:::

最新消息

:::

活動公告
  • 標題:中研院社會所於9月8日(五)14:30~16:30邀請彰師大王郁琮教授演講
  • 公告日期:2017-08-29

 

 講 題:Opportunities and Challenges of Reanalyzing Old Data with New Perspectives: Applications of Longitudinal Mixture Models to the Taiwan Youth Project

主講人:王郁琮教授 (國立彰化師範大學)
時 間:1060908日(星期五)下午230  - 下午430
地 點:社會所802會議室 
~
自由入席,毋需事先報名~


演講摘要:
Conventional longitudinal statistics (e.g., latent growth curve model) suffers from its unrealistic assumption on homogeneous development, frequently violation of normal distribution of growth parameters, oversimplifying the latent structure of developmental trajectories and lack of model-based classification. The newly developed mixture modeling overcomes these obstacles and provides an opportunity to investigate the 
cause and effect of heterogeneity of developmental trajectories. An unique feature of the longitudinal mixture modeling is that it allows one to explore the latent structure underlying the growth parameters (i.e., initial status and growth rates) as well as estimating different forms of trajectory (i.e., linear and nonlinear) simultaneously. Three empirical applications are shown, for demonstration purpose, as follows: 

Study I demonstrates how to capture the latent heterogeneity among development trajectories of adolescents
 depressive symptoms, analyze gender differences in the depression trajectory, and investigate relationships between types of depression and the odds of a conduct disorder a year later using growth mixture model (GMM). Results showed that depression trajectories are classified three ways, namely: typical depression with a high-low-high trajectory, cumulative depression with a lowhigh-high pattern, and emotionally stable with low scores across the 3 years of the study. Typical depression boys demonstrated more dramatic fluctuations than their counterparts and emotionally stable girls demonstrated higher initial depression scores than boys. Cumulative depression adolescents are at a much higher risk for conduct disorders than those with typical depression or those who are emotionally stable

Study II examines the relationship between heterogeneous parenting styles and the developments of depressive symptoms among adolescents using latent class analysis adjoin with growth mixture model. Results from the general growth mixture modeling (GGMM) revealed a nonlinear increase in the intensity of depressive symptoms between early and middle adolescence. More pronounced depressive symptoms in earlier years were also shown to be associated with more rapid development of similar symptoms later in adolescence. Perceived parenting styles, as manifest in parental warmth and harsh discipline, were categorized into 4 latent heterogeneous classes: attentive, reserved, austere, and conflicting. Adolescents living under austere parenting tend to report the most pronounced depressive symptoms from early to middle adolescence; however, the development of symptoms in the austere group was the slowest. 

Study III investigates the discontinuity of development of depressive moods from early adolescence to late adolescents and to early adulthood. Eight-wave depressive symptom data were analyzed, using the three-step latent transition growth mixture model (LT-GMM). Results show that a three-class solution fit the data well for both junior high school (termed high-improving, cumulative, and low-stable) and senior high school (termed heightening, moderate-stable, and low-stable) stages. The depressive symptoms of most individuals maintained at a low level (i.e., low-stable) throughout adolescence and early adulthood; however, nearly a quarter of the adolescents reported depressive symptoms that were moderately or highly severe in senior high school and beyond. Finally, more than 30% of the participants presented different developmental trajectories between junior and senior high school. These results help elucidate the heterogeneity and fluidity associated with the development of depressive symptoms between early adolescence and early adulthood in light of stressful life events in Chinese culture. 

In conclusion, longitudinal mixture modeling is a very comprehensive and sophisticated technique. Our experiences suggest credible prior information of trajectory parameters is crucial to the success of longitudinal mixture modeling. Incorporating significant predictors and valid outcome variables in the full model will facilitate the stability of the estimation. Other alternatives are pattern mixture modeling for missing data analysis and dual trajectories mixture for co-occurring development. Dynamic structural equation modeling (DSEM) and spectral analysis are for intensive longitudinal data.

講者簡介:
王郁琮,美國南加州大學計量心理與統計博士,曾於1997年至2005年於美國華盛頓特區政府智庫(Westat)擔任研究員;2014年至2016年擔任台灣統計方法協會理事長。自20058月起任教於國立彰化師範大學輔導與諮商學系至今。研究專長為青少年心理學、混合模式(Mixture Models)、教學評鑑、學習成效評量、試題差異反應研究。

  • 附件檔案: 海報 JPG
  • 參考連結:
  • 張貼人:105254501
  • 最後修改時間:2017-08-29 PM 2:52

cron web_use_log