Front-door estimand

As explained in Causal AI

A front-door adjustment estimand is a causal effect quantity identified through a mediator that satisfies front-door identification conditions.

A front-door estimand identifies the causal effect of a treatment on an outcome through an observed mediator. It separates causal influence from association when treatment and outcome may share unmeasured causes. section 10.5 section 10.5.2

Side-quests, won items, and purchases

Side-quest engagement (E) affects won items (W), and won items affect in-game purchases (I). Guild membership (G) also influences engagement and purchases, creating a backdoor relationship. The mediator W helps identify the effect of E on I without requiring adjustment for every common cause. section 10.5.1 section 10.5.2

Check these identification conditions before treating the estimand as usable:

• Require full mediation: the mediator intercepts every directed causal path from treatment to outcome. section 11.3.3
• Require first-stage unconfoundedness: no hidden confounder jointly affects treatment and mediator. section 11.3.3
• Require second-stage unconfoundedness: no hidden confounder jointly affects mediator and outcome after accounting for treatment. section 11.3.3
• Use the mediator to avoid backdoor adjustment when many confounders exist, latent confounders prevent adjustment, or unknown confounders remain plausible. section 10.5.2

Figure 10.11 Side-quest engagement leads to winning virtual items like this magic bow. Won items drive more in-game purchases, such as magic arrows for the magic bow, so we introduce a mediator “won items” on the causal path between side-quest engagement and in-game purchases.

The causal diagram places won items between side-quest engagement and in-game purchases, making the mediation structure explicit. section 10.5.1

Identification in code

Represent the graph with directed edges G→E, G→I, E→W, and W→I; query P(I do(E=e)); and provide P(I,E,W) as the observational distribution. y0 can then derive the front-door estimand. DoWhy can identify it from a pandas DataFrame, a DAG, and the treatment and outcome names. section 11.3

Identification is distinct from estimation: once the front-door estimand exists, observed variables can support inference even when some model parameters are not identifiable. A fitted causal generative model estimates this identified quantity rather than removing bias by modeling alone. section 11.6.3 section 11.6.7

How Causal AI covers it

How deep it goes

Derives the estimand with graphical algorithms and explains three assumptions, but leaves numerical estimation to later workflows.

Examples

Uses side-quest engagement, won items, guild membership, in-game purchases, y0, and DoWhy.

What it assumes

Assumes familiarity with DAGs, causal effects, mediators, confounding, d-separation, and observational distributions.

What you can do after

Apply it via y0 or DoWhy to identify a causal effect from a DAG and observational distribution.

How it connects

Connects causal effects, mediators, backdoor adjustment, do-calculus, graphical identification, latent confounding, and instrumental variables.

liveBookmanning.com

Read Causal AI on liveBook Get the book at manning.com

Related concepts

• mediatoruses · A mediator is an intermediate variable through which part or all of an exposure’s causal effect on an outcome operates.
• estimatormeasured by · An estimator is a rule, procedure, or computational model that uses data or measurements to produce an estimate of an unknown quantity, parameter, prediction, or state.
• causal identificationproduced by · Causal identification is the process of determining whether a causal quantity can be uniquely computed from available observational or interventional information and model assumptions.

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.