Factor Structure of the Inventory of Adult Temperament Styles

Temperament refers to stylistic and relatively stable traits that subsume intrinsic tendencies to act and react in somewhat predictable ways to people, events, and stimuli. Temperament can be defi ned by four bipolar styles: extroversion-introversion, practical-imaginative, thinking-feeling, and organizedfl exible. These four styles provide the theoretical structure for the group administered Inventory of Adult Temperament Styles (IATS).Validity evidences are reported on 853 middle class adults (43% female), ages 15 through 54 (M = 26.5, SD = 8.6), who were studying at universities or working in Sao Paulo state (Brazil). The degree items are consistent with the four temperament styles was examined through item response theory and confi rmatory factor analysis. Results confi rm the proposed theoretical structure. Invariance testing suggests that scores refl ecting temperament styles have similar meanings for men and women. Implications as well as future directions for research are discussed.

Contemporary research generally characterizes temperament as stylistic and relatively stable traits that subsume intrinsic (i.e., biological) tendencies to act and react in somewhat predictable ways to people and events (Casalin, Luyten, Vliegens, & Meurs, 2012;Rothbart & Hwang, 2002;Worobey, 2001).Although the terms temperament and personality at times are used synonymously, differences exist in their origins, emergence, and robustness.In contrast to personality, the earlier emergence of temperament (Martin, García, Sanz, Arias, & Matamala, 2009;Thomas & Chess, 1977, 1989;Walters, 2011) is attributable to its strongly biological origin.Moreover, while personality may include sixteen or more qualities including motivation and values (Goldsmith et al., 1987), temperament generally is characterized by fewer qualities -often four bipolar qualities.
Brazilian scholarship on temperament focuses mainly on maternal childrearing (Malhado & Alvarenga, 2012;Melchiori, Alves, Souza, & Bugliani, 2007) or pathological characteristics related to temperament (Fuscaldo, Bisol, & Lara, 2013;Lara, Ottoni, Brusntein, Frozi, & Bisol, 2012;Mochcovitch, Nardi, & Cardoso, 2012;Tavares & Gentil, 2007).However, in Brazil, most temperament tests still lack suffi cient validity evidences and norms to be used with confi dence (Guzzo, Riello, & Primi, 1996;Ito, Gobita, & Guzzo, 2007;Klein, Putnam, & Linhares, 2009).For example, data from the Questionnaire for Types Assessment (QUATI; Zacharias, 2003), designed to assess three types of temperament (extroversion-introversion, sensation-intuition, and thinking-feeling), display adequate test-retest reliability as well as concurrent validity evidence.However, the QUATI's factor structure has been questioned, given evidence that the test measures nine rather than three factors and that evidence for the sensation-intuition factor is limited (Moraes & Primi, 2002).Thus, additional evidence as to the factor structure of temperament tests in Brazil is needed.The purpose of this research was to examine the factor structure of the Inventory of Adult Temperament Styles (IATS).This research also investigates possible gender differences in temperament styles among Brazilians, given pervasive international research showing such differences.

Participants
Data were collected on 853 middle class adults (43% female), ages 15 through 54 (M = 26.5,SD = 8.6), who attended universities or worked in various professions in Brazil's state of Sao Paulo.These persons were contacted by their professors or employees and invited to participate in the study.

Instrument
The theoretical foundation for the Inventory of Adult Temperament Styles (IATS) is based on prevailing temperament models (e.g.Myers et al., 1998;Oakland et al., 1996), that utilize four bipolar temperament styles: extroversion-introversion, practical-imaginative, thinkingfeeling, and organized-fl exible.One hundred items forcedchoice items were created to be consistent with this model.Each item has two opposing options.

Procedure
Approval by the Brazilian Ethics Research Committee was secured (N.277/07).The IATS was group administered to college and professionals at university or work settings.Raw scores were used for data analysis in order to preserve statistical properties of the observed score distribution.The fi t of items with their respective temperament styles was evaluated using the two parameter logistic model from item response theory (IRT).This model was utilized because the magnitude of relationships between items and temperament styles was presumed to vary.A three parameter model was not utilized because the guessing parameter is not of interest when examining dichotomous temperament items.Separate analyses were performed for the extroversion-introversion, practical-imaginative, thinking-feeling, and organized-fl exible styles, to verify the proposed four-factor model.
IRT analyses were conducted using the Xcalibre 4.1 computer program (Guyer & Thompson, 2012) following the maximum likelihood estimation approach.Items were removed if they demonstrated either negative point bi-serial correlations or poor fi t with their respective temperament dimension.Items were considered to display acceptable fi t if the z residual fi t statistic did not exceed the two-tailed critical alpha value of .01.
The next stage of data analysis involved the use of item parceling methods.Item parceling involves combining items into groups (Gorsuch, 1983).Item parceling was warranted because the variance of individual items was restricted due to their dichotomous format.Moreover, parceling yields aggregated scores that have greater variability than scores obtained from individual items.Three parcels of approximately equal size were constructed for each dimension of the four bipolar temperament styles, as the use of three parcels per dimension produces a justidentifi ed construct that ". . .has only one unique solution that optimally captures the relations among the items" (Little, Cunningham, Shahar, & Widaman, 2002, p. 162).
Confi rmatory factor analysis (CFA) was used to exam correspondence of the remaining 79 items to a four factor structure, with each factor representing one bipolar temperament style.The factor loadings of parcels on their respective temperament styles were freely estimated.Additionally, measurement errors and correlations among the styles were freely estimated.CFA was used to test the extent to which the hypothesized model presented in Figure 1 fi ts data obtained from the total sample as well as data from disaggregated male and female samples.Next, parcel invariance was examined across men and women to determine if the data demonstrate confi gural invariance (i.e., whether the measured variables defi ne the same factors and have the same pattern of loadings across males and women) and metric invariance (i.e., whether the magnitude of observed variables' factor loadings are invariant across men and women).
CFA analyses were performed using Amos 16.0 (Arbuckle, 2007) computer software following the method of maximum-likelihood estimation.Multiple fi t indices were used to judge model fi t.Multiple index presentation strategies help control the error rate for accepting misspecifi ed models (Hu & Bentler, 1999).The indices utilized in this study include the comparative fi t index (CFI), root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), and change in chi-square value (∆χ 2 ).The successive models used to test invariance are nested.Thus, a likelihood ratio test (∆χ 2 ) was used to determine if adding theoretically substantive constraints result in a large and statistically signifi cant drop in χ 2 values (Keith, 2005).Additionally, change in CFI and RMSEA values was examined when evaluating invariance.CFI changes of .01 or less and RMSEA changes of .015or less were used as criteria for establishing invariance.

Results
Seventy-nine items were acceptable based on fi t with the two parameter logistic model from IRT. Twenty-one items were removed because they demonstrated either negative point bi-serial correlations or poor fi t with their respective temperament dimension.Twenty-one items assess extroversion-introversion, 19 assess organized-fl exible, 18 assess practical-imaginative, and 21 assess thinking-feeling styles.
All scales (i.e., dimensions) had reasonable a parameter estimates.The typical range for estimates is .30to 1.50 (Guyer & Thompson, 2012), and estimates for all retained items either fell within or approximated this range.The range and average of a parameter estimates within each dimension are as follows: .22 to 1.16 with a mean of .58 for extroversion-introversion, .32 to 1.16 with a mean of .64 for organized-fl exible, .17 to 1.16 with a mean of .53 for practical-imaginative, and .17 to 1.16 with a mean of .53 for thinking-feeling.Results suggest that items within each dimension have discriminative value when assessing individual differences in temperament styles.
All estimates for the b parameter fall within the typical range of -3 to 3 (Guyer & Thompson).The b parameter is related to item means and refl ects the probability of endorsing a response indicative of a particular temperament style.The range and average of b parameter estimates within each dimension are as follows: -2.45 to 1.42 with a mean of -.80 for extroversion-introversion, -1.71 to 2.52 with a mean of -.43 for organized-fl exible, -2.14 to 2.52 with a mean of -.17 for practical-imaginative, and -2.87 to 2.21with a mean of -.24 for thinking-feeling.
Individual items were selected from item pools for each style and assigned randomly to one of three parcels in to reduce unwanted sources of variance and parameter bias when examining model fi t (Little et al., 2002).Parcel data are reported in Table 2. Global fi t statistics, which refl ect the fi t of the CFA model presented in Figure 1 to the observed data, are presented in Table 3. Model parameter estimates, including standardized regression weights for parcels and correlations among latent temperament styles, are presented in Tables 2 and 4, respectively.Although fi t statistics generally indicate reasonable fi t between the hypothesized model and the observed data, the CFI value observed for women falls below .95.All parcels load moderately well on their respective factors.
Results from tests of measurement invariance are presented in Table 5. Confi gural and metric invariance are supported for men and women.An invariant struc-tural confi guration indicates that items measure the same constructs in both groups.Metric invariance indicates that factors are calibrated in a similar way and thus have similar meanings for men and women (Steinmetz, 2011).

Discussion
Assessment of temperament styles provides important information that can be used to better understand people's predisposition to display certain behaviors and help them to achieve life pursuits.Although scholarly literature on temperament exists in Brazil (Guzzo et al., 1996;Klein et al., 2009), its focus largely has been on child rearing practices (Malhado & Alvarenga, 2012;Melchiori et al., 2007) and use in clinical setting (Lara, Ottoni, et al., 2012, Mochcovitch et al., 2012).Its limited focus precludes the use of temperament assessment for other purposes such as vocational guidance or personnel assessment.Therefore, this research examined the factor structure of the Inventory of Adult Temperament Styles (IATS) together with possible gender differences on thinking-feeling styles among Brazilians.
Results indicate that the fi t between the hypothesized model and the observed data generally is adequate, thus confi rming a four factor structure of temperament (extroversion-introversion, practical-imaginative, thinking-feeling, and organized-fl exible styles).These fi ndings are congruent with international research with this temperament model as reported by others (e.g., Benson, et al., 2009;Callueng & Oakland, 2014;Myers & Myers, 1980;Rowinski et al., 2013) and add evidence as to the validity of the IATS's internal structure.
Small to moderate correlations among some temperament styles were found, with organized-fl exible and practical-imaginative styles sharing approximately 17% common variance.Thus, the temperament qualities measured by these two styles are not unique to them and instead share a more general temperament quality -albeit modestly.Other attempts to develop items that measure only a pure, single and unique temperament style have not been successful (Benson et al., 2009;Myers & Myers, 1980;Oakland et al., 1996).This research also examined investigated possible gender differences in temperament styles among Bra-zilians, given pervasive international research showing such differences.Invariance testing suggests that scores refl ecting temperament styles have similar meanings for men and women.In general, international research suggests that men are more likely to prefer a thinking style while women are more likely to prefer a feeling style (Callueng & Oakland, 2014;Francis, 2006;Joyce, 2010;Myers et al., 1998).This fi nding is consistent with previous Brazilian data related to gender and creative styles (Wechsler, 2009), thus confi rming that, in western societies, women display a greater need for attachment (Else-Quest et al., 2006).Thus, differences in women's temperament styles might be refl ecting their clear tendency to look for harmony, to value sentiments over logic and their tendency to be more sympathetic and friendly in their interactions than men.

Limitations and Directions for Future Research
Although results confirm the IATS' theoretical structure, evidence regarding the reliability of scores derived from the IATS is needed.Moreover, external validity evidence regarding the relations of IATS scores with other measures of temperament, as well as relations with other constructs, is necessary.Evidence regarding the utility of IATS scores for applied purposes such as career counseling, organizational development, or treatment planning would be benefi cial so that test users can make informed decisions regarding the consequences of test use.Finally, evidence regarding differences on the frequency of temperament styles in the Brazilian population is needed given the present lack of appropriate tests for measuring Brazilian's temperament styles.

Figure 1 .
Figure 1.Measurement model for the Inventory of Adult Temperament Styles.

Table 1
Descriptions of Temperament QualitiesThis dimension describes individuals' orientations as to when they make decisions.Those with organized preference styles generally prefer to fi nalize decisions and have issues settled as soon as possible while those with fl exible preference styles generally prefer to delay decisions and keep their options open.
Those with organized styles generally Those with fl exible styles generally want to plan/schedule are fl exible in commitments persist, are dependable seek opportunity for play keep personal space neat tolerate disorder of possessions enjoy predictable/structure enjoy surprise/adaptive to change

Table 2
Descriptions of Parcels and Loadings of Parcels on Associated Styles

Table 4
Correlations among Latent Styles of Temperament

Table 5
Summary of Results for Measurement Invariance Testing Across Gender GroupsNote.AIC = Akaike Information Criterion, CFI = Comparative Fit Index, RMSEA = Root Mean Square Error of Approximation, SRMR = Standardized Root mean Square Residual.