LOT Summer School 2018

Introduction to nonlinear regression modeling

Jacolien van Rij

Contact

Introduction to nonlinear regression modeling

Teacher: Dr. Jacolien van Rij

Contact

Email address: j.c.van.rij@rug.nl

http://website teacher/: www.jacolienvanrij.com

Course info

Level: advanced

Participants are expected to have experience with the software R statistics: how to import data, manipulate data frames, and some basic knowledge of plotting. Participants are also expected to have practical experience with linear mixed-effects modeling (e.g., using the lme4 library in R).

Students need to bring a laptop with R installed as well as the most recent version of the packages mgcv and itsadug. (Please contact the organizers of the LOT School if you cannot bring a laptop.)

Course description:

This course will provide a hands-on introduction to Generalized Additive Mixed-Effects Modeling (GAMM; Hastie & Tibshirani, 1990; Wood, 2017), a nonlinear regression method.

When analyzing psychological or psycholinguistic data, we often deal with nonlinear trends in our data. For example, the accuracy on a picture selection task may increase nonlinearly with age. The effects of stimulus properties such as word length, frequency, or semantic similarity can be nonlinearly related to the (log-transformed) reaction times. The EEG amplitude changes nonlinearly over time as response to linguistic stimuli. For analyzing these types of data, GAMMs may be more suited than the commonly used linear mixed-effects regression models, because GAMMs can include complex nonlinear regression lines and nonlinear interaction surfaces. To account for the variation due to participants and items nonlinear random effects can be included.

In the course, I will introduce the basic syntax, but also methods to identify the best model given the data, and demonstrate how to visualize non-linear effects and non-linear interactions. In addition, I will address several potential problems, including, but not limited to, dealing with the common problem for time series data of encountering autocorrelation in the residuals of a model.

In the first part of each lecture, the theory is explained by using real experimental data to illustrate the potential of GAMs for several types of data, including eye tracking (gaze) data, pupil dilation data, and ERP (EEG) data, and behavioral data. Four practical assignments are part of the course. The aim of these assignments is to get practical experience with the method by analyzing a comparable data set on their own laptop. In the second part of each lecture will consist of a lab session in which students can get started with the assignments. In addition, practical issues and questions that arise from the lab sessions be discussed.

Participants of the course are invited to bring their own time series data to see whether this method can help to answer their research questions.

Day-to-day program

Monday: Introduction Generalized Additive Modeling

Tuesday: Linear and nonlinear random effects

Wednesday: Model criticism and model selection

Thursday: Nonlinear interactions

Friday: Interpretation and visualization

Reading list

TBA