February 2011 Regression Workshop

Prerequisites

Please be sure to come with a laptop with a recent version of [http://cran.r-project.org/ R] installed (i.e. 2.12.0 or higher). You should also have the languageR, Design, reshape, and ggplot2 packages and their dependencies installed.

n.b. A quick and easy way to do this at the R prompt:

install.packages(c("languageR", "Design", "reshape", "ggplot2"))

Location

Meliora 269 (Kresge) for 2/2 and Meliora 366 thereafter.

Schedule

Week 1

2/2: Introduction to R for regression (Alex Fine and Dave Kleinschmidt)

• What format should your data be in? How to get your data into the right format, how to import it into R?
• How to inspect variables, add/name/rename/delete variables?
• How to do simple data summaries in R?
• How to prepare your data for regression analysis, understanding your dependent and independent variables (histograms, outliers)
• Here are the datasets Dave and Alex will be working with: attachment:regdata (and more for Alex: attachment:alexdat)

• A lot of what we're going to talk about (plus more!) is covered in the first chapter of [http://www.ualberta.ca/~baayen/publications/baayenCUPstats.pdf Baayen (2008) [PDF]]. If you get bored, read through this and work through the exercises on page 20.

Week 2

2/7: Conceptual Introduction to linear and logistic regression with code examples in R (Florian Jaeger)

• what's regression: mathematical and geometric intro to ordinary regression
• how to run ordinary regressions in R
• conceptual introduction to multilevel/mixed, interpretation of random effects.
• what are the assumptions made by these models? How do they differ from and related to analysis of variance?

• slides: attachment:Rochester-IntroToRegressionModels.pdf and code: attachment:Code-Rochester-IntroToRegressionModels.R

2/9: Some practical issues in ordinary and mixed regression analyses (Florian Jaeger)

• collinearity: what is it? why is it a problem? what to do about it?
• understanding model comparison

Week 3

2/14: Linear and logistic regression examples (Alex Fine and Florian Jaeger)

• we'll go through some example data sets in more detail (please download this: attachment:exampledata.zip)
• you're also welcome to bring your own data to analyze
• this meeting will also serve to collect questions to be addressed in the remaining sessions

2/16: Visualization of your data with ggplot2 in R (Judith Degen and Florian Jaeger)

• the ggplot2 package provides a powerful way to create plots, ranging from simple bar and line plots, to predictions of non-linear models, smoothers, etc.
• we will go through some simple example, but also show how to use ggplot to visualize eye-tracking data or complex relationships between covariates for various types of dependent variables.
• some code that shows some of what you can do with the different geoms: attachment:ggplot_tutorial.R
• for above code to work, you will need to run these functions first: attachment:myFunctions.R
• slides from the first ggplot session (containing mostly some example plots): attachment:ggplot_tutorial.pdf

Week 4

2/21: Coding your contrasts to test the hypothesis you really mean to test (Maureen Gillespie, Northeastern University)attachment:fakedata.txt attachment:CodingLectureRochester.R

2/23: Relaxing the linearity assumption - within and beyond the generalized linear mixed model framework (Florian Jaeger)

• sometimes the linearity assumption of generalized linear mixed models is not warranted.
• what ways are there to test for non-linear trends in the data? what type of non-linearity can we include in our models within the generalized linear mixed models framework?
• what if there is a specific non-linear model we're interested in testing? non-linear mixed models provide a powerful extension to the generalized linear mixed models, but they come with a new set of challenges that we discuss in this session.

• Florian's slides: attachment:Rochester-CommonIssuesAndSolutionsWithRegressionModels.pdf

Week 5

Note: there is no session on Monday. The session planned for 2/28 is moved to 3/2

3/02: Non-linear mixed models (Ting Qian)

attachment:slides attachment:script attachment:data