Introduction to the Generalized Linear Model (GLM)

An introduction to the foundation, estimation, and interpretation of generalized linear models, which are a general form of regression that can handle nonlinear data. This workshop is designed to teach researchers with only a foundation in regression analysis how to analyze, fit, interpret, and visualize models with nonlinear outcomes. Learn logistic regression, count regression, zero-inflated and hurdle regressions, and more!

Instructors:
Kevin M. King (University of Washington)

Workshop Dates and Times:
Recorded on July 24-26, 2024
Now available for self-paced learning

Workshop Format:
Three-Day Online Workshop Recording

Many researchers study outcomes that are not linear. Lots of outcomes are binaries (such as whether or not a participant meets diagnostic criteria), other outcomes are counts (such as how many symptoms of depression a participant meets criteria for), while yet others are skewed, zero inflated, or ordered or even non-ordered categories. Most foundational models in the social, behavioral, and medical sciences assume that the outcome (technically its residuals) is normally distributed. Yet treating nonlinear data as if it were linear can have serious consequences for statistical inferences and for prediction. Fortunately, there are relatively simple ways to account for and explicitly model nonlinearity. This course is designed to introduce researchers to ways of handling data that are not normally distributed.

The focus is on the Generalized Linear Model, which is basically regression with a twist. We’ll learn about the nature of different nonlinear distributions, how they’re estimated in the GLM, how to think about inference, and how to interpret and visualize parameter estimates in the context of nonlinearity. Because the GLM is the foundation of many more advanced statistical methods, the lessons of this course are generalizable to many other modeling frameworks including SEM and MLM, but this course is designed to be accessible for researchers who only have training in regression.

What you’ll learn

  • Foundations – Learn about different nonlinear distributions (binaries, counts, categories) and how they can be predicted using the GLM framework.

  • Implementation – Learn how to develop and estimate GLMs in R, with a focus on model testing, interpretation, uncertainty, and inference.

  • Interpretation and Visualization – Learn how to explain and visualize your focal effects of interest in plain language that your audience can understand.

Syllabus

Day 1

  • Overview

  • Review of Linear Regression

  • Introduction to Maximum Likelihood

  • Understanding non-normal distributions: binaries, counts.

  • Why can’t I just slap a robust estimator on my model and call it a day?

Day 2

  • Introduction to the Generalized Linear Model

  • Logistic regression: predicting a binary outcome

  • Ordinal and Multinomial regression: several logistic regressions in a trench coat

  • Poisson and negative binomial: predicting counts of thins

  • What do you do with extra zeros? Hurdles and zero-inflated models

Day 3

  • Model fit.

  • Marginal effects and counterfactuals

  • Visualization and model interpretation

  • Simulation and statistical power

  • Generalizing to other frameworks (MLM/SEM)

Registration Options

Introduction to the Generalized Linear Model

  • Professional
  • $824
  • Baseline Price for Faculty,
    Staff, and Other Professionals
  • Click Register Below
  • Trainee
  • $824 $549
  • 33% Discount for
    Students and Postdocs
  • Use code "TRAINEE" at Checkout

Note: All registration options include access to video recordings of the workshop for continued access and for self-paced learning for those who cannot attend live. Recordings will be uploaded to the workshop page shortly after each day ends.

 FAQs

  • Investigators who want to learn how to correctly model, visualize, and interpret data that isn’t normally distributed.

  • Beginner

  • Prior training in linear regression analysis is required.

  • R will be emphasized but most models can be fit in MPlus, SPSS, SAS, STATA, and most learning can be generalized to those packages/programs.

  • Registered learners will have access to the Zoom link and passcode to attend the synchronous workshop, as well as video recordings of the lectures, downloadable slideshows, example data files, example syntax files, and other associated materials.