A CROSS-CULTURAL SHORTENED FORM OF THE ZKPQ (ZKPQ-50-CC) ADAPTED TO ENGLISH, FRENCH, GERMAN, AND SPANISH LANGUAGES 

Anton Aluja¹, Jérôme Rossier², Luis F. García¹, Alois Angleitner³, Michael Kuhlman⁴,and Marvin Zuckerman⁴

¹University of Lleida (Spain)

²University of Lausanne (Switzerland)

³University of Bielefeld (Germany)

⁴University of Delaware (United States)

INTRODUCTION

The Zuckerman-Kuhlman Personality Questionnaire (ZKPQ) was designed within the framework of Zuckerman's Alternative Five personality model (Zuckerman, Kuhlman, Teta, Joireman, & Kraft, 1993). This model was developed by comparing structures based on 3, 4, 5, and 6 factor solutions of several personality questionnaires (Zuckerman, Kuhlman, & Camac, 1988; Zuckerman, Kuhlman, Thornquist & Kiers, 1991). These two studies of scale factors were followed by one based upon factor analyses of items from the scales most clearly defining the factors (Zuckerman et al., 1993). From these studies, Zuckerman et al. (1993) concluded that both the three and five-factor models were robust and reliable. Based on these results, a questionnaire was developed using the items most clearly defining the five factors, excluding items showing a strong social desirability influence. The 99-item version contained the following scales: Impulsive Sensation Seeking (ImpSS, 19 items), Neuroticism-Anxiety (N-Anx, 19 items), Aggression-Hostility (Agg-Host, 17 items), Activity (Act, 17 items), Sociability (Sy, 17 items), and Infrequency (Inf, 10 items). The Inf scale is only used to assess the validity of individual records and is not intended to be a trait scale.  Internal consistency for both sexes varied between 0.72 and 0.86 (for more detail on the development of the ZKPQ and its reliability and validity, see Joireman & Kuhlman, 2004, and Zuckerman, 2002).

The raw scores of the ZKPQ, the EPQ (Eysenck, & Eysenck, 1975), and the NEO-PI-R (Costa, & McCrae, 1992) scales were factor analyzed conjointly (Zuckerman et al., 1993). Solutions of three, four, and five factors were obtained. In the three factor rotation, the Sy and Act scales loaded on the Extraversion factor with the EPQ-E, and the NEO-E, and the N-Anx loaded on a Neuroticism factor along with the EPQ-N and the NEO-N scales. ImpSS, Agg-Host, the Psychoticism Scale of the EPQ-R, and the Agreeableness and Conscientiousness Scales of the NEO-PI-R loaded mainly on the third factor. This factor was split into two in the four-factor structure. Conscientiousness, Psychoticism, and Impulsive Sensation Seeking defined one factor, whereas Agreeableness, Aggression-Hostility, and Openness loaded mainly on the other one. Finally, in the five-factor structure, the 30 facets of the NEO-PI-R were used instead of the Big-Five dimension scales. These results corroborated the previous 4-factor structure plus an additional factor exclusively formed by the 6 Openness facets. These results have been replicated in Spanish populations (Aluja, García, & García, 2002; 2004a; Romero, Luengo, Gómez, & Sobral, 2002).

The original version of the ZKPQ has been adapted in different countries and languages: China (Wu, Wang, Du, Li, Jiang, & Wang, 2000), Germany (Ostendorf, & Angleitner, 1994), Italy (De Pascalis, & Russo, 2003), Japan (Shiomi, Kuhlman, Zuckerman, Joreiman, Sato, & Yata, 1996), and Spain (Catalan [Gomà-i-Freixanet, Valero, Puntí & Zuckerman, 2004; Gomà-i-Freixenet, Wismeijer, & Valero, 2005], and Spanish versions [Aluja et al., 2002, 2004a; Herrero, Viña, González, Ibáñez, & Peñate, 2001; Romero et al., 2002]). In general, the factor structure and the psychometric properties have replicated results found for the original English version.  Two German versions of the ZKPQ have been developed and both confirm the basic structure of the test as described in the American results (Angleitner, Riemann, & Spinath, 2004; Schmitz, 2004). Angleitner et al. (2004) also reported the first behavior genetic twin study on the ZKPQ scales. Schmitz (2004) has used the ZKPQ in acculturation studies of immigrants to Germany and studies of the relationships between the ZKPQ dimensions and love styles.

As far as we know, two short versions of the ZKPQ have been developed. The first one (Zuckerman, 2002) is made up of 35 items, with seven items per scale. Items with the highest correlations with the total scores on each of the five factors and the greater response variance were selected. Several items were eliminated because of redundancy of content, and the next highest correlation was substituted. These short scales had Cronbach alphas between 0.62 and 0.79. The second one (Aluja, García, & García, 2003a) is composed of 69 items selected through the following procedure, which combines exploratory and confirmatory factor analysis (Byrne, 1993; Yadama, & Drake, 1995): (1) items with loadings equal to or higher than 0.30 on their own factor were selected; (2) a confirmatory factor analysis was carried out over these selected items. One item from each pair of highly correlated items (i.e. with a modification index of the correlations between the error terms of the items higher than 100) was suppressed; (3) a second factor analysis was conducted using the 76 surviving items. Following the same criteria in point 1, 69 items were retained for the short version. These 69 items were distributed across scales as follows: ImpSS (14), N-Anx (18), Agg-Host (13), Sy (13), and Act (11). Internal consistencies were similar to the 89-item version, and varied between 0.74 and 0.81.

The 69-item version was developed following an empirical criterion with no pre-conceived idea as to the number of items. The resulting scales presented a number of items per scale between 11 (Activity) and 18 (N-Anx) (Aluja et al., 2003a). However, the number of items retained in the 69-item version and the statistical procedure described above made it possible to further reduce the number of items per scale. This approach generated a shorter version maintaining the same psychometric properties, including reliability coefficients, of the ZKPQ as well as the structural validity of the questionnaire. Two instances support this possibility. Firstly, the Activity scale improved its psychometric properties in the short version, although six items were suppressed (Aluja, et al., 2003a). Secondly, the 10-item Extroversion and Neuroticism scales of the EPQ-R presented good psychometric and structural properties (Aluja, García, & García, 2003b). According to these findings, we hypothesised that ZKPQ scales formed by 8 to12 items will only have similar psychometric properties to the original questionnaire. The availability of these short scales has obvious advantages in practical and research settings, since the administration and correction time are considerably reduced (Buchanan, Johnson, Goldberg, 2005).

The aim of the present study was to develop a version of the ZKPQ with a robust structure and acceptable psychometric properties in four languages: English (United States), French (Switzerland), German (Germany), and Spanish (Spain). In developing this version, a procedure similar to that used by Aluja et al. (2003a) was carried out combining exploratory and confirmatory factor analysis. Note that cross-cultural studies are usually focused on differences between cultures. However, this study looks for structural similarities in the four languages analysed.

The present study is focused on an etic (versus emic) research approach. The etic perspective looks for universal traits and is centered on comparison across cultures. The emic perspective seeks to describe personalities with culturally sensitive elements (Triandis, & Such, 2002). The results of the current study pose several questions for personality assessment beyond the cross-cultural validity of the ZKPQ. Developing the same measure for different countries has the advantage of establishing the same metric for all countries analyzed. This characteristic would largely facilitate cross-cultural research with Zuckerman’s Alternative Five Personality Model. 

METHOD
Subjects

The total sample consisted of 4,621 subjects (1,667 males [36.1%], and 2,954 females [63.9%]) from 4 countries. Inclusion in the study was restricted to subjects under 35 years of age in order to reduce the age differences between samples. The numbers of subjects in the four countries were 517 from Germany (Mean age: 26.94; SD: 4.31), 962 from Spain (Mean age: 21.39; SD 2.97), 764 from the French-speaking part of Switzerland (Mean age: 21.68; SD 3.05), and 2,378 from the United States of America (the age variable was not coded for this sample). Since all subjects in the American sample were university students enrolled in introductory psychology courses, the mean age is known to be around 19 years. Frequencies for sex were 117 males and 400 females for Germany, 351 and 611 for Spain, 294 and 470 for Switzerland, and 905 and 1,473 for USA. The samples from Spain, Switzerland, and the United States of America were university students. The German sample was a pool of two sub samples: one of university students, and one taken from the Bielefeld twin study.

Measures

 ZKPQ. The long form of the ZKPQ has 99 items, and measures the following personality factors:  Impulsive Sensation Seeking (ImpSS), Neuroticism-Anxiety (N-Anx), Aggression-Hostility (Agg-Host), Activity (Act), and Sociability (Sy). Reliability coefficients (Cronbach alpha) for the ZKPQ ranged between .72 and .83 in the original American study (Zuckerman et al., 1992; Zuckerman et al., 1993).  The German and Spanish versions of the ZKPQ have been published by Ostendorf, and Angleitner (1994), and Aluja, et al., (2002), respectively.  The French version is under revision (Rossier, Verardi, Massoudi, & Aluja, submitted).  These three non-English versions show a factor structure and reliability indices similar to those for the American ZKPQ.  The Infrequency scale was not used in the statistical analyses of these studies.  

RESULTS

Descriptive statistics, t-tests comparisons and alpha coefficients for the ZKPQ

Table 1 shows descriptive statistics (Mean, standard deviation, Skewness, and Kurtosis), alpha coefficients and differences between calibration and validation samples. Table 1 shows no significant mean differences between the groups, nor were differences found in their standard deviations. Given the large sample size and thus high statistical power for these comparisons, it is quite unlikely that these non-significant results correspond to a Type-II error. Skewness and Kurtosis values were under 1 for the five dimensions. Alphas ranged between 0.74 and 0.86, and were quite similar in both groups.  Additionally, descriptive statistics for the Infrequency scale indicate that the test validity was adequate.  This table suggests that both samples can be properly compared.

 

                                                                               Table 1 

Descriptive statistics (Mean, Standard Deviation, Skewness, and Kurtosis), alpha coefficients, t-tests and

 effect size  between calibration and validation sample.

(a)    Note: Neuroticism-Anxiety (N-Anx), Impulsive-Sensation Seeking (ImpSS), Activity (Act), Sociability (Sy), Aggression-Hostility (Agg-Host), and Infrequency (Inf). ES: Effect size

Exploratory Factor Analysis (EFA)

       Table 2 shows the five factor principal components (PC) solution with orthogonal (Varimax) rotation obtained in the calibration sample. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was 0.87, and Bartlett's Test of Sphericity (BTS) yielded approx. Chi-Square=44,523.30; d.f.: 3,972 (p<0.001).  The KMO measure of sampling adequacy is an index which examines the appropriateness of factor analysis. It should be 0.50, and the higher the better. This measure of sampling adequacy compares magnitude of correlations with the magnitude of partial correlation coefficients. Small values indicate that correlations cannot be explained by other variables, and factor analysis may be inappropriate. Additionally, a principal axis (PA) with direct oblimin rotation solution was obtained in the calibration sample. Tucker’s Congruency coefficients between the latter and the principal components solution were 0.99, 0.99, 0.99, .099, 0.98, for N-Anx, ImpSS, Act, Sy, and Agg-Host, respectively; overall, global congruence was 0.99. Given the virtual equality of both solutions, the PC-Varimax solution was selected since this procedure was the one used in original work on ZKPQ structure (Zuckerman, et al., 1993). Also, it should be noted that an orthogonal solution is more consistent with Zuckerman’s theoretical conceptualization of his five factor model. As can be seen in Table 2, a total of 80 items (ImpSS: 20; N-Anx: 19; Act: 15; Sy: 12 and Agg-Host: 14) had loadings equal to or higher than 0.30 on their own factor, and lower on the remaining factors. These 80 items were employed for the CFA’s described below. Items deleted from further analysis at this stage are underlined in Table 2.

 

                                           Table 2

                                Principal Components factorial matrix and

                                Varimax rotation (calibration sample)

                                Note: F-I: N-Anx; F-II:ImpSS; F:-III: Act; F-IV: Sy; F-V: Agg-Host.

Confirmatory Factor Analysis (CFA)

A CFA was performed in the validation sample on the variance-covariance matrix of the 80 items described above.  This analysis employed the AMOS 4.01 statistical package (Arbuckle, 1999).  The estimation method was Maximum Likelihood, which seems warranted given the size of the sample and the apparent normality of the variables (see Table 1).  Model identification was achieved by fixing regression coefficients of the error terms over the endogenous variables and also the variance of factors to 1 (MacCallum, Browne, & Sugawara, 1996).  Each item was linked to a single factor.  Note that only item 53 loaded on a different factor (ImpSS instead of Sy).  Table 4 shows the fit indices for both orthogonal and oblique solutions for these 80 items. The AMOS procedure was used to examine the correlation of the error component between the 80 items.  This analysis showed that several item pairs were highly to moderately correlated, with Modification Indices (MIs) higher than 100.  The following 25 item pairs with MIs higher than 100 are ordered by the MIs from the largest to smallest (inter-item correlations are in brackets and the item number corresponds to the long form of the ZKPQ): 17-63 (0.66), 57-11 (-0.57), 14-89 (0.59), 1-29 (-0.45), 42-47 (0.43), 19-29 (-0.42), 7-61 (0.51), 9.48 (0.47), 18-28 (0.34), 6-29 (0.36), 6-1 (-0.34), 2-66 (-0.43), 68-87 (0.48), 1-19 (0.34), 51-61 (0.49), 43-48 (-0.43), 45-55 (0.43), 24-55 (0.44), 7-61 (0.43), 9-78 (0.40), 6-19 (-0.31), 2-85 (0.40), 67-91 (0.36), 18-75 (0.21), and 25-76 (0.39).

Examination of these items revealed within-pair redundancy, and as recommended elsewhere (Byrne, 1993; Yadama, & Drake, 1995), one item was deleted from each pair. Within all but two of the above pairs, the item with the lowest loading was deleted. Following this procedure, all ZKPQ scales were composed of 10 or more items, excepting Sy. Two item pairs on this scale (43-48 and 68-87) were strongly correlated, but no item was deleted in order to test all of them through the standardized regression weights. A CFA model was performed on the 57 items remaining after the deletions described above. Again, analysis was performed with the AMOS 4.01 statistical package with Maximum Likelihood estimation. Based on the standardized regression weights, those items with the lowest values were removed until 10 items per scale remained.  Note that all weights were significant. No item from the Sy scale was removed, because the standardized regression weights of this scale were acceptable (ranging between 0.33 and 0.61) and similar to those obtained in the remaining four scales (see Table 3). With these modifications, the questionnaire was reduced to 50 items (10 per scale). These items appear in the appendix, henceforth referred to as the Zuckerman-Kuhlman Personality Questionnaire Cross Cultural 50-item version (ZKPQ-50-CC).

The results of a CFA (Amos, Maximum Likelihood) performed on these 50 items are given in Table 4. Firstly, the differences in the RMSEA were significant between the 80 and the 50 item models. The ECVI, an index to discern among several related models, was considerably lower for the 50-item than for the 80-item model, and the only model with a c²/d.f. ratio lower than 5 was the oblique 50-item model. Also, the GFI, AGFI, improved greatly from 80 to 50 items. However, for both 50 and 80-item analyses, the NFI, TLI, and CFI fell below the usual cut-off value (0.90; Hu, & Bentler, 1999). Differences between orthogonal and oblique models are minimal. This finding is not surprising, since there were no secondary loadings on the EFA solution. Correlations between the latent variables for oblique models were (50/80-item model): ImpSS/N-Anx: -0.11/-0.04; ImpSS/Agg-Host: 0.31/0.34, ImpSS/Act: 0.19/0.14, ImpSS/Sy: -0.40/-0.36, N-Anx/Agg-Host: 0.28/0.24, N-Anx/Act: -0.05/-0.08, N-Anx/Sy: 0.17/0.19, Agg-Host/Act: 0.04/0.03, Agg-Host/Sy: -0.08/-0.09 and Act-Sy: -0.18/-0.19.

Table 3

 Standardized Regression Weights (SRW) for the five dimensions (validation sample) 

Note: Neuroticism-Anxiety (N-Anx), Impulsive-Sensation Seeking (ImpSS), Activity (Act),

                          Sociability (Sy),   Aggression-Hostility (Agg-Host).

Table 4

 Fit indices of the ZKPQ 80-item, and the ZKPQ 50-item models (validation sample)

(a)      c²: Chi-Square. d.f.: Degrees of Freedom.  GFI: Goodness of Fit Index. AGFI: Adjusted Goodness of Fit Index. NFI: Normed Fit Index. TLI: Tucker Lewis Index. CFI: Comparative Fit Index. RMSEA: Root Mean Square Error of Approximation. The LO and HI columns contain the lower and upper limit of a 90% confidence interval on the population value of RMSEA. ECVI: Expected Cross-Validation Index.

(b)      The associated p values were always lower than .0001.

Using the same CFA procedure described above, five models were evaluated, one for each of the separate factors of the ZKPQ-50-CC. This procedure analyses the construct validity of each scale when the effect of the others is removed. The resulting fit indices are given in Table 5, which shows very good values for N-Anx, ImpSS, Acty, and Agg-Host, although slightly lower ones for Sy. With regard to Sy, it should be noted that this scale has two pairs of highly correlated items. Overall, these results show good consistency for four out of five factors in the 50-item model. The standardized regression coefficients obtained with the 50-item model also support the use of the five scales, although future studies could improve the Sy scale by replacing one item of each redundant pair with new items that correlate with the theoretical construct of Sy but are less redundant in relation to other items in the scale. Correlations in the total sample between the scores obtained with the original ZKPQ (89-items) and the ZKPQ-50-CC were 0.90, 0.87, 0.94, 0.95, and 0.92 for the N-Anx, ImpSS, Sy, Agg-Host, and Act scales, respectively.  Thus, the scores obtained with the 50-item instrument are quite comparable to those obtained with the longer version.  The correlations between ZKPQ factor-derived scales for the 50-item version were: ImpSS/N-Anx: 0.05; ImpSS/Agg-Host: 0.19, ImpSS/Act: 0.14, ImpSS/Sy: 0.20, N-Anx/Agg-Host: 0.18, N-Anx/Act: -0.09, N-Anx/Sy -0.09, Agg-Host/Act: -0.01, Agg-Host/Sy: 0.08 and Act-Sy: 0.14. Congruent with Zuckerman's orthogonal model, correlations are quite low.

Table 5

 Fit indexes of each dimension in the ZKPQ 50-item version (validation sample).

(a) c² : Chi Square. d.f.: Degrees of Freedom.  GFI: Goodness of Fit Index. AGFI: Adjusted Goodness of Fit Index. NFI: Normed Fit Index. TLI: Tucker  Lewis Index. CFI: Comparative Fit Index. RMSEA: Root Mean Square Error of Approximation. ECVI: Expected Cross-Validation Index.

(b) The associated p values were always lower  than .0001.

Cross-cultural comparisons of the ZKPQ-50-CC

 Table 6 shows the descriptive statistics (Mean, Standard deviation, Skewness, and Kurtosis), and alphas of the ZKPQ-50-CC for each country. All scales have satisfactory alpha coefficients above .70 in each country, except for the Agg-Host scale in Germany, Spain and Switzerland, and the Sy scale in Germany. However, these latter alpha coefficients were only slightly lower (.60-.68). Scores show a normal distribution. Only the N-Anx scale in the American, German and Swiss samples and the Act scale in Spain show a slightly high kurtosis. Table 6 also shows correlations between the five factors and age for all countries except the United States. On average over the five dimensions and the three countries, age accounts for 1.24 % (d = .20) of the total variance. Thus, the global effect of age is small (Cohen, 1969). In Germany, correlations range from -.09 to -.19 and age accounts on average for 1.90% (d = .26) of the total variance. In Spain, correlations range from -.03 to -.16 and age accounts on average for 1.31% (d = .20) of the total variance, and in Switzerland, correlations range from -.01 to     -.11 and age accounts on average for 0.52% (d = .12) of the total variance.

Table 6 

Descriptive statistics (Mean, Standard deviation, Skewness, and Kurtosis) and alphas by country,

and correlations with the age variable.

(a)     

          Correlations with age can not be computed for the American sample.

(b)      * p < 0.05; ** p < 0.01; *** p > 0.001

   Note: Underlined effect sizes correspond to non-signifcant differences between the two countries being compared.

It might be important to consider the gender variable because differences in personality traits between males and females are well established in the literature (v.g., Lynn, & Martin, 1997). Regarding the ZKPQ, such differences are mainly observed for ImpSS (males score higher), and N-Anx (females score higher) (Zuckerman, 1994; 2002). A two-way ANOVA on N-Anx with country and gender as factors indicated that there was a significant country effect, F(3,4613) = 12.73, p < .001 (d = .18), a significant gender effect, F(1,4613) = 386.36, p < .001 (d = .58), and a significant interaction effect, F(3,4613) = 12.85, p < .001 (d = .18). Only the gender effect was associated with a d higher than .20. Women scored globally higher than men on this scale. For ImpSS, there was a significant country effect, F(3,4613) = 69.83, p < .001 (d = .43), and a significant gender effect, F(1,4613) = 44.57, p < .001 (d = .20), but no significant interaction effect, F(3,4613) = .16, p = .93 (d < .01). Country and gender effects were associated with a d higher than .20.  The American sample scored particularly high and the German sample particularly low on this scale (see Table 6). Moreover, women scored lower than men on this scale. For Act, there was a significant country effect, F(3,4613) = 31.93, p < .001 (d = .28), and a significant gender effect, F(1,4613) = 18.50, p < .001 (d = .12), but no significant interaction effect, F(3,4613) = .38, p = .77 (d < .01). Only the country effect was associated with a d higher than .20. The Spanish sample scored particularly high and the Swiss sample particularly low on this scale. For Sy, there was a significant country effect, F(3,4613) = 44.16, p < .001 (d = .35), and a significant gender effect, F(1,4613) = 15.44, p < .001 (d = .10), but no significant interaction effect, F(3,4613) = 1.99, p = .11 (d = .06). Only the country effect was associated with a d higher than .20. The American and the Spanish samples scored particularly high and the German and the Swiss samples particularly low on this scale. For Agg-Host, there was a significant country effect, F(3,4613) = 38.40, p < .001 (d = .30), a significant gender effect, F(1,4613) = 4.42, p = .04 (d = .06), and a significant interaction effect, F(3,4613) = 11.74, p < .001 (d = .18). Only the country effect was associated with a d higher than .20. The American sample scored particularly high on this factor compared to the three other samples. Globally, the country factor accounted for 2.5% of the total variance, the gender variable for 1.9% of the total variance, and the interaction term for 0.3% of the total variance. Finally, since the age variable also plays a role in personality mean-level (Eysenck, & Eysenck, 1985), it should be controlled. Specifically for the ZKPQ, ImpSS and N-Anx decrease with age (Zuckerman, 1994). When controlling for age, the country variable, excluding the American sample, accounted for 1.5% of the total variance, the gender variable for 2.1% of the total variance, and the interaction term for 0.4% of the total variance. Gender and age group means by country are available upon request from the first author.

         Finally, a principal component with Varimax solution over the 50-item version was obtained separately for each country. Table 7 shows the loadings per item and factor, as well as the congruence indexes between each pair of countries. Congruence coefficients were computed following the Orthogonal Procustes method used by McCrae, Zonderman, Costa, Bond, and Paunonen (1996)[1]. The Scree plot supported a five factor solution in the four countries. The KMO was between 0.80 and 0.86, and the variance accounted for by the five factors was around 34% in all cases.  The 50 items loaded as expected, except for item 48 (Sy) in the German sample (0.11), item 92 (Sy) in both the German (-0.08) and Swiss (-0.01) samples, and item 3 (Agg-Host) in the Swiss sample (0.29) (Again, item number is with regard to the original ZKPQ). Note that total congruencies were higher than 0.90 for all comparisons across countries. To reinforce these analyses, a multi-group CFA was conducted. The oblique 50-item model was simultaneously fitted in the four countries. This model obtained a c² of 9602.629 (p < .001; d.f.= 4660). The AGFI and the CFI were similar to the model presented in table 3 (.83, and .77, respectively). Also, the c²/d.f. (2.06) and the RMSEA (0.022) obtained good values. These data suggest that this structure is invariant across the four countries.

DISCUSSION

            The main aim of this study was to obtain a ZKPQ version with acceptable psychometric properties in four languages. In this way, a 50-item version (10 per scale) of the ZKPQ       (ZKPQ-50-CC) was developed using several multivariate techniques such as Exploratory and Confirmatory Factor Analysis. This short version of the ZKPQ presents an equivalent factor structure in the four languages, since all scales obtained high congruence indexes across the four languages and multi-group CFA results were satisfactory.  As regards reliability, the mean of the five scales was higher than 0.70 in the four countries, and all indexes were similar to those obtained for the original version. 

The results of this study represent, therefore, a double contribution to the ZKPQ literature: 1) demonstrating the cross-cultural validity of Zuckerman’s Alternative Five Personality Model, and 2) developing a short version with good psychometric and structural properties in four different languages (English, French, German, and Spanish). The cross-cultural design and the use of multivariate statistical procedures in its development give this short version the edge over the 35, and 69-item ZKPQ versions, developed with American and Spanish samples, respectively.

            Developing short versions of personality questionnaires is a common practice in personality assessment (Buchanan, et al., 2005). Some instances are the Eysenck Personality Questionnaire (EPQ-R, Eysenck, Eysenck, & Barrett, 1985), the NEO-PI-R (Aluja, García, Rossier, & García, 2005; McCrae, & Costa, 2004), the Sensation Seeking Scale, form V (Aluja, García, & García, 2004b; Hoyle, Stephenson, Palmgreen, Lorch, & Donohew, 2002), the Big Five Goldberg’s adjectives (García, Aluja, & García, 2004; Goldberg, 1992), the STAXI-CA (del Barrio, Aluja, & Spielberger, 2004). These short versions can reduce the administration time for the scales at little cost in terms of reduced reliability. The decision as to the number of items of these short versions varies depending on the authors, but it is usually a data driven procedure based on statistical criteria such as, for instance, classical theory indexes or factor loadings. In the present study, we have conducted a similar procedure to that used by Aluja and colleagues, (2003a). Following this procedure, the Sy scale would have 8 items (MI’s > 100). However, we decided to retain 10 items for this scale for several reasons: a) to maintain the same number of items for all scales, b) all standardized regression weights of the Sy scale were high which gives support to the construct validity and dimensionality of the scale, c) the internal consistency of the 10-item scale in the four countries was acceptable, and d) there was a high correlation between the original and the 10-item Sy scales.

Fabrigar, Wegener, Maccallum and Strahan (1999) claimed that Principal Components (PC) does not distinguish between common, unique and error variance, and the components extracted can not be equated to common factors. If the goal is to arrive at a parsimonious representation of the associations among measured variables, the PC is not appropriate. In these cases, EFA extraction methods, such as Principal Axis (PA), are strongly recommended. This recommendation is reinforced when communalities are low (v.g. < 0.40), a pattern usually found in factor analysis at the item level. Also, most studies have performed an orthogonal rotation. Since psychological variables usually correlate, oblique rotation methods are preferable to orthogonal ones. However, no sharp differences between PC and PA, or between orthogonal and oblique solutions, were found in the present study. Moreover, the ZKPQ five-factor structure was mainly orthogonal in the four countries, since not many secondary loadings appeared, and correlations between factors were low.

The usefulness of confirmatory factor analysis for personality research has been criticized (Raykov, 1998; Vassend, & Skrondal, 1995, 1997).  It has been argued that: a) fit indexes are usually poor; b) there are too many parameters to be estimated, and c) there are a great deal of secondary loadings and highly correlated items. In spite of these limitations, the CFA is a useful tool for analysing questionnaires and the selection of items, and for testing the factorial validity of scores from a measurement instrument (Byrne, 2001; pp. 104-119). Following this idea, fit indexes, multi-group analysis and standardized regression weights were used in the present study. The results strongly support the construct and cross-cultural validity of the items selected for each scale. Although fit indexes could be improved if the number of items was reduced to five per factor, this strategy would result in limited content diversity and reduce the reliability of the scale.

Alpha reliability coefficients were similar to those of the original 89-item version. Moreover, correlations between the short form of the scale and the original versions were around 0.90. This is especially interesting in the case of the ImpSS scale, since the new scale includes only 2 items of the impulsivity facet. However, the remaining 8 items also reflect a kind of impulsivity expressed in spontaneous types of sensation seeking without reflection on possible risks or consequences. Markon, Krueger and Watson (2005) maintain that an "Unconscientious Disinhibition" factor, represented by the ImpSS scale in the ZKPQ, is one of four primary factors found in a hierarchal structure common to normal and abnormal personality. The 10-item ImpSS scale presents a robust one-factor structure, as can be seen in Table 7.

The effect sizes of the mean level comparisons across countries show only slight differences between Americans, German, Spanish, and Swiss mean scores (d ≤ 0.43). The largest differences may be observed between the American and German samples. Globally the country variable accounted for 2.5% of the total variance, which corresponds to a d-value of .32. Age differences between samples could explain this pattern, especially for the ImpSS scale. Indeed, the impact of the country variable decreased to 1.5% of the total variance after controlling for age (d = .24). The gender differences were similar to those usually observed using the ZKPQ (Zuckerman, 2002). Interestingly, no interaction between gender and country with a d-value higher than .20 was observed.  However, it should be remarked that this study is mainly focused on the structural validity of the ZKPQ-50-CC across cultures, and the results support the structure developed in the four countries.

Summing up, the present results support the equivalence of the short version (ZKPQ-50-CC) across the four languages and countries: English (United States), French (Switzerland), German (Germany), and Spanish (Spain). If the validation process had been conducted separately for each language, it is possible that different items would have been selected. Our procedure, in contrast, faciltates the simultaneous selection of the best items in all languages, and tests the resulting factor structure within each one. The ZKPQ-50-CC is, therefore, a useful instrument for assessing personality traits involved in Zuckerman's alternative five-factor model, and facilitates cross-cultural research since the same items are used in all languages. Nevertheless, the cross-cultural convergent validity of this short scale still needs to be tested, by demonstrating that it correlates equally with expected criteria across cultures.