臺灣青少年性格發展之多重指標：從HEXACO六大性格特質切入

Multiple Indicators of Personality Development for Taiwanese Adolescents: The HEXACO Model of Personality as a Framework

青少年階段正處於生理與心理上的變化，除了生理的成熟，也經歷社會化的過程，而了解青少年的性格發展型態，有其重要的意涵。本研究目的為了解臺灣青少年的性格向度在個體間與個體內的發展連續性之特性，分別檢驗四種性格連續性：平均水準連續性、排序連續性、個人層次連續性、以及自比連續性。本研究以連續三年收集臺灣地區405位國中生與595位高職生之HEXACO性格向度的分數，並分別檢驗前述四種連續性。不過，在檢驗前述的四種連續性前，先行檢驗結構連續性，以確定所測量的性格量表在國中生與高職生之三個時間點皆有相同的性格結構後，再進行後續其他連續性的檢驗。綜合所有的結果顯示，在平均水準連續性與個人層次連續性上，僅發現誠實／謙遜有明顯下降，而其餘向度變化均不明顯；而在排序連續性與自比連續性上，發現國中生與高職生皆有高的連續性，其中，在排序連續性上，高職生的情緒性、外向性、嚴謹性及開放性之相關係數高於國中生。最後針對研究結果與研究限制進行討論，深入說明不同連續性指標變化的意涵與可能的原因，並對未來研究方向提出相關建議。

The development of personality traits has received considerable attention from scholars in recent 20 years (Caspi et al., 2005). However, few studies in Taiwan have examined the development of personality in teenagers. Therefore, this study, with a longitudinal design, analyzed the patterns of personality trait development in adolescents by using the Honesty-Humility, Emotionality, Extraversion, Agreeableness, Conscientiousness, and Openness to Experience (HEXACO) personality model. Mean-level continuity, rank-order continuity, ipsative continuity, and individual-level change were employed as indicators of personality development. However, structural continuity was first examined to validate the consistent structure of constructs across time (De Fruyt et al., 2006). Mean-level continuity is a measure of the average difference in the personality traits of the same group of individuals across time or simultaneously between different age groups (Wright et al., 2012). Studies (Ashton & Lee, 2016) have revealed a pattern of increasing agreeableness and a pattern of decreasing honesty-humility during adolescence. However, the development patterns for the remaining dimensions are inconsistent, and they differ between preadolescence and postadolescence. Rank-order continuity indicates the stability of individuals with respect to their relative rankings on certain personality traits within a group or between time points. This measurement is generally calculated using the coefficient of correlation between specific personality traits at two time points (Wright et al., 2012). Roberts and DelVecchio (2000) found that the average correlation coefficient in the age group from 12 to 17.9 years ranged from .46 to .48 (Borghuis et al., 2017; Klimstra et al., 2009). Third, individual-level change is a measure of intraindividual comparison; the differences in a particular personality trait in the same individual across time are evaluated. The reliability change index is primarily employed to compare an observed distribution with the theoretically expected distribution (i.e., normal distribution) and involves calculating standardized scores of the difference scores between two time points. Studies have indicated that the change ratios of adolescents in the United States and Estonia were higher than the theoretically expected distribution (McCrae et al., 2002; Pullmann et al., 2006). Finally, ipsative continuity indicates the continuity of overall personality traits over time and is commonly expressed using the Q correlation coefficient (Wright et al., 2012). there were higher cross-time stabilitiesy in overall personality during postadolescence than during preadolescence (Klimstra et al., 2009; Van Dijk et al., 2020).
This data analyzed here was from a study of the "investigation of adolescent personality and social behavior in daily life" (Hsu & Chang, 2015). There were 594 junior high school students (JHSSs) and 991 vocational high school students (VHSSs) in Taiwan over 3 years. Of the participants who completed the 3-year follow-up, 405 were JHSSs (M = 12.96, SD = 0.62; 185 male) and 595 were VHSSs (M = 15.73, SD = 0.59; 346 male). All participants were instructed to complete the Chinese version of the HEXACO-100 Personality Inventory (Hsu & Chang, 2015) each year. The scale comprises six major dimensions of personality (i.e., honesty-humility, emotionality, extraversion, agreeableness, conscientiousness, and openness), with 16 items in each dimension; these items are scored using a 5-point Likert scale (1 = completely disagree to 5 = completely agree). The Chinese version of each subscale has good reliability, with values of Cronbach's α ranging from .73 to .78.
Before the formal analyses, we employed multi-group confirmatory factor analysis (MGCFA) to validate the structural consistency across 3 years in these two groups. We constructed four models to assess invariance in the measurement structure over time in accordance with the recommendations of Meredith (1993): A configural invariance model, a factor loadings invariance model, an intercept invariance model, and a residuals invariance model. The results revealed structural continuity for the two student groups over 3 years; the JHSSs had intercepts invariance and the VHSSs had residuals invariance.
Mean-level continuity was evaluated using MGCFA by incorporating a model for mean invariance. This model was used to examine average differences in the levels of the measured constructs across time. The results revealed that the fit for mean-level invariance was not good for neither the JHSSs nor the VHSSs. Four types of patterns may be observed when examining changes in mean-level continuity: A declining pattern, a rising pattern, a V-shaped pattern, and a flat pattern. In the VHSSs, the honesty-humility construct had a declining pattern over the 3 years. In addition, the honesty-humility and conscientiousness constructs for the JHSSs and the agreeableness construct for the VHSSs exhibited significant differences only in the first and third years. For the VHSSs, only the openness construct exhibited a rising pattern, with a significant difference discovered between the first and third years. The openness construct for the JHSSs had a V-shaped pattern, and the emotionality construct for the VHSSs had an inverted V-shaped pattern. Finally, the remaining dimensions in the two samples had flat patterns (i.e., nonsignificant mean-level difference over 3 years).
Rank-order continuity was analyzed using correlations between two time points (Roberts & DelVecchio, 2000). The correlation coefficients indicated a high level of effectiveness in capturing rank-order continuity between the same personality dimension in adjacent years for both JHSSs and VHSSs. Furthermore, when analyzing extraversion standardized by Z score, this study found that the correlation coefficient for VHSSs was significantly higher than that for JHSSs for two adjacent years. However, for emotionality, conscientiousness, and openness, the correlation coefficients for VHSSs were significantly higher than those for JHSSs only in the first and second years. Honesty-humility and agreeableness did not differ significantly between the two groups in adjacent years.
This study analyzed the change in personality scores of individuals between two time points by comparing the observed difference distribution with the theoretically expected distribution. With respect to JHSSs, the results revealed significant differences between the observed difference distributions of extraversion and agreeableness from the theoretically expected distributions in two adjacent years. In addition, significant differences in honesty-humility and emotionality were discovered between the second and third years and the first and second years, respectively. With respect to VHSSs, the results revealed significant differences between the observed difference distributions of honesty-humility, extraversion, agreeableness, and openness in the first and second year and in the second and third year from the expected difference distributions. Notably, the proportion of individuals with decreasing scores in honesty-humility was greater than those with rising scores.
Ipsative continuity, expressed using the Q correlation coefficient, is used to evaluate the continuity of the relative rankings of an individual's overall personality traits over time. A higher coefficient indicates greater similarity in the overall personality profile at the two time points. The average Q correlation coefficient for JHSSs was .65 and .72 in adjacent years. The average Q correlation coefficient for VHSSs was .70 and .74 in adjacent years. In addition, when comparing the Q correlation coefficient calculated from the actual data with that obtained from simulation samples with the same means and standard deviation, the Q correlation coefficient of the actual data was greater. This finding suggests a high level of ipsative continuity in both groups.
This study examined personality development in adolescence by evaluating both mean-level and rank-order continuity from the interindividual and intraindividual perspectives. Four indicators-mean-level continuity, rank-order continuity, individual-level change, and ipsative continuity-were analyzed in preadolescent (i.e., JHSS) and postadolescent (i.e., VHSS) samples. The mean-level continuity results revealed no clear changes in the dimensions except for in honesty-humility (decreasing pattern), which agrees with the results of a relevant study (Ashton & Lee, 2016). Adolescents may compete with peers to obtain resources during adolescence, resulting in less frequent cooperative and altruistic behaviors (Ashton & Lee, 2016). The correlation coefficient for rank-order continuity was greater than 0.5 for both 1-year and 2-year intervals. The results indicated a large individual-level change in conscientiousness; small individual-level changes were found in other dimensions, such as honesty-humility, emotionality, extraversion, and agreeableness. The average Q correlation coefficient of ipsative continuity in 2 years of preadolescence was .69, whereas that in 2 years of postadolescence was .72, indicating high ipsative stability in adolescence. The overall results of mean-level continuity and individual-level changes revealed changes within individuals over time but no significant changes in individuals for certain personality traits (e.g. emotionality).
The design of this study can be improved in several way. The use of different participants in the preadolescence and postadolescence phases created a disconnect in the development patterns, resulting in potential errors when explaining the development pattern of the adolescent period. In addition, the onset of adolescence varies; therefore, future studies may yield valuable insights by extending the duration over which individuals are tracked to when they graduate from college or enter the workforce.