面板数据处理方法lecture 1 statistical methods for the analysis of longitudinal data.ppt

Lecture 1(Chapter 1) Introduction This course describes statistical methods for the analysis of longitudinal data, with a strong emphasis on applications in the biological and health sciences Univariate statistics: each subject gives rise to a single measurement, termed response Multivariate statistics: each subject gives rise to a vector of measurements, or different responses Longitudinal data: each subject gives rise to a vector of measurements, but these represent the same response measured at a sequence of observation times Repeated responses over time on independent units (persons) Basic issues and exploratory analyses Definition and examples of LDA Approaches to LDA Exploring correlation Statistical methods for continuous measurements General Linear Model with correlated data Weighted Least Squares estimation Maximum Likelihood estimation Parametric models for covariance structure Generalized linear models for continuous/discrete responses Marginal Models Log Linear Model and Poisson Model for count responses Logistic model for binary responses GEE estimation methods Estimation techniques Random Effects Models (Multi-level models) Transition Models Longitudinal study: people are measured repeatedly over time Cross-sectional study: a single outcome is measured for each individual In LDA, we can investigate: changes over time within individuals (ageing effects) differences among people in their baseline levels (cohort effects) LDA requires special statistical methods because the set of observations on one subject tends to be inter-correlated Reading ability appears to be poorer among older children. Let’s assume that the reading ability of each child has been measured twice. Repeated observations are likely to be correlated, so assumption of independence is violated What if we used standard regression methods anyway (ignore correlation)? Correlation may be of scientific focus Incorrect inference Inefficient estima