TOWARD A MORE PSYCHOMETRICALLY SOUND BRIEF MEASURE OF SCHIZOTYPAL TRAITS: INTRODUCING THE SPQ-BRIEF REVISED

Transcription

1 Journal of Personality Disorders, 24(4), , The Guilford Press TOWARD A MORE PSYCHOMETRICALLY SOUND BRIEF MEASURE OF SCHIZOTYPAL TRAITS: INTRODUCING THE SPQ-BRIEF REVISED Alex S. Cohen, PhD, Russell A. Matthews, PhD, Gina M. Najolia, MA, and Laura A. Brown, MA Psychometric case identification of individuals at risk for developing schizophrenia-spectrum disorders is an important enterprise. Unfortunately, current instruments for this purpose are limited. The present studies sought to improve the Schizotypal Personality Questionnaire- Brief (SPQ-Brief), a measure of schizotypal traits that has come under recent criticism. In the first study, we conducted exploratory factor analysis on the SPQ-Brief using a Likert-style scoring format in a large sample of nonclinical adults. Although acceptable internal consistency was found, the original factor structure was not supported. In the second study, we administered the full version of the SPQ to a separate large nonclinical adult sample and employed exploratory and confirmatory factor analysis to identify critical items that could be used to revise the SPQ-Brief. The end product of these studies is the SPQ-Brief Revised, which offers a subordinate seven-factor and super-ordinate three or four factor solution, employs a Likert-scale format to improve sensitivity, and is brief (34 items). Psychometric detection of individuals at risk for developing schizophreniaspectrum disorders is a critical enterprise for understanding the schizophrenia prodrome and for clarifying potentiators that influence illness trajectory. A number of measures have been advanced for this purpose based on self-report assessment of subclinical schizophrenia symptomatology. This psychometric-identification method is advantageous over other risk-identification procedures (e.g., Genetic identification; Addington, 2004; Tsuang, Stone, Tarbox, & Faraone, 2002) in that it can be applied to large samples in a cost-effective manner. There is considerable variability across measures in length (e.g., Chapman Schizotypy Scales, N of items = 166; Chapman, Chapman, & Kwapil, 1995; Oxford-Liverpool Inventory of Feelings and Experiences [O-LIFE], N of items = 159, Mason, Claridge, & Jackson, 1995; Schizotypal Personality Questionnaire [SPQ], N of items = 74 items, Raine, 1991) and trait coverage (e.g., Chapman Schizotypy From Louisiana State University. Address correspondence to Alex Cohen, Louisiana State University, 236 Audubon Hall, Baton Rouge, LA 70808; acohen@lsu.edu 516

2 THE SPQ-BRIEF REVISED 517 scales cover four domains, O-life covers four-five domains, SPQ covers nine domains). A researcher must balance brevity with breadth when selecting a screening measure. An abbreviated version of a popular instrument, the Schizotypal Personality Questionnaire (SPQ-Brief), was developed that covers three important schizotypal trait domains (Raine & Benishay, 1995). Although a solid literature employing this instrument has been established (e.g., Bedwell, Kamath, & Baksh, 2006; Jahshan & Sergi, 2007; Koo et al., 2006), its psychometric adequacy has come into question (Axelrod, Grilo, Sanislow, & McGlashan, 2001; Aycicegi, Dinn, & Harris, 2005; Compton, Chien, & Bollini, 2007; Mata, Mataix-Cols, & Peralta, 2005). The present project reports data from two studies of the SPQ. In the first study, we examine the psychometric properties of the SPQ- Brief in a large sample of nonclinical young adults using a more sensitive response (i.e., Likert) format. In the second study, we develop an improved version of this brief measure, the SPQ-Brief Revised (SPQ-BR) based on full SPQ data administered to a separate sample of nonclinical adults. The SPQ is an important instrument for schizotypy research. The full version is comprised of 74 true-false items and covers nine distinct domains which closely mirror the diagnostic criteria for Schizotypal Personality Disorder included in the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV; American Psychiatric Association, 1994). The full SPQ captures a broader range of schizotypal traits than other instruments and has shown adequate reliability and convergent validity (Raine, 1991; Reynolds, Raine, Mellingen, Venables, & Mednick, 2000). Factor analysis of the full SPQ supports the breadth of coverage, although there is disagreement over whether a three-factor structure (covering Cognitive-Perceptual, Disorganized, and Interpersonal factors; Reynolds et al., 2000; Wuthrich & Bates, 2006), a four factor model (covering Cognitive-Perceptual, Paranoid, Disorganized, and Interpersonal factors, Compton, Goulding, Bakeman, & McClure-Tone, 2009) or even larger factor structures (e.g., covering Social Anhedonia, Social Anxiety, Eccentric/Odd Behavior, Mistrust, and Unusual Beliefs, and Experiences; Chmielewski & Watson, 2008) are most appropriate. The factor structure seems to depend, in large part, on whether the analysis is conducted at the subscale level or the item level (Chmielewski & Watson, 2008). However, as noted by Raine and Benishay (1995), the length of the SPQ has made it less than ideal for some screening situations. The SPQ-Brief (Raine & Benishay, 1995), developed as a more concise measure of schizotypy, comprises 22 true-false items from the full SPQ covering three domains Cognitive-Perceptual, Disorganized, and Interpersonal. The selection strategy for this measure was based on identifying full SPQ items that were most highly correlated with their respective factors. This approach is consistent with one of the most common methodologies for item selection in scale reduction selecting items that maximize internal consistency, i.e., selecting items with high inter-item correlations (Clark & Watson, 1995; Stanton, Sinar, Balzer, & Smither, 2002). How-

3 518 COHEN ever, as discussed by these and other authors (Boyle, 1991; Kline, 1986), using an internal consistency maximizing approach can have some drawbacks. First, this approach can lead to item redundancy (Stanton et al., 2002). After all, the most internally consistent measure would likely occur when items are simply paraphrases of each other. Second, this approach can lead to factor structure limitations, such that, for example, only positively worded items are retained in the abbreviated measure (Miller & Cleary, 1993). Finally, focusing on maximizing internal consistency can lead to reduced validity (Kline, 1986). Reductions in validity can occur because when trimming items to increase internal consistency, the researcher may inadvertently trim items that serve to represent an increased portion of the theoretical construct space (Smith & Stanton, 1998). Thus, the scale may have less construct validity. Although the SPQ-Brief has been employed in a number of recent studies (e.g., Bedwell et al., 2006; Jahshan & Sergi, 2007; Koo et al., 2006), enthusiasm for this instrument is tempered for several reasons. First, multiple studies have reported a high level of intercorrelation between items from different subscales (Axelrod et al., 2001; Aycicegi et al., 2005; Compton et al., 2007; Mata et al., 2005), and less than ideal internal consistency (<.70) for the Cognitive-Perceptual and Disorganized subscales in three studies (Aycicegi et al., 2005; Compton et al., 2007; Mata et al., 2005). Second, factor analyses of the SPQ-Brief have provided only limited support for a three-factor solution. For example, Compton et al. (2007) reported that a three-factor solution offered little improvement over a single-factor model while Aycicegi et al. (2005) found that a two-factor solution best fit the data. Similarly, considerable item cross loading between the Interpersonal and Cognitive-Perceptual factors has been reported (Axelrod et al., 2001; Mata et al., 2005). Third, the SPQ brief offers limited coverage compared to the full SPQ because it covers only Cognitive-Perceptual, Disorganization, and Interpersonal domains, and not individual subscales. This is important as one considers recent evidence suggesting that the Interpersonal subscale reflects two distinct constructs social anxiety and social anhedonia. These domains appear to reflect very different underlying pathological processes in individuals with schizophrenia and schizotypy (Brown, Silvia, Myin-Germeys, Lewandowski, & Kwapil, 2008; Chmielewski & Watson, 2008; Kirkpatrick, Buchanan, Ross, & Carpenter, 2001; Lewandowski et al., 2006). A final consideration with the full SPQ involves the scoring format which employs a forced-choice true-false decision. Forcing participants to make an all or none evaluation prevents participants from taking into account degrees of trait severity, which is an issue because it means the SPQ may be insensitive to subtle trait manifestations (Peltier & Walsh, 1990; Wuthrich & Bates, 2004). Moreover, some have suggested that individuals may be reluctant to endorse seemingly severe symptoms of mental illness (Piltier & Walsh, 1990). To address these concerns, Wuthrich and Bates (2005) adapted the response format to include a five-point Likert scale and

4 THE SPQ-BRIEF REVISED 519 found improved reliability/sensitivity for the Likert format over the traditional format. This format has not been applied to the SPQ-Brief to our knowledge, but is particularly promising for offering expanded scoring range and improved sensitivity for its limited number of items. The present study employed exploratory and confirmatory factor analytic procedures to examine and create the SPQ-BR. In the first study, the psychometric properties of a Likert-type SPQ-Brief were examined in a large nonclinical adult sample. Given that the psychometrics from this first study were less than ideal, we sought to develop a more valid and conceptually consistent version of the SPQ-Brief by identifying critical items from the full SPQ, using a Likert-type response format administered to a separate large nonclinical adult sample. A primary difference between this study and the original SPQ-Brief development study is that our procedures are based on factor analytic procedures that control for inter-item correlation. Finally, we examined the convergence between SPQ-BR scores and quality of life with the expectation that increasing trait severity would be associated with poorer quality of life (as seen in Cohen & Davis, 2009). In sum, we aimed to develop a revised SPQ-Brief that would (1) employ a more sensitive response format than the original SPQ-brief while providing adequate internal consistency, (2) offer expanded trait coverage beyond the three broad Interpersonal, Cognitive-Perceptual, and Disorganization factor domains while minimizing factor cross-loading, (3) yield a higher order four-factor solution, characterized by Social Anxiety, Social Anhedonia, Disorganization, and Cognitive-Perceptual dimensions, and (4) demonstrate convergent validity with a measure of subjective and objective quality of life. STUDY 1 METHODS: PSYCHOMETRICS OF THE BRIEF SPQ Participants Participants were undergraduate students enrolled at Louisiana State University. Freshmen and sophomore students (N = 8,591) were approached by to participate in an on-line survey, and offered a chance to win monetary compensation as part of a lottery (10 prizes of $25us). Embedded within this survey were a consent form, basic demographic questions, Infrequency Scale items (Chapman & Chapman, 1983) and a modified SPQ. The response rate was approximately 17% (n = 1,507). Of these responses, 9% (n = 138) of the questionnaires was discarded because of incomplete responses (n = 128) or questionable validity (n = 10, as detailed below). The final dataset included 1,356 participants. Demographic and descriptive variables are included in Table 1. This study was approved by the LSU Human Subject Review Board and all participants offered informed consent prior to completing the surveys.

5 520 COHEN TABLE 1. Means and Standard Deviations for the Demographic and Descriptive Statistics Study 1 Study 2 Demographic Variables (n = 1356) (n = 1395) Age ± ± 2.03 % Female 64% 65% Ethnicity % Caucasian 83.9% 81.7% % African American 8.4% 8.4% % Asian-American 2.9% 3.1% % Hispanic 2.7% 3.2% % Other 2.0% 3.6% Schizotypal Personality Trait Scores Ideas of Reference.22 ± ± Social Anxiety.90 ± ± Magical Thinking 1.88 ± ± Unusual Perceptions.85 ± ± Eccentric Behavior 1.77 ± ± No Close Friends 1.22 ± ± Odd Speech 1.99 ± ± Constricted Affect.60 ± ± Suspiciousness 1.61 ± ± Response Frequency Strongly Disagree 24.12% 25.34% Disagree 32.71% 33.77% Neutral 17.48% 16.10% Agree 20.41% 20.15% Strongly Agree 5.27% 4.64% Notes. 1 Scores range from 4 to4; 2 Scores range from 6 to6; 3 Scores range from 8 to8; 4 Scores range from 18 to 18; 5 Scores range from 16 to 16; 6 Scores range from 14 to 14. Schizotypal Traits Schizoptyal traits were assessed using the SPQ-Brief (Raine & Benishay, 1995), a 22-item measure. We adopted a five-point Likert scale system that has been employed in recent research using the full SPQ (Wuthrich & Bates, 2005). Participants response options ranged from Strongly Disagree to Neutral to Strongly Agree. The Likert scale version of the full SPQ has shown high convergence and improved internal reliability (α =.95) when administered in either computer or standard paper and pencil versions, compared to the original version. For each of the SPQ measures included in this study, increasing scores reflect more severe traits. Infrequency Scale To screen out responders who provided random or grossly invalid responses, we included three questions from the Infrequency Scale (Chapman & Chapman, 1983). This abbreviated version, used in our prior research (e.g., Cohen & Davis, 2009), was employed to reduce the overall burden of test administration. This includes items such as I find that I often walk with a limp, which is the result of a skydiving accident. Individuals who endorsed two or more infrequency items were excluded from this

6 THE SPQ-BRIEF REVISED 521 study (66% of total items), a more stringent threshold than the traditional three of 13 items used in prior research (e.g., Gooding, Tallent, & Matts, 2005). Analyses Analyses were conducted in three steps. First, we examined descriptive statistics and ranges of SPQ scores. Second, we examined the internal consistency in SPQ factors scores using the factor structure reported in Raine and Benishay (1995). Third, principal components factor analysis with varimax rotation was conducted to examine the structure of the SPQ- Brief using the 22 items from the original measure, with Likert-type response options. It is important to note that prior studies have reported differences between males and females in SPQ scores (Raine, 1992). RESULTS Descriptive statistics and ranges for the SPQ are provided in Table 1. There was a notable range of responses suggesting that participants, by and large, made use of the full spectrum of options. Interestingly, only a minority (approximately 5%) of responses reflected strong endorsement of schizotypal traits (i.e., strongly agree). However, a large number of responses (approximately 20%) reflected more modest endorsement (i.e., agree). Internal consistency was good for each of the subscales (i.e., Chronbach s α for the Cognitive-Perceptual factor =.79, for the Interpersonal factor =.86, and for the Disorganization factor =.83), although internal consistency was highest for the total score (Chronbach s α=.90). Females reported significantly less severe traits within each of the factors, all t s(1354) > 3.41, p s <.001, although the effect sizes for each of these differences were generally small (<.32; using Cohen s 1988 criteria). Scores from the individual subscales were each significantly positively correlated with each other, range of r(1,354) values =.21.64). Given both three and four factor models have been argued in the literature, using Kaiser s criterion, all factors with eigenvalue over one were retained; in our initial examination we did not constrain our exploratory factor analysis in terms of factors retained. A four-factor solution was extracted with a cumulative variance of 56.21% explained. However, review of the scree plot did not suggest that a scree started at the fourth or fifth factor raising questions about the stability of the four-factor solution. Factor loadings and explained variances for each factor in the four-factor solution are reported in Table 2. Currently there does not exist a commonly agreed upon standard for interpreting factor loadings. As noted by Stevens (2002), in samples over 1,000 factor loadings greater than.16 are statistically meaningful. Furthermore, Comrey and Lee (1992) note that although factor loadings at or above.32 are poor in terms of overlapping variance with the factor, they are still potentially meaningful depending on

7 522 TABLE 2. Principal Component Analysis of SPQ-brief with Likert-Type Response Options (n = 1356) Factor Raine & Benishay Axelrod et al. Factor Factor I feel I have to be on my guard even with friends. Interpersonal Interpersonal.78 Have you found that it is best not to let other people know too much about you? Interpersonal Interpersonal.72 Do you feel that you are unable to get close to people? Interpersonal Interpersonal Do you often have to keep an eye out to stop people from taking advantage of you? Cognitive-Perceptual Interpersonal I tend to keep my feelings to myself. Interpersonal Interpersonal Do you often pick up hidden threats or put-downs from what people say or do? Cognitive-Perceptual Interpersonal Some people find me a bit vague and elusive during a conversation. Disorganized Disorganized Some people think that I am a very bizarre person. Disorganized Disorganized.82 I am an odd, unusual person. Disorganized Disorganized.81 People sometimes comment on my unusual mannerisms and habits. Disorganized Disorganized.72 I sometimes use words in unusual ways. Disorganized Cognitive-Perceptual People sometimes find me aloof and distant. Interpersonal Disorganized I feel very uneasy talking to people I do not know well. Interpersonal Interpersonal.85 I feel very uncomfortable in social situations involving unfamiliar people. Interpersonal Interpersonal.83 I tend to keep in the background on social occasions. Interpersonal Interpersonal.76 I find it hard to communicate clearly what I want to say to people. Disorganized Interpersonal.52 Have you ever noticed a common event or object that seemed to be a special sign for you? Cognitive-Perceptual Cognitive-Perceptual.72 Are you sometimes sure that other people can tell what you are thinking? Cognitive-Perceptual Cognitive-Perceptual.61 Have you ever had the sense that some person or force is around you, even though you cannot see anyone? Cognitive-Perceptual Cognitive-Perceptual.60 Do you ever suddenly feel distracted by distant sounds that you are not normally aware of? Cognitive-Perceptual Cognitive-Perceptual.57 When shopping do you get the feeling that other people are taking notice of you? Cognitive-Perceptual Cognitive-Perceptual Have you had experiences with astrology, seeing the future, UFOs, ESP, or a sixth sense? Cognitive-Perceptual Cognitive-Perceptual.40 Eigen Value Variance Explained 14.57% 14.34% 14.26% 13.01% Note. The conceptual factor for each item as defined orginally by Raines and Benishay as well as by Axelrod et al., are reported for comparision purposes (results for Axelrod et al. are based on factor loadings reported in Table 2, p. 175). Factor loadings less than.35 are not reported.

8 THE SPQ-BRIEF REVISED 523 the research question. For the purpose of this study, factor loadings greater than.40 are considered strong, and we consider items with loadings of.30 or more on two or more factors as cross-loading items. In sum, only limited support for Raine and Benishay s (1995) SPQ-brief measure or for Axelrod et al. s (2001) empirically-derived version of the SPQ-Brief measure was found. Of note, the Interpersonal dimension appears to split into two sub-factors. Additionally, eight of the 22 items (36%) demonstrate cross-loadings, which calls into question whether these items tap into unique underlying constructs, or some other higher order construct. As a follow-up analysis, given that Axelrod et al. (2001) in their analysis forced a three factor solution, we re-ran our exploratory factor analysis, constraining the model to three factors. In this model, the cumulative variance explained was 51.36%. Again though, review of the scree plot did not support this model, and interpretation of the factor loadings was even more complicated, with 12 of the 22 items (54%) demonstrating crossloadings (the full results of these analysis, including factor loadings, can be obtained from the first author). DISCUSSION In contrast to prior studies (Aycicegi et al., 2005; Compton et al., 2007; Mata et al., 2005), adequate internal consistency of the traditional threefactor solution was found. However, the factor structure identified by Raine and Benishay (1995) or Axelrod et al. (2001) was not maintained. Rather, a four-factor solution marked by two separate Interpersonal factors was observed. It is interesting to interpret this factor structure in light of recent evidence (from the full SPQ) that the Interpersonal factor defined in Raine and Benishay (1995) actually reflects distinct social anxiety and social anhedonia (e.g., Constricted Affect, No Close Friends) constructs. Evidence from studies employing instruments other than the SPQ provide further evidence that social anhedonia and social anxiety are separable constructs (Brown et al., 2008; Chmielewski & Watson, 2008; Lewandowski et al., 2006). However, the Interpersonal factors identified in this first study were not neatly comprised of Social Anxiety versus Constricted Affect/No Close Friends factors, as there was considerable cross-contamination from Suspiciousness, Ideas of reference, and Odd speech items. Thus, it seems the original SPQ-Brief Interpersonal factor reflects a heterogeneous mix of traits. Although this lack of replication could reflect our use of a modified response format, the fact that the factor structure has not been found in most prior studies argues against this concern. Regarding the expanded format, there were several interesting findings. Although approximately a quarter of items were endorsed at the Agree level or above, only a small subset of these were in the Strongly Agree range. This supports the notion that the dichotomous response format offers insufficient sensitivity, whereas the expanded Likert format offers

9 524 COHEN greater ability to discriminate. In the next study, we employed factor analysis procedures from the full SPQ, modified using the Likert-like response format, to derive the SPQ-BR. STUDY 2 METHODS Participants Participants were undergraduate students enrolled at Louisiana State University. Freshmen and sophomore students (N = 8,993) were approached via to participate in an on-line survey, and offered a chance to win monetary compensation (10 prizes of $25 U.S.) as part of a lottery. Embedded within this survey were a consent form, basic demographic questions, Infrequency Scale Items (Chapman & Chapman, 1983), and the full SPQ. The response rate was approximately 20% (n = 1,775). Of these responses, 21% ( n = 380) of the questionnaires was discarded due to elevated infrequency scores (n = five, using the two-item criterion detailed above) or incomplete responses (n = 375). It is noteworthy that there were over twice as many invalid questionnaires in this study employing the full SPQ version compared to the first study which used the SPQ- Brief. This underscores the importance of a psychometrically sound brief schizotypy measure to diminish effects of participant fatigue or boredom. The final dataset included 1,395 participants. Demographic and descriptive variables are included in Table 1. This study was approved by the LSU Human Subject Review Board and all participants provided informed consent prior to completing the surveys. Schizotypal Traits The full SPQ (Raine, 1991) was used, which includes 74 statements measuring Ideas of Reference, Excessive Social Anxiety, Odd Beliefs or Magical Thinking, Unusual Perceptual Experiences, Odd or Eccentric Behavior, No Close Friends, Odd Speech, Constricted Affect, and Suspiciousness. We adopted the same five-point Likert scale system used in the first study, ranging from Strongly Disagree to Neutral to Strongly Agree (Wuthrich & Bates, 2005). Quality of Life A modified version of Lehman s Quality of Life Brief Interview (Lehman, 1995), covering the prior month, was employed here. Presented in this study is a summary score reflecting 23 objective quality of life items covering seven domains (home, daily activities, family, social, financial, health, and legal concerns). These items cover objectively anchored assessments

10 THE SPQ-BRIEF REVISED 525 of behavior (e.g., How often do you make plans ahead of time to do something with a friend?) and access to resources (e.g., In the past month, did you have enough money for fun activities?). A separate summary score reflecting nine items assessing subjective satisfaction across these seven domains (using a standardized 7-point Likert scale) is also derived from the measure. Although these measures are typically used in patient studies (e.g., Bellack, Bennett, Gearon, Brown, & Yang, 2006), prior lab research supports convergence with the SPQ-Brief in university samples (Cohen & Davis, 2009). Increasing scores reflect higher functioning. Analyses A multi-step approach was taken to create the SPQ-BR. First, the sample was randomly split into three equal sub-samples. These sub-samples were mutually exclusive and exhaustive, and each sub-sample was used for one of the three analysis steps. In Step 1, a principal components factoring analysis with varimax rotation was conducted to examine the structure of the SPQ using the full 74 items from the original measure, with Likert-type response options. Based on the results of these analyses, poorly functioning items; (a) low factor loadings (loadings less than.40), (b) split factor loadings (loadings greater than.35 on two or more factors), (c) factors consisting of only one item, or (d) poor conceptual fit with other items loading on the factor; were excluded. In relation to this fourth criterion, if, for example, for a particular factor four items demonstrated a clear conceptual relationship (e.g., all were related to magical thinking) and a fifth, unrelated item (e.g., an item about constricted affect) also loaded on that factor, the unrelated item was removed. In Step 2, the factor structure for the resulting subset of items was cross-validated with the second sub-sample using a principal components factoring analysis with varimax rotation. In Step 3, confirmatory factor analysis was used to further cross-validate the SPQ-BR using the third sub-sample of respondents. As part of Step 3 we sought to confirm four models including (1) a uni-dimensional solution, (2) an uncorrelated factor structure, (3) Raine and Benishay s (1995) conceptual three-factor model, and (4) a four-factor solution similar to Raine and Benishay s (1995) model but with Constricted Affect/No Close Friends and Social Anxiety as distinct factors. In Step 4, we examined the convergent validity of the SPQ- BR by computing correlations with QOL scores, with the expectation that they would be inversely and significantly correlated. RESULTS Descriptive statistics and ranges for the SPQ are provided in Table 1. Internal consistency was good for each the full SPQ subscales: Ideas of Reference (α =.82), Social Anxiety (α =.87), Magical Thinking (α =.80), Unusual

11 526 COHEN Perceptions (α =.81), Eccentric Behavior (α =.90), No Close Friends (α =.85), Odd Speech (α =.86), Constricted Affect (α =.81) and Suspiciousness (α =.86). The distribution of responses was almost identical to that seen in Study 1. As in the prior study, females reported significantly less severe traits for most of the subscales all t s(1354) > 2.16, p s <.05, except for Ideas of Reference, Magical Thinking, and Suspiciousness subscales, which were statistical trends, all t s(1354) > 1.79, p s <.07. The effect sizes for each were small (Cohen s d s <.33). Scores from the individual subscales were each significantly positively correlated with each other, range of r(1354) values =.19.80). Step 1: Initial Exploratory Factor Analysis Using the first subsample of respondents (n = 465) an initial fourteenfactor solution was extracted using the full 74 items. The cumulative variance explained with this fourteen-factor model was 63.34%. Review of the scree plot suggested that a more parsimonious solution might be found. All items were reviewed for low factor loadings (loadings less than.40), split factor loadings (loadings greater than.35 on two or more factors), factors consisting of only one item, or poor conceptual fit with other items loading on the factor. Based on these criteria, 42 items were excluded (approximately 65% of these excluded items were excluded for demonstrating split loadings). Step 2: Exploratory Factor Analysis Cross-Validation Using the second subsample of respondents (n = 465) a second exploratory factor analysis was conducted on the remaining 32 items. A seven factor solution was extracted using the 32 items retained after Step 1. The cumulative variance explained with this seven-factor model was 61.65%. Review of the scree plot suggested a scree started at the eighth factor, providing additional support for the seven-factor solution. Factor loadings and explained variances for each factor in the seven-factor solution are reported in Table 3. In the factor analysis, five of the observed factors map directly to subscales suggested by Raine (1991) Eccentric Behavior, Social Anxiety, Magical Thinking, Odd Speech, and Unusual Perceptions. Raine s other four subscales are represented in the two remaining factors. Within each of these two remaining factors, two subscales are collapsed Ideas of Reference and Suspiciousness are collapsed in the first factor, and No Close Friends and Constricted Affect are collapsed in the second factor. Based on the results of this factor analysis, average scale scores were computed for each factor. Means, standard deviations, internal consistency reliability estimates, and bivariate correlations for the seven subscales are reported in Table 4. There are several notables. First, internal consistency was adequate for each subscale. Second, each of the subscales was mod-

12 THE SPQ-BRIEF REVISED 527 estly correlated with each other subscale. With only two exceptions, the magnitudes of the correlations were between.29 and.53, roughly equating to a medium effect size level. Magical Thinking showed smaller correlations with two measures: No Close Friends/Constricted Affect and Social Anxiety, suggesting that these dimensions may be relatively independent. Finally, it is worth noting that No Close Friends/Constricted Affect and Social Anxiety measures showed only modest correlation with each other, suggesting they are by no means redundant with each other. Step 3: Confirmatory Factor Analysis (CFA) of the SPQ-BR To confirm the factor structure of the SPQ-BR, the third subsample of respondents (n = 465) was analyzed using the structural equation modeling software package AMOS 7 (Arbuckle, 2006). Item level listwise deletion resulted in a working sample of 441 participants. We conducted our analyses across four nested models. In Model 1 we sought to examine whether our 32 items loaded on one uni-dimensional latent factor (we expected this model to have poor fit). In Model 2 we tested a model of 7 under-related latent factors, with each latent factor representing one of the factors extracted in our exploratory factor analysis (we expected this model to have poor fit). For each latent factor, the items for each of the subscales were included as indicators for that latent factor. In Model 3, based on work by Raine and Benishay (1995) we predicted that three higher order constructs would exist Interpersonal, Cognitive-Perceptual, and Disorganized. We predicted that the Interpersonal second order construct would be a function of two subscales: No Close Friends/Constricted Affect and Social Anxiety. We predicted that the Cognitive-Perceptual second order construct would be a function of three subscales: Ideas of Reference/Suspiciousness, Magical Thinking, and Unusual Perceptions. We predicted that the Disorganized second order construct would be a function of two subscales: Eccentric Behavior and Odd Speech. We predicted that Model 3 would demonstrate acceptable fit. In Model 4 we sought to examine the possibility that the SPQ is better conceptualized as having 4 factors. As such, in Model 4, we predicted a similar factor structure to Model 3 but with the No Close Friends/Constricted Affect and Social Anxiety factors reflecting distinct factors (Brown et al., 2008; Chmielewski & Watson, 2008; Lewandowski et al., 2006). We predicted that Model 4 would demonstrate acceptable fit. For each model four measures of model fit were calculated: χ 2, comparative fit index (CFI), root mean square error of approximation (RMSEA), and standard root mean residual (SRMR). A non-significant χ 2 indicates good model fit; however, χ 2 is sensitive to sample size. The CFI and RMSEA are less sensitive to sampling characteristics and take degrees of freedom into account. A CFI value of.90 or higher (Medsker, Williams, & Holahan, 1994), a RMSEA value of.06 or lower, and an SRMR value of.08 or lower are also indicative of good model fit (Hu & Bentler, 1999). Results for the

13 528 TABLE 3. Principal Component Analysis of Revised SPQ-Brief Revised with Likert-Type Response Options (n = 465) Factor Raine (1991) Subscales Do you sometimes feel that people are talking about you? (IR4) Ideas of Reference.78 Do you sometimes feel that other people are watching you? (IR5) Ideas of Reference.74 When shopping do you get the feeling that other people are taking notice of you? (IR6) Ideas of Reference.68 I often feel that others have it in for me. (S1) Suspiciousness.62 Do you sometimes get concerned that friends or co-workers are not really loyal or trustworthy? (S2) Suspiciousness.62 Do you often have to keep an eye out to stop people from taking advantage of you? (S3) Suspiciousness.55 Do you feel that you cannot get close to people. (CF1) No Close Friends.83 I find it hard to be emotionally close to other people. (CF2) No Close Friends.77 Do you feel that there is no one you are really close to outside of your immediate family, or people you can confide in or talk to about personal problems? (CF3) No Close Friends.67 I tend to keep my feelings to myself. (CA1) Constricted Affect.66 I rarely laugh and smile. (CA2) Constricted Affect.57 I am not good at expressing my true feelings by the way I talk and look. (CA3) Constricted Affect.51 Other people see me as slightly eccentric (odd). (EB1) Eccentric Behavior.81 I am an odd, unusual person. (EB2) Eccentric Behavior.76 I have some eccentric (odd) habits. (EB3) Eccentric Behavior.75 People sometimes comment on my unusual mannerisms and habits. (EB4) Eccentric Behavior.70 Do you often feel nervous when you are in a group of unfamiliar people? (SA1) Social Anxiety.83 I get anxious when meeting people for the first time. (SA2) Social Anxiety.82

14 529 I feel very uncomfortable in social situations involving unfamiliar people. (SA3) Social Anxiety.80 I sometimes avoid going to places where there will be many people because I will get anxious. (SA4) Social Anxiety.63 Do you believe in telepathy (mind-reading)? (MT1) Magical Thinking.83 Do you believe in clairvoyance (psychic forces, fortune telling)? (MT2) Magical Thinking.82 Have you had experiences with astrology, seeing the future, UFO s, ESP, or a sixth sense? (MT3) Magical Thinking.76 Have you ever felt that you are communicating with another person telepathically (by mind-reading)? (MT4) Magical Thinking.63 I sometimes jump quickly from one topic to another when speaking. (OS1) Odd Speech.77 Do you tend to wander off the topic when having a conversation? (OS2) Odd Speech.76 I often ramble on too much when speaking. (OS3) Odd Speech.70 I sometimes forget what I am trying to say. (OS4) Odd Speech.64 I often hear a voice speaking my thoughts aloud. (UP1) Unusual Perception.76 When you look at a person or yourself in a mirror, have you ever seen the face change right before your eyes? (UP2) Unusual Perception.63 Are your thoughts sometimes so strong that you can almost hear them? (UP3) Unusual Perception.59 Do everyday things seem unusually large or small? (UP4) Unusual Perception.47 Eigen Value Variance Explained 10.49% 10.38% 8.96% 8.73% 8.41% 8.31% 6.34% Note. The conceptual factor for each item as defined originally by Raine (1991) are reported for comparison purposes. Bolded items were included in the original SPQ-Brief measure. Factor loadings less than.35 are not reported. Codes reported in parenthesis correspond with indicators reported in Figures 3 and 4.

15 530 COHEN TABLE 4. Means, Standard Deviations, Internal Consistency Reliability Estimates, and Bivariate Correlations for the SPQ-Brief Revised (n = 465) Mean SD # items Ideas of Reference/ Suspiciousness No Close Friends/ Constricted Affect ** 3. Eccentric Behavior **.34** 4. Social Anxiety **.47**.37** 5. Magical Thinking **.15**.37**.13** 6. Odd Speech **.30**.50**.33**.33** 7. Unusual Perceptions **.37**.46**.29**.49**.46** **p <.01. four models are reported in Table 5. Also reported in Table 5 are χ 2 -differences tests to test for significant improvement in model fit based on the successive models tested. The 3-factor and 4-factor models were both found to be invariant across gender in terms of structural weights and covariances. As expected, neither Model One or Model Two demonstrated acceptable fit. Model 3 and Model 4 both demonstrated acceptable fit. However, based on the χ 2 -differences tests, neither model significantly explained the data better compared to the other model; both the three-factor and four-factor model adequately explain the data. Results for the three-factor model are reported in Figure 1 and the results for the four-factor model are reported in Figure 2. Convergent Validity. The correlations between the SPQ-BR and the QOL scores are presented in Table 6. Each of these correlations was statistically significant such that increasing schizotypal traits were associated with poorer quality of life. DISCUSSION In study 2, the SPQ-BR was constructed, using a Likert-style response format, and featuring seven trait subscales. Although these subscales are by no means redundant with each other, confirmatory factor analysis suggests that they have a three or four factor higher-order structure that is generally consistent with prior research (see below for elaboration). The TABLE 5. Model Fit Statistics Based on Four Nested Models Tested (n = 441) 2 df CFI RMSEA SRMR 2 (df) Uni-dimensional Model Un-correlated model (0) a 3-Factor Model (10)** 4-Factor Model (1) Note. The 3-factor and 4-factor models were both found to be invariant across gender in terms of structural weights and covariances **p <.01. a χ 2 -difference not tested given no change in degrees of freedom.

16 THE SPQ-BRIEF REVISED 531 FIGURE 1. The three-factor super-ordinate structure of the revised SPQ-Brief Revised. Item wordings for all indicators are reported Table 3. psychometric properties of this instrument are promising. Moreover, the brevity of the SPQ-Brief was maintained, such that the final questionnaire comprised only 32 items. Finally, convergent validity was demonstrated in that increased severity of schizotypal traits, measuring by the SPQ-BR, was associated with notable declinations in quality of life. GENERAL DISCUSSION The chief advance of the present studies is the realization of a psychometrically sound brief measure of schizotypal traits. The practical benefits of using a brief over the full instrument were demonstrated in the dramatically increased incomplete response rate for the longer SPQ (>20%) compared to the briefer measure (<10%). The SPQ-BR is an improvement over the SPQ-Brief in that it covers a broader set of pathological traits seven as opposed to three domains. Moreover, it offers a coherent factor structure manifesting in a three or four-part (see below) super-ordinate factor structure. These factor structures have not been supported in studies of the original SPQ-Brief (Compton et al., 2007; Aycicegi et al., 2005). Our use of an expanded response format greatly improves the potential range of responses. Given that the SPQ is typically employed as a screening tool to identify extreme scorers for laboratory studies, improved resolution of

17 532 COHEN FIGURE 2. The four-factor super-ordinate structure of the revised SPQ-Brief Revised. Item wordings for all indicators are reported in Table 3. SPQ traits vis a vis expanded score ranges should improve accuracy in identifying schizotypic individuals. Implications of these findings, with regard to assessment and theoretical understanding of schizotypy, are expounded below. The SPQ-BR covers a broader range of schizotypal traits than any of the other commonly used measures (save the traditional full SPQ). This is particularly important given that schizotypy is considered a heterogeneous construct comprised of three-factors that are closely tied to the positive, negative, and disorganized symptom structure observed in schizophrenia more generally (Liddle, Barnes, Morris, & Haque, 1989). However, recent research employing the full SPQ has raised questions about whether TABLE 6. Correlations Between SPQ-BR Total and Subscale Scores and Subjective and Objective Quality of Life (QOL) Constricted Cognitive Affect/No Close Total Perceptual Disorganized Friends Interpersonal Objective QOL Subjective QOL Note. All r values are statistically significant p <.001.

18 THE SPQ-BRIEF REVISED 533 schizotypy reflects more than three factors (Chmielewski & Watson, 2008; Compton, Goulding, Bakeman, & McClure-Tone, 2009). Our data suggest that schizotypy is best explained by seven subordinate factors subsumed under three or four super-ordinate factors. This is an issue of importance as changes to the operational definition of schizotypal personality disorder are considered for the fifth version of the DSM. Our data suggest that a more refined definition of seven traits, as opposed to nine, reflecting three or four broad clusters is worth considering. With regard to the super-ordinate factor solution of the SPQ-BR, our data can not speak to whether this is best represented by the original three-factor solution recommended by Raine and Benishay (1995) or by a modified four-factor solution, because the confirmatory fit statistics for these models were nearly identical. At odds is the relationship between social anxiety and more traditionally-defined negative traits (i.e., Constricted Affect, No Close Friends). There are three possibilities worth entertaining regarding their relationship. First, it could be that Social Anxiety and Constricted Affect/No Close Friends reflect a common construct. Parsimony would argue for this insofar as the three-factor solution is less complicated than the four-factor solutions. Second, it could be that social anxiety and negative schizotypy reflect distinct (yet modestly correlated) dimensions that capture different aspects of schizotypal pathology. In support of this notion, several recent schizotypy studies employing the Chapman scales (Chapman et al., 1995) have found that social anxiety is actually more strongly related to magical ideation and perceptual aberrations than social anhedonia (Brown et al., 2008; Lewandowski et al., 2006). Third, as noted by Siever, Bernstein, and Silverman (1991), it could be the case that social anxiety reflects a nonspecific comorbid condition that is largely independent of schizotypal processes. This would suggest that the Constricted Affect/No Close Friends factor is a truer indicator of negative schizotypy, and that a three factor schizotypy model can be maintained by excluding Social Anxiety. In evaluating these three possibilities, we favor the third because (a) social anxiety is not specific to schizotypy and is typical of a broad swath of mental illnesses, and (b) original conceptualizations of negative schizotypy were based on social apathy where an individual neither enjoys nor is distressed by Interpersonal interactions (i.e., experiences low levels of social anxiety; Kraepelin, 1971). As evidence of this, the Chapman s Social Anhedonia Scale was deliberately constructed to reflect social apathy as opposed to social anxiety (Eckblad, Chapman, Chapman, & Mishlove, 1982). The present study provides limited insight into this issue, which thus remains for future research. Consistent with prior research using the full SPQ (Cohen & Davis, 2009), SPQ-BR scores were associated with poorer quality of life. Nearly all of these correlations were in the small to medium effect size range, suggesting that the effects of schizotypy are by no means trivial. Correlations were particularly pronounced between the interpersonal factors and reduced self-reported prosocial behaviors (e.g., contacting friends and signifi-

19 534 COHEN cant others) as well as satisfaction with the amount of friendship in one s life. These findings also document the convergent validity of the SPQ-BR. Use of instruments that capture a broad range of schizotypal traits may help lay the foundation for improved insight into the neurodevelopmental trajectory of schizophrenia. There is evidence to suggest that a sizable portion of schizotypic individuals show symptom exacerbation and eventual onset of schizophrenia-spectrum disorder. Thus it could be the case that different schizotypic manifestations may be associated with different psychosis-spectrum disorders or at least different illness courses. Evidence from a 10-year longitudinal study supports this claim in that positive schizotypy was predictive of psychosis spectrum disorders more generally, while 21% of individuals with social anhedonia met criteria for a schizophrenia-spectrum disorder by study s end (Kwapil, 1998). Longitudinal studies to date provide limited information to answer these questions because they have not included the full spectrum of schizotypal traits (i.e., generally neglecting disorganized schizotypy). Furthermore, trait definitions used in these studies are often not comprehensive. For example, negative schizotypal definitions often employ the Chapman Social Anhedonia Scale, which measures social anhedonia without regard to constricted affect (e.g., Gooding et al., 2005; Kwapil, 1998; Kwapil, Barrantes-Vida, & Silvia, 2008). A psychometric identification tool with a broader scope, such as the SPQ-BR, would help overcome these limitations. The factor structure was invariant to gender suggesting that the SPQ- BR is appropriate for use with both males and females. Consistent with Raine (1992), both studies presented here found that males showed somewhat more severe schizotypal traits across each factor dimension. Why males score higher is unclear from our data. Males typically show emergence of symptoms earlier than females so it is possible that females would show similar levels of SPQ rated pathology if assessed at an age reflecting peak window of risk (Goldstein, 1997). Thus, males higher scores could reflect more an age (or more precisely, a stage of neurodevelopment) rather than a gender issue. It is also worth noting that males with diagnosable schizophrenia-spectrum pathology tend to show more negative/deficit symptoms (Kirkpatrick et al., 2001), so it stands to reason that their scores on the Interpersonal dimension would exceed those of females. Regardless of the underlying reason, the present data suggest that males and females should be considered separately when employing the SPQ-BR. Several limitations warrant mention here. Foremost, the schizotypal participants were college students. Although student samples are typical of this line of research, they may not represent the entire population. Second, assessment of schizotypal traits was dependent on self-report measures, which are problematic in that they provide information on only subjective-perceived attitudes. Clinical ratings, while costly for a sample size this large, would be an important adjunct for use in further research. Third, our data were cross-sectional in nature, an approach that limits understanding how symptoms manifest over time. Finally, our analyses

20 THE SPQ-BRIEF REVISED 535 were underpowered to determine the influence of ethnicity on SPQ ratings. It would be important to determine whether our factor structure is invariant across ethnicity. The present data provide important reliability and validity data supporting the use of the SPQ-BR as an instrument for the psychometric assessment of schizotypy. Application of this measure may provide important information regarding schizophrenia-spectrum disorders, and may be helpful for understanding the neurodevelopmental progression of the disorder. REFERENCES Addington, J. (2004). The diagnosis and as- ety and social anhedonia to psychometrically sessment of individuals prodromal for identified schizotypy. Joursessment schizophrenic psychosis. CNS Spec- nal of Social & Clinical Psychology, 27, trums, 9, American Psychiatric Association. (1994). Di- Chapman, L. J., & Chapman, J. P. (1983). agnostic and statistical manual of mental Infrequency scale. Madison, WI: Un- disorders (DSM-IV) (4th ed.). Wash- published test. ington, DC: American Psychiatric Press. Chapman, J. P., Chapman, L. J., & Kwapil, Arbuckle, J. L. (2006). Amos (Version 7.0) T. R. (1995). Scales for the measurement Chicago: SPSS. of schizotypy. In A. Raine, T. Axelrod, S. R., Grilo, C. M., Sanislow, C., & Lencz, & S. A. Mednick (Eds.), Schizotypal McGlashan, T. H. (2001). Schizotypal personality (pp ). New personality questionnaire-brief: Factor York: Cambridge University Press. structure and convergent validity in Chmielewski, M., & Watson, D. (2008). The inpatient adolescents. Journal of Per- heterogeneous structure of schizotypal sonality Disorders, 15, personality disorder: Item-level factors Aycicegi, A., Dinn, W. M., & Harris, C. L. of the schizotypal personality questionnaire (2005). Validation of Turkish and English and their associations with versions of the schizotypal per- obsessive-compulsive disorder symptoms, sonality questionnaire-b. European dissociative tendencies, and Journal of Psychological Assessment, normal personality. Journal of Abnormal 21, Psychology, 117, Bedwell, J., Kamath, V., & Baksh, E. (2006). Clark, L. A., & Watson, D. (1995). Constructing Comparison of three computer-administered validity: Basic issues in objective cognitive tasks as putative en- scale development. Psychological Assessment, dophenotypes of schizophrenia. Schizophrenia 7, Research, 88, Cohen, A. S., & Davis, T. E. I. (2009). Schizotypal Bellack, A. S., Bennett, M. E., Gearson, S. J., symptoms and quality of life in a Brown, C. H., & Yang, Y. (2006). A ran- large nonclinical sample. Comprehensive domized clinical trial of a new behavioral Psychiatry, 50, treatment for drug abuse in peo- Comery, A. L., & Lee, H. B. (1992). A first ple with severe and persistent mental course in factor analysis (2nd ed.). illness. Archives of General Psychiatry, Hillsdale, NJ: Erlbaum. 63(4), Compton, M. T., Chien, V. H., & Bollini, A. M. Boyle, G. J. (1991). Does item homogeneity (2007). Psychometric properties of the indicate internal consistency or item brief version of the schizotypal person- redundancy in psychometric scales? ality questionnaire in relatives of Personality and Individual Differences, patients with schizophrenia-spectrum 12, disorders and non-psychiatric controls. Brown, L. H., Silvia, P. J., Myin-Germeys, I., Schizophrenia Research, 91, Lewandowski, K. E., & Kwapil, T. R (2008). The relationship of social anxi- Compton, M. T., Goulding, S. M., Bakeman,