Making sense of marginal effects
summary
Marginaleffects provides a single interface for 100+ model types in R and Python and demonstrates concrete workflows: use avg_comparisons to compute a lift (example: naive lift 7.5%), detect heterogeneity (desktop ≈ 11%, mobile ≈ −58%), correct sample-population bias with weights to get a revised estimate (example: −13%), and produce publication-ready figures via plot_predictions and plot_comparisons (functions return a ggplot object). Documentation and Stata-validated results accompany the package.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:05Clear comparisons in data science
- 3:35Probability vs odds in logistic regression
- 5:02marginaleffects package
- 6:27Calculating lift from A/B tests
- 8:58Group-specific comparisons
- 10:04Bias correction with weights
- 11:49Visualizing results
- 13:35Plotting lift for stakeholders
- 14:40Demo story
- 15:43Call to action
related talks
01h 19mModel criticism with ArviZ
15mInterpretable Machine Learning with R
42mInterpreting Black-Box Models with Applications to Healthcare
31mMedia Mix Modeling:How to Measure the Effectiveness of Advertising
39mBeyond Bias: Crafting Fairness in Pythonic Machine Learning Systems
42mManualRivas
34mInterpreting Machine Learning Models: Why and How!
01h 05mEvaluating Model Explainability Methods