Longitudinal Data Analysis, Including Categorical Outcomes
This workshop will focus on the analysis of longitudinal data, also known as "panel data." In either case, the data consist of repeated observations over time on the same units. The approach will use mixed models. Models for continuous outcomes will first be presented, including description of the multilevel or hierarchical representation of the model. Use of polynomials for expressing change across time, treatment of time-invariant and time-varying covariates, and modeling of the variance-covariance structure of the longitudinal outcomes will be described.
For dichotomous, ordinal and nominal outcomes, this workshop will focus next on the mixed logistic regression model and generalizations of it. Specifically, the following models will be described: mixed logistic regression for dichotomous outcomes, mixed logistic regression for nominal outcomes, and mixed proportional odds and non-proportional odds models for ordinal outcomes. The latter models are useful because the proportional odds assumption of equal covariate effects across the cumulative logits of the model is often unreasonable.
Finally, missing data issues will be covered. Mixed models allow incomplete data across time and assume that these missing observations are "missing at random" (MAR) under maximum likelihood estimation. Approaches that can go further, and don't necessarily assume MAR, are through the use of pattern mixture and selection models. Applications will be described of mixed pattern mixture and selection models.
In all cases, methods will be illustrated using software. Stata will be used for most examples, with some use of SAS, and SuperMix for categorical outcomes.
Prerequisites: Participants should be thoroughly familiar with linear regression and have some knowledge of logistic regression.
Fee: Members = $1700; Non-members = $3200