQ subscales and items describing ideas of references and lack of close interpersonal relationships were found to predict increased transition to psychosis in a sample including 245 young help-seeking high-risk patients (Salokangas et al., 2013). The SPQ can thus be used to characterize schizotypy in help-seeking high-risk people. 1.3. The Schizotypal Personality Questionnaire The SPQ has been translated and validated in many countries (e.g. Reynolds et al., 2000), and there is evidence of measurement equivalence across cultures, supporting cross-cultural validity of SPQ scores (Fonseca-Pedrero et al., 2014a, 2014b). In its standard form, the SPQ is a 74-item self-report with a true/false format (Raine, 1991), which was developed to assess schizotypal personality disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Revised Third Edition ( American Psychiatric Association, 2000). The 74 items of the SPQ are grouped into nine subscales: three pertaining to attenuated positive symptoms (ideas of reference, odd beliefs or magical thinking, and unusual perceptual experiences); three pertaining to attenuated negative symptoms in the area of social anxiety and anhedonia mainly (excessive social anxiety, no close friends, constricted affect); three pertaining to paranoid ideation and eccentric behavior (odd or eccentric behavior, odd speech, suspiciousness). These nine subscales can be further grouped into second-order factors. There is a general agreement that the SPQ measures a multimensional construct, including a cognitive-perceptual domain (ideas of reference, odd beliefs or magical thinking, unusual perceptual experiences, and suspiciousness subscales); an interpersonal domain (excessive social anxiety, no close friends, constricted affect, and suspiciousness subscales); and a disorganized domain (odd or eccentric behavior and odd speech subscales) (Raine et al., 1994; Reynolds et al., 2000). However some authors proposed a two-factor second-order structure of the SPQ, with a positive symptoms dimension (ideas of reference, odd beliefs or magical thinking, unusual perceptual experiences, suspiciousness, odd or eccentric behavior and odd speech subscales) separated from a negative symptoms dimension (excessive social anxiety, no close friends, constricted affect subscales) (Siever and Gunderson, 1983). Other authors proposed a more complex, four-factor model consisting of a cognitive-perceptual domain (odd beliefs or magical thinking, unusual perceptual experiences subscales), a

paranoid thinking domain (ideas of reference, suspiciousness, excessive social anxiety subscales), a negative symptoms domain (suspiciousness, excessive social anxiety, no close friends, constricted affect subscales), and a disorganized domain (odd or eccentric behavior and odd speech subscales) (Stefanis et al., 2004; Compton et al., 2009). In past studies, the Stefanis et al. (Stefanis et al., 2004) fourfactor model of the SPQ received some support over the more traditional three-factor model (Raine et al., 1994), while evidence in favor of a two-factor study was provided in the past (Siever and Gunderson, 1983) and recently (Gross, 2014). More recently, in a study including 1445 young American participants (mean age 19.5 7 3.2), Gross (2014) found superiority of the four-factor model over the rivals using CFA, but PCA supported a two-factor model with a primarily positive factor and an interpersonal factor, similarly to the solution that was found by Siever and Gunderson (1983). 1.4. Confirmatory factor analysis of the bifactor model Testing a bifactor model of SPQ as an alternative to a strictly unidimensional model and to a second-order model, it is necessary to assure usability of the total score of the SPQ for a categorical scope. Indeed, the SPQ total score is often used to characterize schizotypy for research purposes (e.g. Spitznagel and Suhr, 2004; Hori et al., 2012; Bedwell et al., 2013). A bifactor model also explains how much of the variance can be attributed to item subdimensions, or is simply a reflection of a common factor to all items. Bifactor models were successfully used to test the existence of a general factor of psychopathology common to prevalent mental disorders in adults (Lahey et al., 2012). 1.5. Aims of the study In this study data from a non-clinical sample were used to apply CFA to the 74 items of the original, long form of the SPQ in order to determine whether these items converge into the nine subscales that are thought to define the basal factor structure of the tool. After demonstration of the validity of the first-order, ninesubscale structure of the SPQ, the sample was randomly split into a training set and a test set to explore the bifactor model and its alternatives. In the training set, parallel analysis was used to determine the optimal number of components. Thereafter, in the test set, the unidimensional model, the correlated model, the second-order and the bifactor models of the SPQ were tested to evaluate whether the items converge into a major single factor defining the schizotypy-proneness of the participants, to be used for grouping purposes. The unidimensional model was used to test the case of a strict continuum among the nine subscales defining schizotypy. The bifactor model was used to test the hypothesis of a shared variance attributable to an internal dimension of the sub-domains (i.e., genetic commonality between the different sub-domains). The second-order model was used to test the hypothesis of a shared variance attributable to a dimension external to the correlated sub-domains (e.g., level of distress raised by the schizotypal experiences). The correlated model tested independence of the correlated primary traits.

2. Methods This study is part of the Cagliari – Psychosis: Investigation on Risk Emergence (CAPIRE, which means “to understand” in Italian),

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

an investigation aimed at testing the reliability and validity of the screening tools developed to assess and diagnose the mental states at risk of psychosis. The institutional review board approved the study protocol in accordance with the guidelines of the 1995 Declaration of Helsinki (as revised in Tokyo in 2004). The study was carried out between winter 2011 and spring 2012. 2.1. Participants The sample included the participants to the first two waves of the CAPIRE study. The reference population studied was the young adults attending the Cagliari University (n ¼31,729), who were enrolled for the study. These undergraduate samples were enrolled via a snowball procedure (Vogt, 2005), a method that avoids the bias of self-selection that occurs when recruiters only tap their personal social network (Snijders, 1992). Further details on the CAPIRE study were published elsewhere (Preti et al., 2013, 2014). Participants were individually invited to fill in a booklet including socio-demographic information and the SPQ. Participants were allowed as much time as they wanted to answer the questionnaires; the time dedicated to answering the questionnaires was not recorded. Overall, 962 people were contacted: 120 declined after having had a look at the booklet; 842 people accepted to fill in the questionnaire; 689 participants actually returned the booklet; 40 cases were rejected because their questionnaires were left blank in some part; 649 participants were included in the study out of the 842 people who had accepted to participate (77%), and out of the 962 people who had been asked to take part in the study (67% overall participation rate). Participation was voluntary and no fee or other compensation was given for taking part in the study. All participants provided informed consent. The final sample included 305 males (47%) and 344 females (53%). Participants were 24 years old (SD ¼3.4) on average. In the sample 19 participants declared to be married (2.9%), and 325 reported to be in a stable relationship (50%). The participants whose parents had a high school diploma were 287 (44%), while the participants whose parents had a university degree or a higher qualification were 86 (13%). 2.2. Measures The validated Italian version of the 74-item SPQ was used for this study (Raine, 1991; Fossati et al., 2003). There is evidence that the SPQ possesses adequate psychometric properties (FonsecaPedrero et al., 2008, 2014a, 2014b). Cronbach's alpha for the individual subscales ranged from.71 to.78 (mean.74) in the original validation study (Raine, 1991). 2.3. Statistics No data were missing in the database; error rates were less than 1% and all were corrected based on the questionnaires. All data were coded and analyzed using the Statistical Package for Social Sciences (SPSS) version 20. Scale reliability was measured by Cronbach's alpha, to favor comparison with past studies. For group comparisons, values of.70 are considered satisfactory; values around.60 are considered acceptable when dealing with subscales derived from a single questionnaire (Nunnally, 1978). 2.3.1. Item-level confirmatory factorial analysis (CFA) The overall sample was used to test the first-order, nine-subscale structure of the SPQ. Item level CFA was carried out with the lavaan package (Rosseel, 2012) running in R version 3.1 (R Core

943

Team, 2013). The lavaan package was shown to generate the same results as other software packages (Narayanan, 2012). The items were treated as categorical variables and analyzed by polychoric correlations (Jöreskog, 1994). The mean- and variance-adjusted (diagonally) weighted least squares (WLSMV) estimator was used (Flora and Curran, 2004). The ratio of chi-square to the degrees of freedom (df) was calculated in addition to chi-square to evaluate model fitting, with ratios larger than 3 indicating poor fit (Byrne, 1989). Additional parameters for fit estimation were: the comparative fit index (CFI) and the root mean square error of approximation (RMSEA). RMSEA values of 0.08 or lower and CFI values of 0.90 or higher are considered acceptable (Browne and Cudeck, 1993; Hu and Bentler, 1999). 2.3.2. Principal component analysis at the subscale level After demonstrating the validity of the first-order, nine-subscale structure of the SPQ, the sample was randomly split into a training set and a test set to explore the bifactor model and its alternatives. In the training set, parallel analysis was used to determine the optimal number of components. In parallel analysis, the scree plot of the observed data was compared with that of a random matrix of the same size as the original. The best solution is based on the number of components with eigenvalues higher than those generated by the random data. Components are linear sums of the variables; factors are latent variables that may explain the correlations or co-variances between variables. Components are more easily interpretable than factors, and their models are less indeterminate (Steiger, 1990; Velicer and Jackson, 1990a; Velicer and Jackson, 1990b). Therefore, further analyses were based on the results of principal component analysis (PCA). The parallel analysis and the PCA were carried out with the psych package (Revelle, 2014) running in R 3.1 (R Core Team, 2013). 2.3.3. Confirmatory factorial analysis at the subscale level The model extracted by PCA from SPQ data in the training set was then applied to the test set, to further explore the structure of the SPQ. In the simplest, unidimensional model, the SPQ was assumed to measure a single dimension of schizotypy. In the second-order models, the sub-domains extracted by PCA were expected be correlated and the covariance among them was expected to be explained by a second-order factor (Fig. 1, model C). In the bifactor models, the general factor and the sub-domains were expected to be orthogonally independent, i.e. to be uncorrelated one to the other (Fig. 1, model D). In both the bifactor and the second-order models the error terms associated with each item were uncorrelated. The same model-without convergence into a general or second-order factor-was used to test the correlated model, which is also a test of reproducibility of the solution extracted by the PCA in the training set (see Fig. 1, model B). For sake of completeness, the two-factor model extracted from the parallel analysis was also compared with the standard threefactor model of Raine et al. (Raine et al., 1994) and with the fourfactor model described by Stefanis et al. (2004) and confirmed by Compton et al. (2009) and Fonseca-Pedrero et al. (2014a, 2014b) (see introduction for details on subscales composition by factor). For all (two-, three-, and four-factor) solutions we tested the correlated, the second-order and the bifactor implementation of the model. Maximum likelihood estimation was used to test CFA models. The ratio of chi-square to the degrees of freedom (df) was calculated in addition to chi-square to evaluate model fitting, with ratios larger than 3 indicating poor fit (Byrne, 1989; Hu and Bentler, 1999, p.2). Additional parameters for fit estimation were: the CFI,

944

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

the RMSEA, and the standardized root mean square residual (SRMR). RMSEA values of 0.08 or lower, SRMR of .09 or lower, and CFI values of 0.90 or higher are considered acceptable (Browne and Cudeck, 1993; Marsh, 2005). Model identification was verified according to the Bekker et al. (1994) method. Bekker et al. (1994) developed an algebraic method to check local identification based on the Wald Rank Rule, the rank of a Jacobian matrix. This algebraic method evaluates an augmented version of the Jacobian matrix, which is the matrix of first-order derivatives of the discrepancy function with respect to the free parameters. A necessary condition for the identification is that the Jacobian matrix has at least as many rows as columns. Thus, the rank of the Jacobian matrix is compared to the number of free parameters in the factorial model, i.e. the number of columns in the matrix. When the rank of the Jacobian matrix is different from the number of free parameters in the factorial model, the model is not identified. Calculation was based on a script available in lavaan version 0.5–16 or higher (see Appendix). Models were compared on the basis of goodness-of-fit indices, the Akaike Information Criterion (AIC) (Akaike, 1987) and the Bayesian Information Criterion (BIC) (Schwarz, 1978). Models with the lowest AIC and BIC should be preferred (Schermelleh-Engel et al., 2003; Wicherts and Dolan, 2004). For comparable models, we also calculated the Akaike weights, which can be interpreted as the probability that the given model is the best model (Burnham and Anderson, 2002; Wagenmakers and Farrell, 2004). Calculation was done with the library MuMIn (Barton, 2014) running in R (R Core Team, 2013). The best bifactor model was then compared with the corresponding second-order model with a Satorra-Bentler scaled chisquare difference test (Satorra, 2000). There is evidence that there is sufficient power to differentiate a bifactor model from the corresponding second-order model with a sample size slightly lower than 500 (Chen et al., 2010). The raw data used in the PCA/CFA analyses and the codes that were used for the analyses are provided in the Supplemental Material.

3. Results Summary statistics for the SPQ subscales in the sample are shown in Table 1.

for most scales, with the exception of the “odds beliefs or magical thinking” and the “odd or eccentric behaviors” subscales, for which the median was zero (see also density plot, Fig. A1 in the appendix). Participants were more likely to endorse one or more positive answers on the “suspiciousness” and the “excessive social anxiety” subscales than in the other ones (Table 1). Overall, frequency of item endorsing varied from 5% to 60% (Table 2). Item total correlation was higher than 0.30 in 58 items out of 74 (78%). Internal consistency was excellent (Cronbach's alpha ¼0.92), and no item impacted on Cronbach's alpha (deviations from the overall value were not higher than 0.001 point). In the sample, the ten percent high cutoff was 32, and 70 participants scored equal or above this cutoff on the SPQ. According to the original study (Raine, 1991), no more than a half of those scoring at the top 10 percent of SPQ would receive a diagnosis of schizotypal personality disorder. This leaves a conservative base-rate estimate for this disorder at 5.4% for this sample, i.e. close to the prevalence that Raine (1991) found in his sample of undergraduate students by applying the Structured Clinical Interview for DSM-III-R Personality Disorders (SCID-II; Spitzer et al. 1987). 3.2. First-order factor structure in the overall sample The first-order, nine-subscale model was confirmed by CFA: χ2 (with corrected robust estimation) was 3459.02 (df¼ 2591), po 0.0001; χ2/df ¼1.3; CFI¼0.937; RMSEA¼ 0.023 (95%CI: 0.021 to 0.025). Except for χ2, which may be unduly influenced by sample size, all other goodness-of-fit indicators were in the range of acceptability. The estimated reliability indexes of the subscales were excellent, with “constricted affect” showing the lowest values (but in the good range still). 3.3. Creation of the training set and of the test set The sample was randomly split into a training set (n ¼324) and a test set (n ¼325). The two sub-samples did not differ in terms of gender proportion (females: 54.3% versus 51.7%; χ2 ¼ 2.25, df ¼1, p¼ 0.13); age (24.5 73.6 vs 24.1 73.3; Welch two-sample t-test ¼1.49, df ¼340, p¼ 0.13); or socioeconomic status, as measured by the highest level of parental education (compulsory school: 43.5% vs 41.5%; χ2 ¼3.00, df ¼2, p¼ 0.22).

3.1. Descriptive statistics section 3.4. Parallel analysis and PCA of the SPQ in the training set Participants endorsed the complete interval of the potential scores in the subscales, with the exception of the “no close friends” and “constricted affect” subscales. Median was close to the mean

Parallel analysis suggested two components (Fig. A2 in Appendix).

Table 1 Descriptive statistics for the Schizotypal Personality Questionnaire subscales in the sample (n¼ 649). Subscale

n items

Mean

SD

Median

Interquartile range

Max–Min

% endorsing 0

Ideas of reference Excessive social anxiety Odd beliefs or magical thinking Unusual perceptual experiences Odd or eccentric behaviors No close friends Odd speech Constricted affect Suspiciousness Total score on the SPQ

9 8 7 9 7 9 9 8 8 74

2.2 2.6 1.1 1.6 1.1 1.2 2.6 1.6 2.4 16.4

2.2 2.2 1.5 1.7 1.7 1.5 2.4 1.5 2.1 11.2

2.0 2.0 0 1.0 0 1.0 2.0 1.0 2.0 14.0

4 3 2 2 2 2 3 2 3 15

0–9 0–8 0–7 0–9 0–7 0–8 0–9 0–7 0–8 0–58

30.4% 17.7% 51.6% 32.7% 52.2% 43.9% 22.5% 27.7% 19.1% 2.0%

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

Table 2 Descriptive statistics for the SPQ in the overall sample (n¼ 649) and factor loading in the first-order, nine-subscale model. Subscale

Itema Frequency ItemTotal

0.87 1. 10. 19. 28. 37. 45. 53. 60. 63.

30.4% 11.4% 28.7% 31.7% 19.9% 12.3% 15.9% 38.1% 32.4%

0.23 0.34 0.51 0.43 0.31 0.38 0.39 0.49 0.50

Table 2 (continued ) Subscale

Cronbachb Loading Subscales reliability Cronbach's alpha

Ideas of reference 0.922 0.921 0.920 0.920 0.921 0.921 0.921 0.920 0.920

0.410 0.628 0.818 0.699 0.557 0.701 0.719 0.835 0.875

Excessive social anxiety 22.5% 46.7% 48.8% 26.0% 28.5% 14.6% 60.9% 18.0%

0.41 0.34 0.35 0.40 0.42 0.41 0.28 0.41

0.921 0.921 0.921 0.921 0.921 0.921 0.922 0.921

8.8% 19.9% 27.3% 16.0% 17.6% 8.8% 9.2%

0.32 0.42 0.41 0.36 0.37 0.27 0.32

0.921 0.921 0.921 0.921 0.921 0.921 0.921

0.86

4. 13. 22. 31. 40. 48. 56. 61. 64.

45.1% 26.3% 6.2% 6.0% 5.2% 6.2% 21.0% 23.6% 17.9%

0.34 0.37 0.28 0.33 0.29 0.32 0.30 0.36 0.37

0.921 0.921 0.921 0.921 0.921 0.921 0.921 0.921 0.921

0.921 0.921

0.784 0.779 0.79

8. 17. 26. 35. 43. 51. 68. 73.

38.1% 28.0% 4.2% 7.9% 5.1% 14.3% 14.2% 47.5%

0.35 0.40 0.13 0.18 0.27 0.30 0.19 0.19

0.921 0.921 0.922 0.922 0.922 0.921 0.922 0.922

0.655 0.745 0.435 0.476 0.715 0.640 0.475 0.406

9. 18. 27. 36. 44. 52. 59. 65.

28.5% 18.6% 57.6% 29.0% 12.3% 59.3% 14.5% 19.3%

0.46 0.51 0.37 0.40 0.43 0.31 0.51 0.47

0920 0.920 0.921 0.921 0.921 0.921 0.920 0.920

0.741 0.867 0.607 0.642 0.766 0.485 0.910 0.760

b

0.88

Item number refers to the order in the SPQ. Cronbach's alpha if item was excluded.

Table 3 PCA of the SPQ: two-component solution, loading, eigenvalues and explained variance.

0.770 0.817 0.724 0.748 0.742 0.717 0.725

Unusual perceptual experiences

0.44 0.41

Constricted affect

a

0.88

3. 12. 21. 30. 39. 47. 55.

13.9% 13.3%

Cronbachb Loading Subscales reliability Cronbach's alpha

Suspiciousness

0.723 0.668 0.614 0.772 0.839 0.881 0.606 0.892

Odd beliefs or magical thinking

Itema Frequency ItemTotal

69. 72.

0.86 2. 11. 20. 29. 38. 46. 54. 71.

945

0.578 0.686 0.660 0.744 0.763 0.719 0.547 0.613 0.652

Subscale

Component 1

Component 2

Ideas of reference Excessive social anxiety Odd beliefs or magical thinking Unusual perceptual experiences Odd or eccentric behaviors No close friends Odd speech Constricted affect Suspiciousness Eigenvalues Explained variance

0.78 0.20 0.79 0.85 0.70  0.13 0.62  0.14 0.52 3.11 35%

0.00 0.59  0.17  0.18  0.04 0.86 0.17 0.92 0.36 2.10 23%

Items retained in the model are in bold. 0.93

Odd or eccentric behaviors 5 14. 23. 32. 67. 70. 74.

17.6% 31.1% 28.5% 9.1% 11.1% 12.8% 5.2%

0.35 0.49 0.50 0.39 0.37 0.32 0.32

0.921 0.920 0.920 0.921 0.921 0.921 0.921

0.811 0.844 0.898 0.959 0.785 0.791 0.809

6. 15. 24. 33. 41. 49. 57. 62. 66.

11.6% 23.6% 12.8% 9.6% 16.6% 18.0% 15.6% 6.3% 5.9%

0.19 0.37 0.31 0.28 0.15 0.16 0.36 0.15 0.21

0.922 0.921 0.921 0.921 0.922 0.922 0.921 0.922 0.922

0.490 0.745 0.764 0.685 0.356 0.339 0.864 0.451 0.576

No close friends

0.82

Odd speech

3.5. Confirmatory bifactor model of the SPQ in the test set

0.90 7. 16. 25. 34. 42. 50. 58.

27.9% 49.0% 52.5% 34.8% 17.7% 25.4% 21.7%

0.51 0.38 0.37 0.39 0.45 0.45 0.46

0.920 0.921 0.921 0.921 0.920 0.920 0.920

0.812 0.669 0.606 0.667 0.754 0.736 0.779

The two-component solution distinguished an attenuated positive symptoms dimension from an attenuated negative symptoms dimension, with suspiciousness loading on the first component mostly (Table 3 and Figs. A3, A4 and A5 in the appendix). The promax rotation provided a more clear separation of the factors than the varimax rotation. Indeed, in the promax rotation, the “suspiciousness” subscale had a greater loading in the “positive symptoms” factors than in the varimax solution, where the loading on the suspiciousness subscale was more distributed between the two factors (see Fig. A6 in the appendix).

The strictly unidimensional model did not reach adequacy on the basis of the goodness-of-fit indices. Similarly, neither the twofactor correlated model nor the second-order implementation of the two-factor model reached the threshold for fit. The two-factor bifactor model reached adequacy according to goodness-of-fit indices (Table 4). As far as the three-factor and the four-factor models were concerned, fit was reached for the correlated model and for the

946

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

second-order model; the corresponding bifactor models were not identified. The correlated three-factor model, the correlated fourfactor model and the bifactor implementation of the two-factor model had quite similar AIC and BIC, with some advantage for the correlated four-factor model (Fig. 2). According to the Akaike weight, the four-factor model had the best fit in the correlated and the second-order implementations of the model (Fig. 2). Among the bifactor implementations of the models, the twofactor model had the lowest AIC and BIC of all models, with a 93% probability of being the best bifactor model according to the Akaike weight. Comparison between the two-factor bifactor model and the corresponding second-order model confirmed superiority of the bifactor model: Satorra-Bentler's chi-square difference¼89.28, df¼ 7, p o.0001. In the two-factor bifactor model, factor loading was acceptable for all items in the general factor. However, the general factor did not explain all variance on some of the subscales. The “ideas of references” and the “unusual perceptual experiences” subscales loaded equally on the general factor and the “positive symptoms” sub-domain, while the “odd beliefs or magical thinking” subscale loaded more on this sub-domain than on the general factor. Regarding the “negative symptoms” sub-domain, the “constricted affect” subscale loaded equally on it and on the general factor, while the “no close friends” subscale loaded more on the “negative symptoms” sub-domain than on the general factor (Table 5). The same comparison is reported in Table A2 of Appendix for the different implementations of the four-factor model. Due to cross-loading, the contribution of some of the subscales to the variance is poorly interpretable. We repeated the analysis in the whole sample (n ¼649), and the results did not change: the four-factor model had the best fit as far as the correlated and the second-order implementations of the model were concerned; the corresponding bifactor model was not identified, as was not the bifactor implementation of the threefactor model. The bifactor implementation of the two-factor model had the best fit of all models (Table A1 in Appendix).

4. Discussion This study found that the Stefanis et al. (2004) four-factor model had the best fit as far as the correlated and the secondorder implementations of the model are concerned. However, the best evidence from the parallel PCA in the training set was in favor of the two-factor model. The bifactor implementation of the twofactor model proved good fit in the subsequent CFA in the test set.

The “odd beliefs or magical thinking” and the “no close friends” subscales loaded more on the sub-domains than on the general factor, suggesting that they measure sub-domains that run independently from what is measured by the general factor. 4.1. Comparison with past studies In past studies, the Stefanis et al. (Stefanis et al., 2004) fourfactor model of the SPQ received some support over the more traditional three-factor model (e.g., Compton et al., 2009; FonsecaPedrero et al., 2014a, 2014b). In this study, too, we found that Stefanis et al. (2004) four-factor model of the SPQ is superior to alternative models when tested in its correlated or second-order implementation. The main objection to using this model, in both its basic version (“correlated trait” model) and the second-order implementation, is the presence of cross-loading, which hampers the interpretation of the scores. In particular, the “excessive social anxiety” and the “suspiciousness” subscales load on two factors, and this makes it quite impossible to decide what is the reason for high or low scores on one or both of these factors. On a clinical ground, it is plausible that paranoid thinking may favor both the development of positive symptoms (magical thinking, perceptual abnormalities) and of interpersonal difficulties, inducing introvertive anhedonia as a defense against perceived hostility from others. However, on a measurement level, a tool that uses the same sub-domain to measure two different and clinically separated dimensions is problematic. It is difficult to clearly attribute variance to one dimension or another. The bifactor model (as one of two equally supported models) may be preferred, since it is easier to interpret with respect to subscale allocation (no cross-loadings) and allows for those diagnostic purposes (calculation of a total score) for which the SPQ is frequently used. Evidence in favor of a two-factor study was provided in the past (Siever and Gunderson, 1983) and recently (Gross, 2014). In particular, the results of this study are congruent with the finding obtained by Gross et al. (Gross, 2014) in a sample including 1445 young participants (mean age 19.5 73.2) and evaluated with both the SPQ and the Wisconsin Schizotypy Scales (WSS). The WSS has a consistent two-factor structure, with evidence in favor of a positive and a negative dimension underlying its four scales (Browne and Cudeck, 1993; Kwapil et al., 2008; Gross, 2014). As far as the SPQ is concerned, Gross et al. (2014), too, found superiority of the four-factor model over its rivals using CFA, but PCA supported a two-factor model with a primarily positive factor and an interpersonal factor, with the same distribution of the loading by subscales as we found.

Table 4 Confirmatory factor analysis of SPQ in the test set: Goodness of fit indexes for the proposed models. Goodness of fit indicators 2

Wald Rank Rule 2

Model

χ

df

χ /df

CFI

RMSEA (90%CI)

SRMR

n columns

Rank

Unidimensional model Two-factor models Correlated traits Second-order model Bifactor model Raine's Three-factor Correlated traits Second-order model Bifactor model Stefanis's Four-factor Correlated traits Second-order model Bifactor model Threshold for good fit

218.92, p o 0.0001

27

8.1

0.815

0.148

(0.130–0.166)

.085

18

18

Identified

143.28, p o 0.0001 143.28, p o 0.0001 54.00, p o 0.0001

26 25 18

5.5 5.7 3.0

0.887 0.886 0.965

0.118 0.121 0.078

(0.099–0.137) (0.102–0.140) (0.055–0.103)

.074 .074 .032

19 20 27

19 19 27

Identified Not identified Identified

83.61, p o 0.0001 83.61, p o 0.0001 56.85, p o.0001

23 23 17

3.6 3.6 3.3

0.942 0.942 0.962

0.090 0.090 0.085

(0.070–0.111) (0.070–0.111) (0.061–0.110)

0.052 0.052 .043

22 22 28

22 22 27

Identified Identified Not identified

42.24, po 0.0001 72.46, p o 0.0001 67.59, p o0.0001 p 40.05

19 21 16

2.2 3.4 4.2 r3

0.974 0.950 0.950 Z .90

.066 .087 .100 o.08

(0.042–0.091) (.065–.109) (.076–.125) o .08

.033 .050 .049 o .09

26 24 29 n col. ¼

26 24 27 Rank

Identified Identified Not identified

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

947

Fig. 2. Distribution of the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC) and the Akaike weight by model. Models with the lowest AIC and BIC and the highest Akaike weight should be preferred.

Gross et al. (2014) did not test a bifactor model, but simply tested the correlated model. We think that the superiority of the bifactor implementation of the two-factor model over other models is a helpful contribution to the investigation of schizotypy. The two-factor model is more clear-cut in defining the role of the different sub-domains, since it has no cross-loadings. Since the SPQ is used for measurement purposes and classification, a model that avoids cross-loadings could be more helpful than a model that makes it difficult to separate different dimensions. 4.2. Implications for research The bifactor implementation of the two-factor model helps defining the variance attributable to the general factor underlying

schizotypy as a vulnerability factor from the variance attributable to sub-domains that may have an independent role in the risk of psychosis. This is a very important focus of research for the early detection of individuals at risk for psychosis and to understand the etiological risk factors for psychosis and its related conditions (Ettinger, et al., 2014; Lenzenweger, 2015). In particular, we found that some of the subscales loaded equally (“ideas of references”, “unusual perceptual experiences”, “constricted affect”) or even more (“odd beliefs or magical thinking”, “no close friends”) on the corresponding sub-domains than on the general factor. We may advance that delusion-proneness and social aloofness/anhedonia represent specific dimensions that may add to the risk of psychosis beside the schizotypy-specific vulnerability. Indeed, if the general factor has to be considered as

Table 5 Confirmatory factor analysis of SPQ in the test set (n¼325): loadings and estimated Cronbach's alpha for the bifactor and alternative models. Unidimensional model Subscale

Ideas of reference Excessive social anxiety Odd beliefs or magical thinking Unusual perceptual experiences Odd or eccentric behaviors No close friends Odd speech Constricted affect Suspiciousness

Estimated Cronbach's alpha

Correlated two-factor model

Second-order model

Positive symptoms

Negative symptoms

Second-order factor

Positive symptoms

0.44 0.57

0.77

0.57

Bifactor model

Negative symptoms

General factor

Positive symptoms

0.56 0.63

0.52

0.57

0.73 0.54

0.77

0.65

0.69

0.39

0.69

0.38

0.70

0.69

0.71

0.40

0.71

0.47

0.58

0.58

0.59

0.34

0.59

0.50

0.28

0.46 0.69 0.48 0.71

0.84

0.72 0.66 0.72 0.69

0.84

0.67

0.72 0.38 0.72 0.39 Load on second-order factor 0.79

0.72 0.67 0.72 0.69 0.57

1.00

0.84

0.70

0.51 0.70 0.52 0.70

0.84

Negative symptoms

0.10

0.68 0.18 0.44 0.21

0.83

0.67

N.B. Regarding the loadings of the subscales, the software computed them directly for the unidimensional, the correlated traits and the bifactor models. As for the secondorder model, each subscale's loading on the second-order factor was calculated by the product of its loading on the first factor and the first-order factor's loading on the second-order factor (Brown, 2006, p. 335). For example, loading of Ideas of reference (IR) on the second-order factor is the product of the loading of IR on the first factor it belongs to (Positive symptoms) and the loading of the Positive symptoms factor on the second-order factor: 0.77  0.57 = 0.44.

948

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

expressing the schizotypy proneness of the subject, the two subdomains may represent additional risk of psychosis beside the one entailed by the variance related to the schizotypy-specific vulnerability. There is some evidence that anhedonia may have a discrete taxonomic nature in the general population (Gross, 2014) and that it clusters separately from other factors accounting for positive schizotypy (Cella et al., 2013). In particular, there is evidence that social anhedonia is a reflection of an anomaly in social and affective reward processes that are specific to schizophrenia and distinguishable from other “negative” affect based on social anxiety or depression (Cohen et al., 2015). Delusion-proneness as well may be schizotypy-related or independent from it: it can be influenced by mood variation in people at risk of affective disorders, or by cognitive faults in the subjects having risk factors for schizophrenia (Bora et al., 2009).

this study, did not take into account the information of SPQ at item level. However, we feel that on a measurement level, the study provided some support for the use of the total scores of the SPQ to characterize both clinical and non-clinical samples as far as schizotypy is concerned.

4.3. Limitations and strengths of the study

Acknowledgments

This study has some limitations that should be taken into account. As in most past studies higher-order models were tested at the subscale level, but item-level factor analysis might produce different results (Chmielewski and Watson, 2008). In an exploratory factor analysis carried out in 1123 college students, Fonseca-Pedrero et al. (2014a, 2014b) found that the SPQ items were grouped in a theoretical structure of seven second-order factors. Models were tested on college students. Students are generally in an age range when the risk of developing psychosis is at its highest, and they may be more forthcoming in providing answers to socially undesirable topics, such as symptoms of psychopathology (Lincoln and Keller, 2008), although they might be not representative of the general population. Moreover, there is no certainty that the factor structure identified in non-clinical samples is generalizable to clinical samples. In particular, healthy participants, such as college students, are less likely to exhibit the disorganized behavior expected to be measured by the SPQ. Another limitation is the lack of a third, independent sample to be used for testing the generalizability of the best solutions extracted by the CFA in the test set (i.e., the Stefanis et al. (2004) four-factor model of the SPQ and the bifactor implementation of the two-factor solution extracted by the PCA in the training set). These limitations are balanced by some strengths of this study. Item-level CFA has confirmed the first-order, nine-subscale model of the SPQ, which had never occurred in past studies. The subsequent evaluation of the models at the subscale level was carried out in independent samples, by distinguishing a training set from a test set. Past studies often relied on a single sample to evaluate their model. Models were compared not merely on the basis of goodness-of-fit indicators, that may be misleading (SchermellehEngel et al., 2003; Wagenmakers and Farrell, 2004), but also by using information criteria and by establishing whether the model was identified or not.

The CAPIRE study had the following coordinators, scientific consultants and collaborators: A. Preti, M.D.; A. Raballo, M.D., Ph. D.; D.R. Petretto, Psy.D.; C. Masala, M.D.; M.G. Carta, M.D.; M. Vellante, Psy.D.; S. Siddi, Psy.D.; M.T. Cascio, Psy.D.; I. Corrias, Psy. D.; M. Gabbrielli, Psy.D.; V. Lai, Psy.D.; T. Muratore, Psy.D.; E. Pintus, Psy.D.; M. Pintus, Psy.D.; S. Sanna, Psy.D.; R. Scanu, Psy.D.; D. Tronci, Psy.D. Dr S. Siddi is the recipient of a Grant from the Regione Sardegna, Italy (Grant n. PRRMAB-A2011-19251). Dr M.T. Cascio is the recipient of a Grant of the Regione Sardegna, Italy (Grant n. T2-MAB-A2008-138).

5. Conclusions In conclusion the study found support for a bifactor implementation of the two-factor model of the SPQ. On a clinical ground the study confirms that two main dimensions of positive symptoms and negative symptoms underlie schizotypy in nonclinical samples. However some of the sub-dimensions measured by the SPQ, such as delusion-proneness and social aloofness/anhedonia, may represent specific dimensions that may add to the risk of psychosis beside the schizotypy-specific vulnerability. It should be noted that the bifactor model, as it was implemented in

Financial support The study was supported by internal funds.

Declaration of interest None.

Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.psychres.2015.11. 010.

References Akaike, H., 1987. Factor Analysis and AIC. Psychometrika 52, 317–332. American Psychiatric Association, 2000. Diagnostic and statistical manual of mental disorders (4th ed., text rev.). Washington, DC: Author. Barrantes-Vidal, N., Grant, P., Kwapil, T.R., 2015. The role of schizotypy in the study of the etiology of schizophrenia spectrum disorders. Schizophr. Bull. 41 (Suppl 2), S408–S416. Barrantes-Vidal, N., Lewandowski, K.E., Kwapil, T.R., 2010. Psychopathology, social adjustment and personality correlates of schizotypy clusters in a large nonclinical sample. Schizophr. Res. 122, 219–225. Barrantes-Vidal, N., Ros-Morente, A., Kwapil, T.R., 2009. An examination of neuroticism as a moderating factor in the association of positive and negative schizotypy with psychopathology in a nonclinical sample. Schizophr. Res. 115, 303–309. Barton, K., 2014. MuMIn: Multi-model inference. R package version 1.10.0. 〈https:// cran.r-project.org/web/packages/MuMIn/index.html〉. Bedwell, J.S., Chan, C.C., Trachik, B.J., Rassovsky, Y., 2013. Changes in the visualevoked P1 potential as a function of schizotypy and background color in healthy young adults. J. Psychiatr. Res. 47, 542–547. Bekker, P.A., Merckens, A., Wansbeek, T.J., 1994. Identification, equivalent models, and computer algebra. Academic Press, San Diego. Birchwood, M., Macmillan, F., 1993. Early intervention in schizophrenia. Aust. New. Zealand J. Psychiatry 27, 374–378. Bora, E., Yucel, M., Pantelis, C., 2009. Cognitive impairment in affective psychoses: a Meta-analysis. Schizophr. Bull. 36, 112–125. Brown, T.A., 2006. Confirmatory Factor Analysis for Applied Research. Guilford Press, New York. Browne, M.W., Cudeck, R., 1993. Alternative ways of assessing model fit. In: Bollen, K.A., Long, J.S. (Eds.), Testing Structural Equation Models. Sage, Newbury Park, CA, pp. 136–161. Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A Practical Information-theoretic Approach. Springer, New York. Byrne, B.M., 1989. A primer of LISREL: basic applications and programming for confirmatory factor analytic models. Springer-Verlag, New York.

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

Cella, M., Serra, M., Lai, A., Mason, O.J., Sisti, D., Rocchi, M.B.L., Preti, A., Petretto, D.R., 2013. Schizotypal traits in adolescents: Links to family history of psychosis and psychological distress. Eur. Psychiatry 28, 247–253. Chapman, L.J., Chapman, J.P., Numbers, J.S., Edell, W.S., Carpenter, B.N., Beckfield, D., 1984. Impulsive nonconformity as a trait contributing to the prediction of psychotic-like and schizotypal symptoms. J. Nerv. Ment. Dis. 172, 681–691. Chapman, L.J., Chapman, J.P., Raulin, M.L., 1976. Scales for physical and social anhedonia. J. Abnorm. Psychol. 85, 374–382. Chapman, M.L., 1978. Body-image aberration and social anhedonia. J. Abnorm. Psychol. 87, 399–407. Chen, F.F., West, S.G., Sousa, K.H., 2010. A Comparison of Bifactor and Second-Order Models of Quality of Life. Multivar. Behav. Res. 41, 189–225. Chmielewski, M., Watson, D., 2008. The heterogeneous structure of schizotypal personality disorder: item-level factors of the Schizotypal Personality Questionnaire and their associations with obsessive-compulsive disorder symptoms, dissociative tendencies, and normal personality. J. Abnorm. Psychol. 117, 364–376. Cohen, A.S., Callaway, D.A., Najolia, G.M., Larsen, J.T., Strauss, G.P., 2012. On “risk” and reward: investigating state anhedonia in psychometrically defined schizotypy and schizophrenia. J. Abnorm. Psychol. 121, 407–415. Cohen, A.S., Mohr, C., Ettinger, U., Chan, R.C., Park, S., 2015. Schizotypy as an organizing framework for social and affective sciences. Schizophr. Bull. 41 (Suppl 2), S427–S435. Compton, M.T., Goulding, S.M., Bakeman, R., McClure-Tone, E.B., 2009. Confirmation of a four-factor structure of the Schizotypal Personality Questionnaire among undergraduate students. Schizophr. Res. 111, 46–52. Craig, T.K., Garety, P., Power, P., Rahaman, N., Colbert, S., Fornells-Ambrojo, M., Dunn, G., 2004. The Lambeth Early Onset (LEO) Team: randomised controlled trial of the effectiveness of specialised care for early psychosis. Br. Med. J. 329 (7474), 1067–1070. Debbane, M., Barrantes-Vidal, N., 2014. Schizotypy From a Developmental Perspective. Schizophr. Bull. 41, S386–S395. Ettinger, U., Meyhöfer, I., Steffens, M., Wagner, M., Koutsouleris, N., 2014. Genetics, cognition, and neurobiology of schizotypal personality: a review of the overlap with schizophrenia. Front. Psychiatry 5, 18. Flora, D.B., Curran, P.J., 2004. An Empirical evaluation of alternative methods of estimation for confirmatory factor analysis with ordinal data. Psychol. Methods 9, 466–491. Fonseca-Pedrero, E., Lemos-Giráldez, S., Paino, M., Santarén-Rosell, M., Sierra-Baigrie, S., Muñiz, J., 2012. Reliability and Sources of Validity Evidence of the Oviedo Schizotypy Assessment Questionnaire-Abbreviated (ESQUIZO-Q-A). Span. J. Psychol. 15, 840–849. Fonseca-Pedrero, E., Paíno, M., Lemos-Giráldez, S., García-Cueto, E., CampilloÁlvarez, Á., Villazón-García, Ú., Muñiz, J., Oviedo, U. De, 2008. Schizotypy assessment: State of the art and future prospects. Int. J. Clin. Health Psychol. 8, 577–593. Fonseca-Pedrero, E., Fumero, A., Paino, M., de Miguel, A., Ortuño-Sierra, J., LemosGiráldez, S., Muñiz, J., 2014a. Schizotypal personality questionnaire: new sources of validity evidence in college students. Psychiatry Res. 219, 214–220. Fonseca-Pedrero, E., Compton, M.T., Tone, E.B., Ortuño-Sierra, J., Paino, M., Fumero, A., Lemos-Giráldez, S., 2014b. Cross-cultural invariance of the factor structure of the Schizotypal Personality Questionnaire across Spanish and American college students. Psychiatry Res. 2014. http://dx.doi.org/10.1016/j.psychres.2014.06.050, Aug 13. pii: S0165-1781(14)00572-1. Fossati, A., Raine, A., Carretta, I., Leonardi, B., Maffei, C., 2003. The three-factor model of schizotypal personality: Invariance across age and gender. Pers. Individ. Differ. 35, 1007–1019. Fusar-Poli, P., Borgwardt, S., Bechdolf, A., Addington, J., Riecher-Rössler, A., SchultzeLutter, F., Keshavan, M., Wood, S., Ruhrmann, S., Seidman, L.J., Valmaggia, L., Cannon, T., Velthorst, E., De Haan, L., Cornblatt, B., Bonoldi, I., Birchwood, M., McGlashan, T., Carpenter, W., McGorry, P., Klosterkötter, J., McGuire, P., Yung, A., 2013. The psychosis high-risk state: a comprehensive state-of-the-art review. JAMA Psychiatry 70, 107–120. Gross, T.R., 2014. Comparing the factor structure of the Wisconsin Schizotypy Scales and the Schizotypal Personality Questionnaire. Pers. Disord. 5, 397–405. Holzinger, K.J., Swineford, F., 1937. The Bi-factor method. Psychometrika 2, 41–54. Hori, H., Teraishi, T., Sasayama, D., Matsuo, J., Kawamoto, Y., Kinoshita, Y., Kunugi, H., 2012. Relationships between season of birth, schizotypy, temperament, character and neurocognition in a non-clinical population. Psychiatry Res. 195, 69–75. Hu, L., Bentler, P.M., 1999. Cutoff criteria for fit indices in covariance structure analysis: Conventional criteria versus new alternatives. Struct. Equ. Model. 6, 1–55. Jöreskog, K.G., 1994. On the estimation of polychoric correlations and their asymptotic covariance matrix. Psychometrika 59, 381–389. Kwapil, T.R., Barrantes-Vidal, N., Silvia, P.J., 2008. The dimensional structure of the Wisconsin Schizotypy Scales: factor identification and construct validity. Schizophr. Bull. 34, 444–457. Lahey, B.B., Applegate, B., Hakes, J.K., Zald, D.H., Hariri, A.R., Rathouz, P.J., 2012. Is there a general factor of prevalent psychopathology during adulthood? J. Abnorm. Psychol. 121, 971–977. Large, M., Sharma, S., Compton, M.T., Slade, T., Nielssen, O., 2011. Cannabis use and earlier onset of psychosis: a systematic meta-analysis. Arch. Gen. Psychiatry 68, 555–561. Lenzenweger, M.F., 2015. Thinking clearly about schizotypy: hewing to the schizophrenia liability core, considering interesting tangents, and avoiding

949

conceptual quicksand. Schizophr. Bull. 41 (Suppl 2), S483–S491. Lincoln, T.M., Keller, E., 2008. Delusions and hallucinations in students compared to the general population. Psychol. Psychother. 81, 231–235. Linscott, R.J., van Os, J., 2013. An updated and conservative systematic review and meta-analysis of epidemiological evidence on psychotic experiences in children and adults: on the pathway from proneness to persistence to dimensional expression across mental disorders. Psychol. Med. 43, 1133–1149. Marsh, H.W., H.K., G.D., 2005. Goodness of fit evaluation in structural equation modeling. In: Maydeu-Olivares, A., McCardle, J. (Eds.), Psychometrics: A Festschrift to Roderick P. McDonald. Lawrence Erlbaum Associates, Inc, Mahwah, NJ, pp. 275–340. Mason, O., Claridge, G., 2006. The Oxford-Liverpool Inventory of Feelings and Experiences (O-LIFE): further description and extended norms. Schizophr. Res. 82, 203–211. Morgan, C., Charalambides, M., Hutchinson, G., Murray, R.M., 2010. Migration, ethnicity, and psychosis: toward a sociodevelopmental model. Schizophr. Bull. 36, 655–664. Narayanan, A., 2012. A Review of Eight Software Packages for Structural Equation Modeling. Am. Stat. 66, 129–138. Nunnally, J.C., 1978. Psychometric Theory, 2nd. Preti, A., Sisti, D., Rocchi, M.B.L., Siddi, S., Cella, M., Masala, C., Petretto, D.R., Carta, M.G., 2014. Prevalence and dimensionality of hallucination-like experiences in young adults. Compr. Psychiatry 55, 826–836. Preti, A., Vellante, M., Gabbrielli, M., Lai, V., Muratore, T., Pintus, E., Pintus, M., Sanna, S., Scanu, R., Tronci, D., Corrias, I., Petretto, D.R., Carta, M.G., 2013. Confirmatory factor analysis and measurement invariance by gender, age and levels of psychological distress of the short TEMPS-A. J. Affect. Disord. 151, 995–1002. R Core Team, 2013. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.Rproject.org/. Rado, S., 1953. Dynamics and classification of disordered behavior. Am. J. Psychiatry 110, 406–416. Raine, A., 1991. The SPQ: a scale for the assessment of schizotypal personality based on DSM-III-R criteria. Schizophr. Bull. 17, 555–564. Raine, A., Reynolds, C., Lencz, T., Scerbo, A., Triphon, N., Kim, D., 1994. Cognitiveperceptual, interpersonal, and disorganized features of schizotypal personality. Schizophr. Bull. 20, 191–201. Reise, S.P., 2012. The Rediscovery of Bifactor Measurement Models. Multivar. Behav. Res. 47, 667–696. Reise, S.P., Moore, T.M., Haviland, M.G., 2010. Bifactor models and rotations: exploring the extent to which multidimensional data yield univocal scale scores. J. Pers. Assess. 92, 544–559. Revelle, W., 2014. Procedures for Personality and Psychological Research. Northwestern University, Evanston, Illinois, USA. Reynolds, C.A., Raine, A., Mellingen, K., Venables, P.H., Mednick, S.A., 2000. Threefactor model of schizotypal personality: invariance across culture, gender, religious affiliation, family adversity, and psychopathology. Schizophr. Bull. 26, 603–618. Rosseel, Y.L., 2012. An R Package for Structural Equation Modeling. J. Stat. Softw. 48, 1–36. Salokangas, R.K.R., Dingemans, P., Heinimaa, M., Svirskis, T., Luutonen, S., Hietala, J., Ruhrmann, S., Juckel, G., Graf von Reventlow, H., Linszen, D., Birchwood, M., Patterson, P., Schultze-Lutter, F., Klosterkötter, J., 2013. Prediction of psychosis in clinical high-risk patients by the Schizotypal Personality Questionnaire. Results of the EPOS project. Eur. Psychiatry 28, 469–475. Sara, G.E., Large, M.M., Matheson, S.L., Burgess, P.M., Malhi, G.S., Whiteford, H.A., Hall, W.D., 2014. Stimulant use disorders in people with psychosis: a metaanalysis of rate and factors affecting variation. Aust. New Zealand J. Psychiatry 49, 106–117. Satorra, A., 2000. Scaled and adjusted restricted tests in multi-sample analysis of moment structures. In: Heijmans, D.D.H., Pollock, D.S.G., Satorra, A. (Eds.), Innovations in Multivariate Statistical Analysis: A Festschrift for Heinz Neudecker. Kluwer Academic, Dordrecht, pp. 233–247. Schermelleh-Engel, K., Moosbrugger, H., Müller, H., 2003. Evaluating the Fit of Structural Equation Models: tests of Significance and Descriptive Goodness-ofFit Measures. Methods Psychol. Res. Online 8, 23–74. Schmid J, L.J., 1957. The development of hierarchical factor solutions. Psychometrika 22, 53–61. Schwarz, G., 1978. Estimating the Dimension of a Model. Ann. Stat. 6, 461–464. Siever, L.J., Gunderson, J.G., 1983. The search for a schizotypal personality: historical origins and current status. Compr. Psychiatry 24, 199–212. Snijders, T., 1992. Estimation on the basis of snowball samples: how to weight. Bulletin of. Sociol. Methodol. 36, 59–70. Spitzer, R.L., Williams, J.B.W., Gibbon, M., 1987. Structured Clinical Interview for DSM-III-R Personality Disorders: SC1d-U. New York State Psychiatric Institute, New York. Spitznagel, M.B., Suhr, J.A., 2004. Neuropsychological impairment associated with symptoms of schizotypy: role of depressive and paranoid symptoms. J. Nerv. Ment. Dis. 192, 382–384. Stefanis, N.C., Smyrnis, N., Avramopoulos, D., Evdokimidis, I., Ntzoufras, I., Stefanis, C.N., 2004. Factorial composition of self-rated schizotypal traits among young males undergoing military training. Schizophr. Bull. 30, 335–350. Steiger, J.H., 1990. Some Additional Thoughts on Components, Factors, and Factor Indeterminancy. Multivar. Behav. Res. 25, 41–45. Tarbox, S.I., Pogue-Geile, M.F., 2011. Clinical Psychology Review A multivariate perspective on schizotypy and familial association with schizophrenia: A

950

A. Preti et al. / Psychiatry Research 230 (2015) 940–950

review. Clin. Psychol. Rev. 31, 1169–1182. Valmaggia, L.R., McGuire, P.K., Fusar-Poli, P., Howes, O., McCrone, P., 2012. Economic impact of early detection and early intervention of psychosis. Curr. Pharm. Des. 18, 592–595. van der Gaag, M., Smit, F., Bechdolf, A., French, P., Linszen, D.H., Yung, A.R., McGorry, P., Cuijpers, P., 2013. Preventing a first episode of psychosis: meta-analysis of randomized controlled prevention trials of 12 month and longer-term followups. Schizophr. Res. 149, 56–62. Varese, F., Smeets, F., Drukker, M., Lieverse, R., Lataster, T., Viechtbauer, W., Read, J., van Os, J., Bentall, R.P., 2012. Childhood adversities increase the risk of psychosis: a meta-analysis of patient-control, prospective- and cross-sectional cohort studies. Schizophr. Bull. 38, 661–671. Velicer, W.F., Jackson, D.N., 1990a. Component Analysis versus Common Factor Analysis: Some issues in Selecting an Appropriate Procedure. Multivar. Behav.

Res. 25, 1–28. Velicer, W.F., Jackson, D.N., 1990b. Component analysis versus common factor analysis: Some further observations. Multivar. Behav. Res. 25, 97–114. Vogt, W.P., 2005. Dictionary of Statistics and Methodology: A Nontechnical Guide for the Social Sciences. Sage, London. Wagenmakers, E.-J., Farrell, S., 2004. AIC model selection using Akaike weights. Psychon. Bull. Rev. 11, 192–196. Wicherts, J.M., Dolan, C.V., 2004. A Cautionary Note on the Use of Information Fit Indexes in Covariance Structure Modeling With Means. Struct. Equ. Model.: Multidiscip. J. 11, 45–50. Yung, A.R., Nelson, B., 2013. The ultra-high risk concept-a review. Can. J. Psychiatry 58, 5–12. Zubin, J., Spring, B., 1977. Vulnerability–a new view of schizophrenia. J. Abnorm. Psychol. 86, 103–126.