Construction and psychometric evaluation of a short form of the Broad Autism Phenotype Questionnaire2

Broad Autism Phenotype (BAP) represents a group of personality traits expressed in limitations in social relations and pragmatic speech dimension, and rigid behavior. The Broad Autism Phenotype Questionnaire (BAPQ) measures personality traits which are crucial in defining the BAP. In the present research, three studies were conducted with the general aim to create a short form of the BAPQ. Study 1 was carried out to determine the factor structure of the BAPQ in a sample of 501 students and to select items for the short form. Obtained components: Aloofness, Rigidity, and Pragmatics, corresponding to the structure of the instrument proposed by authors, accounted for 26.61% of variance. Study 2 was conducted to examine factor structure of the BAPQ short form (BAPQ-SF), in a sample of 298 students. This solution explained 45.76% of the total variance. The aim of Study 3 was to determine psychometric characteristics of the BAPQ-SF in a sample of students (N = 294). Three-factor model of the BAPQ-SF was confirmed. Correlations of the BAPQ-SF with the Autism-Spectrum Quotient (AQ) and the Delta 10 suggest convergent and discriminant validity of the BAPQ-SF.

& Kwapil, 2007) showed positive correlations between autistic tendencies measured by the Autism Spectrum Quotient (AQ; Baron-Cohen et al., 2001) and schizotypy on the whole. At the subscales level, the strongest correlation was obtained between the Social skills domain of the AQ and the interpersonal domain of schizotypy which encompasses negative schizotypal traits. According to the findings of Ingersoll, Hopwood, Wainer, and Donnellan (2011), the Broad Autism Phenotype Questionnaire (BAPQ; Hurley, Losh, Parlier, Reznick, & Piven, 2007) measures personality traits which are crucial in defining BAP, and with regard to the type of subscales it most closely corresponds to the theoretical concept of BAP when compared to two other instruments, AQ (Baron-Cohen et al., 2001) and The Social Responsiveness Scale (Constantino, 2002;Ingersoll, 2010;Ingersoll et al., 2011).
An increase in use of brief measurement instruments has taken place in personality research during the last decades. By being less time-consuming for participants, shorter measures ensure higher participation rates (Konstabel, Lönnqvist, Walkowitz, Konstabel, & Verkasalo, 2012) and are less likely to trigger negative participants' reactions such as fatigue, boredom, or frustration (Robins, Hendin, & Trzesniewski, 2001), which may distort their answers. Further, shorter measurement instruments are more suitable for broad scope research in which measures of multiple constructs are employed. The general aim of the present research was defined based on these considerations.

General aim and specific aims
General aim of the research was construction of a Serbian short form of the BAPQ for the assessment of autistic personality traits in general population. The three studies, which are parts of this research, had their specific goals. The aim of the first study was to determine the factor structure of the BAPQ and to select items for the short form of the questionnaire. The aim of the second study was to examine factor structure of the BAPQ short form (BAPQ-SF). In the third study, the goal was to establish psychometric characteristics of the BAPQ-SF.

General procedure
Participants in each of the three studies were students of the Belgrade University. Bearing in mind that autistic traits are related to scientific/professional field in which a person is engaged, so that these traits are more pronounced among mathematicians, physicists and engineers compared to persons in the fields of humanities (Baron-Cohen et al., 1998;Baron-Cohen et al., 2001), we aimed to include students of various scientific disciplines in our samples. Thus, participants in the research were students of the following fields: special education and rehabilitation, traffic and transport engineering, political sciences, mathematics, physics, organizational sciences, medicine, agriculture, police studies, and geography. The samples in three studies were unrelated and did not overlap.
The students provided answers to the claims from the BAPQ scale at their faculties before or after lectures. Before data collection, participants were asked to participate in a research on their sociability and their relations with other people. All addressed students agreed to participate. The researchers guaranteed the participants anonymity and that the obtained results will only be used for scientific purposes.

Study1
There is still no extensive data on psychometric characteristics of the BAPQ and other instruments measuring broad autistic phenotype in larger, non-clinical samples. Therefore, we decided to conduct a study with the aim to determine factor structure of the BAPQ in a sample of students of the University of Belgrade, with regard to the experience from other studies with similar samples (Ingersoll, 2010;Ingersoll et al., 2011) and to select items for the short form of the instrument. This study was aimed to be the first step of our research on the manifestations of BAP in Serbia.

Method
Sample. The questionnaire was filled out by 625 students of the University of Belgrade. The participants who gave answers to all the items were singled out, and the sample used for analyses consisted of 501 students. The participants were from 19 to 24 years of age (M = 21.49, SD = 1.17). With regard to gender, 29.2% of the participants were male, and 70.8% female.

Results
Data adequacy for factor analysis was determined by Kaiser-Meyer-Olkin measure (KMO .87), which exceeds the recommended value of .60, and by Bartlett's test of sphericity which reached statistical significance pointing to the factorability of the correlation matrix, χ 2 (630) = 4555.55, p = .000. With the aim to determine the BAPQ structure, Principal axis factoring with Oblimin rotation was applied on 36 items of the scale. Considering previous research of the instrument structure (Hurley et al., 2007;Ingersoll et al., 2011;Sasson et al., 2013;Wainer et al., 2011), three factors were specified to be extracted (Table 1). Three extracted components together account for 26.61% of the total variance. The first factor (12 items, α = .88) accounts for 17.72% of the variance. The second factor (8 items, α = .69) accounts for 5.09%, and the third factor (8 items, α = .64) accounts for 3.79% of the variance. Items 3, 5, 6, 10, 21, 24, 26 and 34 were left out because they do not reach the minimum load level of .30.
Correlations between obtained factors are weak to moderate (Table 2).

Discussion
The obtained item loadings on three extracted factors in our analysis are generally in accordance with the contents of the BAPQ subscales. All items from Factor I correspond to items which constitute the Aloof subscale in the original version of the BAPQ, with the exception of item 7 which refers to good understanding of interlocutors, and which belongs to the Pragmatic subscale in the original version, while it was placed in Factor I according to the results of our analysis. Sasson et al. (2013) obtained similar results. According to the results of their research, item 7 is distributed in both Aloof and Pragmatic subscales, but with higher loading on the Aloof subscale. Factor II includes all items as factor Rigid obtained by the authors of the scale. The only difference is that items 3, 6, 24 and 26 are left out in our analysis due to low values of factor weight (< .3). Wainer et al. (2011) also left out items 24 and 26, as well as items 19 and 30. Our Factor III consists of 8 items, while the factor Pragmatic in Hurley et al. (2007) includes 12 items. Each of the 8 items which belong to our Factor III correspond to the items from the factor Pragmatic defined by the authors of the scale (Hurley et al., 2007).
Item 7 which according to its content, corresponds to Factor 3, did not reach the load level of .30 on this factor in our analysis. Item 7 is highly loaded on the first factor. With nearly equal loadings, item 11 is distributed both in factor I and factor III. The content of this item refers to pragmatic aspects of speech, which is why we have decided to observe it as an item belonging to the Pragmatic factor. In Sasson et al. (2013), item 11 is distributed in both Pragmatic and Aloof with nearly equal loadings.
Based on the results of factor analysis, and on the presented item analyses, a short version of the BAPQ was devised in this study. The BAPQ short version includes 14 items obtained by eliminating items which belong to Factor I with loadings lower than .65 and communalities lower than .40 and items on factors II and III which had loadings lower than .35 and communalities lower than .20.

Study 2
The aim of Study 2 two was to explore the factor structure of the short version of the BAPQ devised in Study 1.

Method
Sample. In the second study, the sample consisted of 298 students aged between 19 and 24 (M = 21.28, SD = 1.28). Male participants constituted 14.1% of the sample, and female 85.9%.

Results
By means of Exploratory factor analysis, it was determined that most correlation coefficients in correlation matrix have the value over .30, that the value of KMO measure of sampling adequacy is satisfactory .87, and that the result of Bartlett's spherical test reaches the level of statistical significance χ 2 (91) = 1419.49, p = .000. Exploratory factor analysis -Principal axis factoring with Oblimin rotation with fixed number of three factors was performed. Simple structure was obtained. The three factors accounted together for 45.76 % of the total variance. The first factor, which includes items 1, 9, 16, 25, 31, and 36, accounted for 31.50% of the variance (Cronbach's α = .86). The second factor which includes items 2, 11, 29, and 32, accounted for 8.28% of the variance (Cronbach's α = .71). Four items of the third factor: 13, 19, 22 and 30, accounted for 5.99% of the variance (Cronbach's α = .66) Results of the exploratory factor analysis are presented in Table 3. Correlations between the extracted factors are presented in Table 4. The obtained moderate intercorrelations (from .29 to .42) are in accordance with theoretical assumptions on the structure of the BAP (Hurley et al., 2007).  Table 5 shows the values of reliability and inter-item correlation in three extracted groups of items.

Discussion
The results of factor analysis confirmed grouping of items in three factors which determine BAP: Aloof, Rigid, and Pragmatic. As the original version, the short version of the BAPQ with 14 items also has solid reliability, when the number of items and criteria in social sciences are taken into account (Cortina, 1993). The employed criteria for the retention of items have proven to be adequate in a sense that the created short form preserved acceptable levels of internal consistency. Also, the percentage of the explained variance is higher compared to the factor analysis presented in Study 1.

Study 3
We conducted Study 3 with the aim to examine psychometric properties, i. e. validity and reliability of the BAPQ short version developed in studies 1 and 2. We sought to establish validity of the questionnaire by testing its construct, convergent, and discriminant validity. To examine the construct validity of the short form of the BAPQ confirmatory factor analysis was performed. We tested the three-factor model consisting of aloofness, pragmatic language deficits, and rigidity, as proposed by developers of the BAPQ.
To establish convergent validity of the BAPQ-SF we examined its correlations with the AQ ( Baron-Cohen et al., 2001). AQ is one of the most widely used measures of BAP. It comprises five dimensions: Social skills, Communication, Attention to detail, Attention switching and Imagination. By comparing the BAPQ and AQ structure it can be seen that they only partially overlap. Social skills, Communication and Attention switching conceptually correspond to the BAPQ subscales of aloofness, rigidity and pragmatic language deficit, respectively. We expected moderate to high correlations between corresponding subscales. On the other hand, we expected low correlations between the BAPQ-SF and Imagination and Attention to detail because these two dimensions are not included in the BAPQ.
To assess discriminant validity of the questionnaire correlation with the Delta 10 (Knežević, Opačić, Kutlešić, & Savić, 2005) was investigated. The Delta 10 is a measure of disintegration, which is conceptualized as a basic personality trait of psychosis proneness. Disintegration encompasses following facets: General Executive Impairment, Perceptual Distortions, Enhanced Awareness, Depression, Paranoia, Mania, Flattened Affect, Somatic Dysregulations, and Magical Thinking. The facets are postulated to stem from disintegration of the information processing systems responsible for reality testing, hence the name of the trait (Knežević et al., 2016). Autism and schizophrenia are distinct, but partially overlapping phenomena (e.g., Chisholm, Lin, Abu-Akel, & Wood, 2015). Likewise, autistic traits are correlated with various indicators of psychosis proneness in general population (Claridge & McDonald, 2009;Hurst et al., 2007;Russell-Smith, Bayliss, & Maybery, 2013;Russell-Smith, Maybery, & Bayliss, 2011). Given these findings, we expected the BAPQ-SF and the Delta 10 to show low to moderate correlation as an evidence of discriminant validity.

Method
Sample. Questionnaires were administered to a sample of 294 students aged 19 to 26 years (M = 21.16, SD = 1.46). Male students comprised 43.2%, and female students 56.8% of the sample.
AQ (Baron-Cohen et al., 2001). The questionnaire consists of 50 items. Subjects rate to what extent they agree or disagree with the statements on a 4-point Likert scale. The item scores are summed to obtain total score and scores for subscales corresponding to dimensions of the questionnaire. Total scale Cronbach's ɑ = .68.
The Delta 10 (Knežević, Opačić, Kutlešić, & Savić, 2005). The questionnaire measures personality trait of disintegration.The short version of the questionnaire consisting of 20 items was used. Respondents rate the degree in which each statement refers to them on a 5-point Likert scale (Cronbach's ɑ = .93). The total score is obtained by averaging responses on all items. A higher total score indicates a higher level of disintegration.

Results
Correlations between the BAPQ-SF, the AQ, and the Delta 10 are presented in Table 6. As can be seen, the BAPQ-SF total score and its subscale scores show moderate to high positive correlations with the AQ total score and its subscales, with the exception of the attention to detail and imagination subscales of the AQ which are weakly or statistically insignificantly related to the BAPQ-SF scales. The BAPQ-SF total score is moderately positively related to the Delta. At the subscales level, Pragmatic subscale exhibits moderate, and Aloof and Rigid subscales exhibit weak correlations with the Delta. Confirmatory factor analysis was performed using IBM AMOS, version 24. Prior to CFA, data were screened for multivariate normality, outliers and missing data. There were no missing data or outliers, and Mardia's coefficient of multivariate kurtosis suggested no departure from normality. Model specifications included correlated factors, uncorrelated error variances, and the variance of one item on each factor fixed to one. Upon inspection of fit indices, a correlation between error terms between two items was allowed. Maximum likelihood estimation was employed. The path diagram of the model with allowed error term correlation, showing standardized regression weights and correlations between factors is presented in Figure 1. It is generally recommended to use several fit indices when assessing the model fit. We used the following indices: χ 2 , the Comparative Fit Index (CFI), the Tucker-Lewis Fit Index (TLI), the Standardized Root Mean Square Residual (SRMR), and the Root Mean Square Error of Approximation (RMSEA). We have followed the recommendations by Hair Black, Babin, & Anderson (2010) who determine cut-off values of fit indices depending on the number of observed variables and on the sample size. For our analysis, significant chi-square is expected and does not imply rejection of the model, given the relatively large sample size. Criteria for goodness-of-fit to be applied in the analysis, according to Hair et al. (2010), are as follows: CFI or TLI > .92, SRMR = .08 or less, RMSEA < .07. At first, the obtained fit indices suggested that the model did not adequately fit the observed data (χ 2 = 190.29, p = .000, CFI = .91, TLI = .89, SRMR = .06, RMSEA = .073). Upon inspection of the modification indices, a correlation between error terms of two items of Aloof subscale was added to the model, which resulted in adequate model fit with the following values of fit indices: χ 2 = 166.09, p = .000, CFI = .93, TLI = .92, SRMR = .06, RMSEA = .066. The allowed error term correlation is justified because the meaning of the two items is practically identical, which is not the case with other items in Aloof subscale. Cronbach's α coefficients of internal consistency obtained in this study were as follows: .85 for the BAPQ-SF total, .85 for Aloof, .72 for Pragmatic, and .66 for Rigid.

General Discussion
The results obtained in three studies support validity of the BAPQ short form as a measure of the theoretical concept of the BAP (Bishop et al., 2004;Bolton et al., 1998;Piven et al., 1997).
Factor Aloof corresponds to the theoretical concept of social isolation. Social aloofness is present in general population among persons with more prominent autistic characteristics, and is characterized by difficulties in establishing social contacts and making friends. Social aloofness is characterized by lack of interest in social interactions, withdrawal in making contacts, and "keeping aloof" (Gerdts & Bernier, 2011). Items with high loadings on factor Rigid can be related to resistance to change, as well as to an individual's decreased flexible reaction ability. Rigidity, preservation, difficulty in initiating new activities, tendency to routines and rituals are common characteristics of persons who belong to autism spectrum and broad autism phenotype (Hill & Frith, 2003;Losh, Childress, Lam, & Piven, 2008). Factor Pragmatic consists of items which refer to pragmatic deficits in communication in a social context. By analyzing the content of items, we can assume that it includes intonation, basic voice loudness, focusing on the conversation topic, changing the topic according to interlocutor's expectations, and efficiency in conveying a message, which all together represent respecting conversation rules in the process of communication.
Moderate to high correlations between the BAPQ short form and the AQ total scores and between corresponding subscales support convergent validity of the instrument. Correlations of the similar size between the BAPQ and the AQ total scores were obtained by other authors Nishiyama et al., 2014). Low to moderate correlations between the BAPQ-SF and its subscales and the Delta 10 obtained in this study are in line with the results of previous studies on the relationship between autistic traits as measured by the AQ and psychosis proneness (Claridge & McDonald, 2009;Hurst et al., 2007;Russell-Smith et al., 2013;Russell-Smith et al., 2011). This provides evidence for discriminant validity of the BAPQ-SF. Russell-Smith et al. (2013) point out that further research is needed to investigate the degree in which the observed correlations stem from actual overlap between autistic tendencies and schizotypy or reflect insufficient specificity of the instruments. Our findings that Pragmatic subscale is more strongly related to Disintegration (r = .50), than Aloof and Rigid subscales (r = .31 for both subscales) suggest that pragmatic language deficiencies represent an important area of intersection of the two phenomena or their operationalizations.
Our results provide evidence for construct validity of the BAPQ-SF. Confirmatory factor analysis results suggest that the structure of the scale corresponds to the three-factor structure of the original scale, with items in the short form loading on the same factors as in the original scale. The questionnaire and its subscales exhibit good to acceptable reliability as measured by internal consistency. Advantages of the short form of the BAPQ over the original scale, that stem from smaller number of items which require less time to respond, are the following: greater probability that a potential respondent will decide to participate, experiences of fatigue and boredom due to repetition of similar items are less likely to occur and to have an impact on respondents' answers (Konstabel et al., 2012;Robins et al., 2001). The short form of the instrument may be especially suitable for research encompassing a multitude of variables. On the other hand, potential limitations of the BAPQ short form should be noted. It could be less sensitive and less precise than the original form in the identification of persons who exhibit significant levels of the BAP. However, as it may be used primarily for the efficient screening, accompanied, when needed, by subsequent more detailed analyses using the BAPQ original form or similar instruments, these shortcomings may be overcome.

Limitations
There is limitation related to the use of a convenience sample of university students, which probably led to the reduction of results variability and limited the possibilities of their generalization. Further limitations of the research are non-balanced gender structure of the sample, and the lack of information on presence of autistic spectrum disorders or schizophrenia spectrum disorders in participants' families which could provide further important insights.