Causal assumption

As explained in Causal AI

A causal assumption is a constraint about how variables or events are causally related that guides causal inference.

Suppose you want to know whether increasing side-quest engagement would change in-game purchases. A correlation between engagement and purchases does not answer that intervention question. A causal assumption constrains how the variables are related so that data can support a causal inference, not just a prediction. section 1.1 section 10.2.2

The first practical step is to state what you believe causes what. A causal directed acyclic graph (DAG) makes those beliefs visible as edges between variables. It is an account of the data-generating process, not a diagram verified simply by fitting it to data. The graph says *what* causes an effect, but not *how* the effect is produced: different logic gates, for example, can share the same causal DAG. section 3.1.2 section 3.1.3

Choose assumptions at the level the question requires:

• Associational models make statistical assumptions about observed relationships. Those assumptions alone do not establish an intervention’s consequences. section 10.1.2
• A causal DAG makes interventional assumptions about what causes what. Its structure can supply the level of assumptions needed for an intervention query. section 10.1.2 section 10.2.2
• A structural causal model (SCM) can add assumptions about *how* causes act—for instance, a proportional increase in height for each unit increase in femur length. These mechanism assumptions can support further inferences, including counterfactual queries, but create more opportunities for modeling error. section 6.1 section 6.1.5

Which uncertain edge matters?

For an average treatment effect of side-quest engagement on in-game purchases, uncertainty about whether time spent playing causes skill level or the reverse need not matter to the chosen backdoor estimand. An omitted common cause of engagement and purchases does matter: it would violate the assumption that adjustment handles the relevant confounding. Judge an uncertain graph against the assumptions needed for the particular estimand, rather than requiring every edge to be settled. section 11.3.4

Figure 11.7 Direct influence of a latent variable on the treatment and outcome would violate the assumption underpinning the backdoor adjustment estimand. If you are not confident in an estimand’s assumptions, target another.

The diagram shows how a latent variable influencing both treatment and outcome threatens the backdoor adjustment assumption. section 11.3.4

A workable sequence is to pose the causal query, build a model, check whether the query is identifiable from that model and the available data, then estimate it. If identification fails, seek additional observed variables, reconsider the assumptions, or leave the question unanswered. After estimation, test sensitivity to assumption violations. Observational data cannot verify every causal assumption; a robustness check that adds an unobserved confounder can show how quickly the estimated effect changes. section 10.2.2 section 11.3.4 section 11.5.5

More detail is not automatically better. Two SCMs can agree on observational and interventional distributions yet produce different counterfactual inferences because they encode different mechanisms. Use additional assumptions when they help answer the target question, and account for the risk if they are wrong. section 10.6.1 section 6.1.5

How Causal AI covers it

How deep it goes

Distinguishes graphical assumptions about what causes what from structural assumptions about how, then checks their role in identification and sensitivity.

Examples

Uses logic gates, femur length and height, and game engagement and purchases to distinguish graphical, mechanism, and estimand-specific assumptions.

What it assumes

Builds from the distinction between correlation and causation; uses causal DAGs and intervention queries to develop more detailed assumptions.

What you can do after

Evaluate/understand which assumptions support a chosen estimand and probe how unobserved confounding could change its estimate.

How it connects

Links causal DAGs and structural causal models to the causal hierarchy, identification, backdoor adjustment, and counterfactual inference.

liveBookmanning.com

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

Also explained in

• 

  liveBookmanning.com
  Causal Inference for Data ScienceA causal assumption is a condition about variables, relationships, or the data-generating process that is taken to hold when drawing causal conclusions.

Compare how 2 books explain causal assumption →

Related concepts

• structural causal modelused by · A structural causal model represents causal relationships among variables through equations that determine each variable from other variables and exogenous factors.
• graphconfigures · A graph is a mathematical structure consisting of vertices connected by edges that represent relationships or connections.
• causal graphical modelused by · A causal graphical model is a probabilistic or structural model whose graph represents causal relationships and whose mechanisms support reasoning about interventions.
• linear regressionincludes · Linear regression is a statistical modeling method that estimates a continuous response as a linear combination of predictor variables and fitted coefficients.
• causal inferenceproduces · Causal inference is the process of determining or estimating how causes, exposures, or interventions affect outcomes, using data and assumptions about causal relationships.

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.