[[alternative]]Development of Nonlinear Mixed-Effects Modelling Techniques for Pavement Data Analysis

計畫編號：NSC101-2221-E032-033 研究期間：201208~201307 研究經費：637,000[[abstract]]在各科學領域的研究資料很多是屬於多層次資料，分析此類的資料通常採用多層次線性混合效應 模式。因為資料的結構具有層級性，多層次模式的資料探索分析、統計模式的建立及模式評估比標準 複迴歸複雜。本研究擬以過去多年在預測模式構建研究方面之基礎，計畫以二年二期的方式，更進一 步地發展一套系統化的「非線性混合效應模式」(Nonlinear Mixed-Effects Models, NLME)構建技術與標 準分析流程，並擬利用美國AASHO 道路試驗的原始柔性鋪面資料，配合視覺圖法、線性與非線性混 合效應模式之分析與應用，深入探討現有鋪面設計法中有關標準軸重當量(ESAL)觀念之適用性，並擬 與非線性迴歸技術與當代迴歸技術之結果相比較。 研究中亦將選用數個多層次資料來描述如何應用此發展出來的視覺圖技巧。經由應用此視覺圖技 巧，研究者將能回答在探索多層次資料及建立多層次模式研究中常被重視的問題，如：描述個人層及 群體層的原型(線性或非線性)，辨識重要的預測變數及異常的受試者，選擇合適的統計模式及隨機效 果的共變模式，建議可能的殘差共變模式及模式的評估與檢測。主要的研究內容包括： 1. 國內外相關研究之文獻蒐集與整理。 2. 柔性鋪面道路試驗資料擷取與公式應用。 3. 利用非線性迴歸技術與當代迴歸技術來建立預估模式。 4. 利用視覺圖法來協助資料探索分析。 5. 利用線性混合效應模式來建立預估模式。 6. 利用非線性混合效應模式來建立預估模式。 7. 利用敏感度分析來協助前述模式之評估與驗證。 8. 系統化分析流程之彙整與應用。 9. 檢視標準軸重當量之觀念並探討其適用性。 10. 加入季節性監測資料與時間序列分析之可行性研究。[[abstract]]Multilevel data are very common in many fields. Hierarchical Linear Models (HLMs) or Linear Mixed-Effects (LMEs) models are often utilized to analyze multilevel data. Because of the hierarchy of data structure, the exploratory analysis, statistical modeling, and examination of model-fit of multilevel data are more complicated than those of standard multiple regressions. Based on the past experience in the development of modeling techniques, this study is focused on the development of Nonlinear Mixed-Effects Modeling techniques for aiding multilevel pavement data analysis. The entire project consists of two phases to be completed within two years. The main focus of this study is to develop a systematic modeling approach using visual-graphical techniques, LMEs, and Nonlinear Mixed-Effects Models (NLMEs) for the analysis of the original AASHO Road Test flexible pavement data. Different predictive models using nonlinear regression and modern regression techniques will also be developed for comparison purposes. The proposed visual-graphical techniques will be demonstrated via several multilevel data. From the application of the proposed methods, investigators can answer the research questions that most addressed in multilevel studies. These questions include characterizing or describing the patterns at both the group and individual level, identifying the important predictors and unusual subjects, choosing suitable statistical models, selecting random-effects structures, suggesting possible residuals covariance models, and examining the model-fits. Subsequently, the applicability of the well-known 18-kip equivalent single axle load (ESAL) concept will be further examined. The major tasks include: 1. Literature review of LMEs, NLMEs, and visual graphical methods. 2. Preparation the original AASHO Road Test flexible pavement data and the analysis of the existing design equation. 3. Models development using nonlinear regression and modern regression techniques. 4. Conducting exploratory data analysis using visual graphical methods. 5. Models development using Linear Mixed-Effects Models. 6. Models development using Nonlinear Mixed-Effects Models. 7. Conducting sensitivity analysis and examination of the model-fit 8. Integration of the proposed systematic NLMEs modeling approach. 9. Reexamine the applicability of the 18-kip ESAL concept. 10. Conducting feasibility study of integrating LTPP seasonal monitoring data and time series analysis for future study.[[sponsorship]]行政院國家科學委員