Probabilities and Odds Ratios in Logistic Regression: What you can and what you can’t do

Course description

Following the debate in social science methodology about the ‘usefulness’ of Odds Ratio’s and Average Marginal Effects and many other quantities of interest when using Discrete Response Models (DRM) or Latent Variable Models (LVM), this course will provide knowledge on what you can and cannot do when interpreting DRM or LVM. We address the problems scholars encounter when analysing discrete outcomes, indicate which quantities of interest could be used, and under which circumstances one could use one or the other. All in all, the course indicates what you can and what you can’t do with DRM or LVM depending upon your research question.

Content

Day 1: refresh knowledge on description/prediction and causal inference, Setting the problem: interaction effects when examining ‘marginal effects’, OR’s, probits, probabilities (Linear Probability Models, Discrete Response Models, Latent Variable Models, Probit Models).
Day 2: Comparing effects of different variables within equations (heteroskedasticity and unequal thresholds; unobserved heterogeneity, mavericks; scaling effects, collapsing effects). Consequences and solutions.
Day 3: Comparing groups or time periods (interpreting interaction effects); Consequences of threshold differences and different scaling effects.
Day 4: Comparing effects between equations (spurious and mediation effects).

Learning Highlights

Students will learn pros and cons of using Linear Probability Models, logit, probit, and other models with categorical dependent variables, as well as interpretation of OR’s, probits, AME’s, MEM’s, Discrete and instantaneous changes, Standardized (full/half) effects, Maximum effects, ranking and ratio’s of quantities of interest. Moreover, students will learn the description of problems encountered when interpreting discrete response models and how to deal with them.

Target audience

The course will be useful to PhD students, postdocs and researchers interested in learning these methods.

Required training or equipment

Basic knowledge of multivariate regression techniques using Stata is needed.

Students must bring their own laptop with Stata programme installed.

Lecturer

Maike van Damme

Maike is a senior researcher in sociology at CED, working on social stratification and socio-demography in cross-national perspective. She has been teaching on models with discrete responses for many years and published on such models frequently.

Más información