What will I learn here?
Generalised Linear Models differ in their
conditional distribution β
not every outcome variable is normally distributed. This tool shows the 3D posterior landscape
P(y | x, Ξ²β, Ξ²β) for six GLM families: How does the distribution shape change when
the predictor x or the parameters vary?
Recommended exploration
- Click through all six model tabs β observe how the distribution shape changes depending on the outcome type
- Vary Ξ²β (Slope) β how does the conditional distribution change along the x-axis?
- For Poisson: increase Ξ²β β the distribution shifts and widens (Variance = Mean)
- For Neg. Binomial: vary dispersion r β how does it differ from Poisson under overdispersion?
- Rotate and zoom the 3D view β explore the landscape from different angles
The six GLM families
Normal (OLS): continuous data β symmetric bell curve, Ο constant
Poisson: count data (0,1,2,β¦) β log link, Variance = Mean
Neg. Binomial: count data with overdispersion β Variance > Mean
Gamma: positive continuous data β right-skewed, log link
Binomial: number of successes out of n trials β logit link
Logistic: binary 0/1 data β probability between 0 and 1
Further Reading
Dunn & Smyth (2018) β Generalized Linear Models with Examples in R
Lambert (2018) β accessible introduction to Bayesian thinking