透過您的圖書館登入
IP:3.138.33.87
  • 學位論文

群聚資料下混合效應模式的半參數化貝氏分析

Semiparametric Bayesian Analysis of Mixed Models for Clustered data

指導教授 : 蕭朱杏

摘要


這篇論文的內容是針對群聚資料(包括了長期追蹤資料),提出半參數化混合效應模式(semiparametric mixed models)的完整貝氏分析方法。所討論的半參數化模式包含了相加性混合效應模式(additive mixed models)與變化係數混合效應模式(varying-coefficient mixed models)。我利用懲罰性節點法(penalized spline)來發展無參數迴歸函數的事前分布以進行貝氏分析。所提出的事前分布有別於目前文獻中其他形式的事前分布,主要差異就是它是可積分的(proper);而且不像平滑節點法(smoothing spline)需要利用樣本中不重複的所有解釋變數值來決定節點數目與位置,我所提出的是由分析者自己決定節點數目與位置。這兩個差異使得所提出的方法具有計算上的優勢,特別是處理樣本數很大的資料集,也因為可積分的性質使得模式選取可以利用比較不同模式的邊際概似函數來達成。此外,相較於古典統計學派的方法,所提出的分析法可比較容易的延伸到非連續的資料結構上,並不需要依賴特異的平滑參數估計準則。模式參數的估計過程是利用吉布斯抽樣法(Gibbs sampler)模擬產生事後分布的樣本以進行推論。我仔細的討論不同資料結構下的詳細抽樣步驟,這包括了連續、二元與計數資料。特別是計數資料由於不具有事前與事後分布的共軛特性(conjugate),我提出了更有效率的更新過程以完成吉布斯抽樣法。模擬研究顯示所提出的貝氏分析法能夠正確的估計出無參數迴歸函數的趨勢。除了考慮估計的問題外,更進一步也討論了模式選取的問題。主要是探討半參數化與參數化模式的邊際概似函數估計方法,因而使得所進行的模式比較不再侷限於巢式(nested)的關係。最後,我將所提出的方法應用到AIDS世代研究(Kaslow et al. 1987)的例子上。分析結果顯示會影響HIV感染者體內CD4+細胞數目的解釋變數是感染者感染前的CD4+細胞數目,而HIV感染者體內CD4+細胞數目隨時間變化的趨勢是不斷線性遞減的。在同時考量參數化與半參數化模式後,參數化的線性混合效應模式是資料較支持的可能模式。

並列摘要


In this thesis I consider Bayesian semi-parametric analysis of mixed-effects models for clustered data. Particularly, I consider the additive mixed model and varying-coefficient mixed model, and use nonparametric arbitrary smooth functions to represent the covariate effects. I model the nonparametric functions using the qth-degree polynomial penalized splines with fixed knots, and specify the prior for the corresponding smoothing parameter of each function. A computationally efficient Markov chain Monte Carlo (MCMC) algorithm is proposed to simulate posterior samples for inference. In addition to the continuous response setting, the binary and count data are also considered and discussed in detail. Special attention is necessary due to the non-conjugacy for binary data with logit link and count data with log link. I also develop a modified Metropolis-Hastings algorithms to mix the Markov chain and increase the speed. The simulation studies show that the posterior mean via nonparametric approach captures well the true functional forms. In addition to the estimation, I also address the problem of model choice between the competing parametric and semi-parametric specifications using marginal likelihoods and Bayes factors. Finally, the data of multicenter AIDS cohort study (Kaslow et al. 1987) are considered for illustration.

參考文獻


Albert, J. H., and Chib, S. (1993), “Bayesian Analysis of Binary and Polychotomous Response Data,” Journal of the American Statistical Association, 88, 669-679.
Besag, J. E. (1989), “A Candidate’s Formula: A Curious Result in Bayesian Prediction,” Biometrika, 76, 183.
Bliss, C. (1934), “The Method of Probits,” Science, 79, 38-39.
Breslow, N. E., and Clayton, D. G. (1993), “Approximate Inference in Generalized Linear Mixed Models,” Journal of the American Statistical Association, 88, 9-25.
Chiang, C.-T., Rice, J. A., and Wu, C. O. (2001), “Smoothing Spline Estimation for Varying Coefficient Models With Repeatedly Measured Dependent Variables,” Journal of the American Statistical Association, 96, 605-619.

延伸閱讀