Course No.：21509Z
Period：20
Credits：1
Course Category：Advanced Course
Aims & Requirements:
Many research studies generate data with a multi-level structure. For example, we conduct a survey on college-entrance exam scores of students, randomly sampled from the universities in Beijing. Since the same university tends to have similar collage-entrance exam scores, students within a single university will be expected to be less variable in scores than students drawn from different universities. Such the data are said to be multilevel or hierarchical. One important example of multilevel data is a longitudinal study with repeated measurements on subjects over time. The major advantage of longitudinal over cross-sectional studies is that the former can distinguish changes over time within subjects from the difference among subjects in their baseline levels.
The main issue in the analysis of multilevel data is possible correlation of subjects in the same cluster. Failure of accounting for this correlation can lead to invalid statistical inferences. I will introduce three models for handling clustered or longitudinal data: marginal models, random effects models, and transition models for longitudinal data and conditional fixed-effects models for clustered data. Emphasis will be placed on understanding key assumptions in these methods and on using existing statistical software packages for implementing those methods.
Primary Coverage：
Lecture 1. Introduction
Lecture 2. Two-level linear models
Lecture 3. Three-level linear models
Lecture 4. Multi-level models for cross-nested data
Lecture 5. General linear models for longitudinal continuous outcomes
Lecture 6. Generalized linear models for longitudinal binary data
Lecture 7. Review and exam
Author：Xiaohua Zhao
