Foundations and Utility of Generalized Linear Models

Description du service

Note: Registration for this session is not yet open. In the social sciences, we are often faced with the reality that our data and the variables we use do not fit neatly into standard linear models that predict simple straight lines or normally distributed errors. To accommodate such data, better understand relationships between variables, and answer the big questions we want to ask, generalized linear models are introduced. For historical reasons, such models in the social sciences are referred to as “maximum likelihood models”. Through his review of the foundations of the linear model, Dr. Gill will build on participants’ common understanding of the standard linear model, by introducing and elaborating on the basics of generalized linear models. A strong theoretical basis for these models will be presented and explained and distinctions drawn that highlight how the standard linear model can be adapted to accommodate a broader class of outcomes (including counts, dichotomous outcomes, bounded variables and more). Dr. Gill will guide participants towards a clearer understanding of the contexts in which different regression types should be considered and different models employed so that participants can make more informed analytical choices to better understand their data and interpret their findings. The workshop will close with a brief introduction to traditional machine learning and deep learning. Dr. Gill will explain how maximum likelihood (or generalized linear) models can feed machine learning and how these models compare to the neurons forming deep learning neural networks. He will also discuss the human input components of each type of learning, and the potential errors such models can produce as a result of the way they are structured. Prerequisites: basic statistics, some linear modeling, some matrix algebra. For additional details see: https://www.kindinstitute.net/workshop-outlines/gill-ii-january-2027.