帳號:guest(18.191.150.192)          離開系統
字體大小: 字級放大   字級縮小   預設字形  

詳目顯示

以作者查詢圖書館館藏以作者查詢臺灣博碩士論文系統以作者查詢全國書目
作者(中文):盧永騰
作者(外文):Lu, Yung-Teng
論文名稱(中文):多重中介因子模型下的高適應性中介效應分析
論文名稱(外文):Adaptive mediation analysis under multiple mediator scenarios
指導教授(中文):謝文萍
林聖軒
指導教授(外文):Hsieh, Wen-Ping
Lin, Sheng-Hsuan
口試委員(中文):黃彥棕
戴安順
口試委員(外文):Huang, Yen-Tsung
Tai, An-Shun
學位類別:碩士
校院名稱:國立清華大學
系所名稱:統計學研究所
學號:109024512
出版年(民國):111
畢業學年度:111
語文別:中文
論文頁數:65
中文關鍵詞:因果推論中介效應分析穩健估計機器學習
外文關鍵詞:Causal InferenceMediation analysisRobust estimationMachine Learning
相關次數:
  • 推薦推薦:0
  • 點閱點閱:87
  • 評分評分:*****
  • 下載下載:0
  • 收藏收藏:0
自然中介效應(natural indirect effect)與自然直接效應(natural directly effect)是因果推論領域中常常被用來量化某個中介因子對一段因果關係中介程度的估計量。本文考慮到當今因果推論會觸碰的資料中,不乏多維度、高維度甚至結構化的資料,而當該複雜型態的資料被考量為中介因子時,若需做中介效應的探討,將會遇到中介因子的預測模型難以建立,以及真實資料的性質難以被正確配適等問題。鑒於此,考量傳統的估計方法盡數使用線性相關的估計手段,本文嘗試結合機器學習相關的非線性估計方法,進行複雜資料型態的預測。本文進行數個模擬研究,假設在真實中介因子維度不多時,嘗試維度縮減(dimension reduction)是否能還原出相對於估計框架更可運用的有關中介因子的資訊,並基於模擬研究的結論,結合上述的機器學習估計方法、維度縮減,以及考慮到模型配適情形可能不理想時可以使用的多重穩健(multiple robust estimator)估計方法中Mark van der Laan等人提出的TMLE(targeted maximum likelihood estimation),提出一套複雜中介因子的中介效應分析問題下的一個建議的估計方式。
Natural indirect effect and natural direct effect are often used in the field of causal inference to quantify mediation effects in causal relationship. The thesis focus on the situations of multivariate, high-dimensional and even structured data were considered as mediators, which problems such as difficulty in establishing the prediction model of the mediators and difficulty in correctly specified in the real data were raised. In comparison with traditional estimation methods mostly use linear approaches, the author attempts to utilize the nonlinear estimation methods, also machine learning algorithms to handle the problems induced by the complex data. Also, we conducts several simulation studies, assuming that when there are not many dimensions of true mediators, try to determine whether dimension reduction can restore more applicable information about mediators. And in consideration of the multiple robust estimation methods that can be used when the models were not all successfully adapted the data, Mark van der Laan et al., proposed the TMLE (targeted maximum likelihood estimation) as a estimation framework, which were also utilized in the suggested estimation methods under the mediation effect analysis problem of complex mediators we obtained.
目錄
1 緒論 1
1.1 因果推論以及中介效應分析---------------------- 1
1.2 現代資料之於傳統分析方法---------------------- 2
1.3 適應性強的中介效應分析架構--------------------- 2
2 研究背景及相關回顧 3
2.1 單一中介因子模型與因果參數估計------------------ 3
2.2 多重中介因子模型下的因果參數估計----------------- 4
2.3 多重穩健估計------------------------------ 4
2.4 機器學習方法以及多重穩健估計方法應用於中介效應分析----- 5
2.4.1 機器學習方法-------------------------- 6
2.4.2 穩健的估計方式之間結合的發展方向------------- 9
3 研究方法 12
3.1 單中介因子分析模型-------------------------- 12
3.2 多重中介因子分析模型------------------------- 15
3.3 多重中介因子下的傳統估計法:G-computation----------- 17
3.4 中介因子分析的穩健估計法:TMLE----------------- 18
3.4.1 單中介因子的 TMLE 流程------------------ 20
3.4.2 多重中介因子的 TMLE 流程----------------- 22
3.5 整合機器學習的新中介效應估計框架----------------- 24
3.5.1 基於機器學習的 G-computation 框架------------- 24
3.5.2 基於機器學習的 TMLE-------------------- 25
4 模擬研究 27
4.1 模擬設定 — 觀測中介因子----------------------- 27
4.2 模擬環境假設與資料生成方法--------------------- 28
4.3 觀察到的中介因子為真實中介因子的線性映射------------ 29
4.3.1 單變數中介及觀察中介因子------------------ 29
4.3.2 複數變數中介及複數觀察中介因子-------------- 30
4.3.3 複數變數中介及多觀察中介因子--------------- 31
4.3.4 複數變數中介及高維度觀察中介因子------------- 32
4.4 觀察到的中介因子為真實中介因子的非線性映射---------- 32
4.4.1 觀察到的中介因子為真實中介因子的非線性映射:情境一- 36
4.4.2 觀察到的中介因子為真實中介因子的非線性映射:情境二- 36
4.4.3 小結------------------------------- 37
4.5 模擬研究—結論----------------------------- 44
5 資料分析 45
5.1 資料描述-------------------------------- 45
5.2 分析方法-------------------------------- 46
5.3 分析結果-------------------------------- 47
6 討論 48
7 結論 49
參考文獻 50
A 附錄:單一中介因子下的效應辨識 52
B 附錄:多重中介因子模型下的效應辨識 53
C 附錄:單一中介因子下自然間接效應的 TMLE 方法 54
D 附錄:多重中介因子下自然間接效應的 TMLE 方法 56
E 附錄:估計部分主程式碼 58
F 附錄:模擬部分主程式碼 61
1. Fix, E. and Hodges, J. L. (1989). Discriminatory analysis. nonparametric dis-
crimination: Consistency properties. International Statistical Review/Revue
Internationale de Statistique, 57(3):238–247.
2. Glynn, A. N. and Quinn, K. M. (2010). An introduction to the augmented inverse
propensity weighted estimator. Polit. Anal., 18(1):36–56.
3. Ho, T. K. (1998). The random subspace method for constructing decision forests.
IEEE transactions on pattern analysis and machine intelligence, 20(8):832–844.
4. Hoerl, A. E. and Kennard, R. W. (1970). Ridge regression: Biased estimation for
nonorthogonal problems. Technometrics, 12(1):55–67.
5. Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., and Liu, T.-Y.
(2017). Lightgbm: A highly efficient gradient boosting decision tree. Advances
in neural information processing systems, 30.
6, Lin, S.-H. and VanderWeele, T. (2017). Interventional approach for path-specific
effects. J. Causal Inference, 5(1).
7. Robins, J. M. and Greenland, S. (1992). Identifiability and exchangeability for
direct and indirect effects. Epidemiology, 3(2):143–155.
8. Roweis, S. T. and Saul, L. K. (2000). Nonlinear dimensionality reduction by locally
linear embedding. Science, 290(5500):2323–2326.
9. Tenenbaum, J. B., de Silva, V., and Langford, J. C. (2000). A global geometric
framework for nonlinear dimensionality reduction. Science, 290(5500):2319–
2323.
10. Van der Laan, M. J., Polley, E. C., and Hubbard, A. E. (2007). Super learner.
Statistical applications in genetics and molecular biology, 6(1)
11. Van Der Laan, M. J. and Rubin, D. (2006). Targeted maximum likelihood learning.
The international journal of biostatistics, 2(1).
12. VanderWeele, T. J. and Vansteelandt, S. (2014). Mediation analysis with multiple
mediators. Epidemiol. Method., 2(1):95–115.
13. Wold, S., Sjöström, M., and Eriksson, L. (2001). Pls-regression: a basic tool of
chemometrics. Chemometrics and intelligent laboratory systems, 58(2):109–130.
14. Zheng, W. and van der Laan, M. J. (2012). Targeted maximum likelihood estima-
tion of natural direct effects. Int. J. Biostat., 8(1)
 
 
 
 
第一頁 上一頁 下一頁 最後一頁 top
* *