Interpretable Statistical Models
Interpretable statistical models make their assumptions and relationships available for inspection.
Interpretability is not only a convenience for analysts. In consequential prediction settings, it can be a requirement for trust, review, and governance.
An interpretable model can help answer questions like:
- Which factors are influencing the prediction?
- How uncertain is the estimate?
- Does the direction of an effect make sense to domain experts?
- Which assumptions are doing the most work?
- What happens when new observations arrive?
These models are especially useful when paired with domain knowledge and an explicit theory of the data generating process.
Related: