Three Techniques for Dealing with Overdispersion in Generalized Linear Models

16 November 2015

1. Use a more flexible distribution

Instead of Poisson, use a Negative Binomial

2. Use quasi-likelihood

The dispersion parameter can be estimated from the Pearson’s Chi-Squared Statistic

3. Use a robust variance estimator

This is also known as a “sandwich” estimator

Recent Posts

• How to do linear regression in Excel using LINEST
• Posterior Predictive Checks
• Benchmarking R, Python, and Julia
• Machine Learning Textbooks
• Variational Lower Bounds for Convex Functions