SCS 2017: Longitudinal and Nested Data

From Wiki1

(Redirected from SCS Courses 2017)
Jump to: navigation, search
At their best, graphics are instruments for reasoning. — Edward Tufte
Where there is no uncertainty, there cannot be truth. — Richard Feynman

This is the home page for the SCS short course on Modeling and Analysis of Longitudinal and Nested Data offered Wednesday evenings from 6pm to 9pm from March 1 to March 29, 2017 in York Lanes 213.

Contents

Quick Links

Calendar

Day Files and Links
Day 1

March 1

Ideas in Regression: Why Models Matter

This is a 'review' of fundamental ideas in regression that are helpful and relevant to understand the role of longitudinal and nested data in scientific inference.

Getting Started with R and R Studio

Day 2

March 8

Linear Models for Nested Data with Normal Error

Part 1: Hierarchical to Mixed Models
Lab_1.R Copy this R script to R Studio and work your way through it.

Linear Models for Longitudinal Data with Normal Error

Longitudinal Models
Lab_2.R exploring longitudinal models
Day 3

March 15

Linear Models for Longitudinal Data with Normal Error (cont'd)

Generalized Splines

Non-Linear Models for Longitudinal Data with Normal Error

Non-Linear Models: Asymptotic Functions of Times

Introduction to Bayesian Approaches: MCMC and HMC

Bayesian vs Frequentist Inference: Statistics on Trial(s)

Installing Rstan for HMC with STAN

Some links for Stan
The 'rstan' package was 'moved to CRAN' a few months ago. This should make installation much easier but almost all the information on the web still refers to pre-CRAN installation.
Vignette for Stan in R
Getting Started: installation -- only partially updated, I think, but should work.

See the documentation site for Stan and download the pdf manual. It will look forbidding at first but it's remarkably comprehensive and easy to use once you learn to navigate in it.
Late Late Show

Musings on R:
Working with messy data | R script
Day 4

March 22


Using STAN for Linear Longitudinal Models with Normal Error

Day 5

Introduction to Stan

Models for TBI:
Recovery after TBI | r script | stan script 1 | stan script 2 | stan script 3 | stan script 3b | stan script 3c | stan script 4: multivariate model

Various models for TBI.

Shortitudinal Data
Shortitudinal Data | r script | stan script 1 | stan script 2: Gamma prior on small sigma

Correcting the bias in the estimate of compositional/contextual effects with small cluster size. This is a simple example illustrating how Stan can be used to specify models with latent predictor variables.

Information Criteria
This is an evolving topic. For a while WAIC (Watanabe-Akaike Information Criterion) was considered the state of the art for Bayesian models but perhaps this is being overtaken by the 'loo' package that works specifically with Stan. There are good readable article on it.

Notes

Personal tools