liveBook · Manning

Bayesian model

As explained in Causal AI

Bayesian modeling is the construction and analysis of probabilistic models using prior distributions, likelihoods, and posterior distributions.

When a promotion might help a customer or make retention less likely, causal assumptions such as monotonicity can narrow the possibilities without fully resolving the counterfactual. section 10.6.2

A Bayesian model handles this uncertainty by placing a prior over structural causal models (SCMs), updating it with observational and interventional data, and producing a posterior over counterfactual inferences. section 10.6.2

A Bayesian workflow for an incompletely identified counterfactual is:

  • Specify diverse SCMs consistent with the causal assumptions. section 10.6.2
  • Assign higher prior probability to more plausible models. section 10.6.2
  • Update the models with level 1 observational and level 2 interventional data. section 10.6.2
  • Sample posterior SCMs and run counterfactual inference for each sample. section 10.6.2

How Causal AI covers it

How deep it goes
Specifies a prior-to-posterior SCM workflow and sampling procedure, without detailing posterior computation or a general identification algorithm.
Examples
Uses a promotion-retention case involving persuadable, lost-cause, and sleeping-dog customers.
What it assumes
Assumes causal assumptions plus level 1 and 2 data, while introducing probability as a way to represent identification uncertainty.
What you can do after
Evaluate/understand counterfactual uncertainty by sampling plausible SCMs after updating them with observational and interventional data.
How it connects
Connects causal identification with counterfactual inference, structural causal models, monotonicity, observational data, and interventional data.

Also explained in

Compare how 3 books explain Bayesian model →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Causal AI. The section links point to the parts of the book it draws on.