Total effect

As explained in Causal Inference for Data Science

A total effect is the overall causal effect of a treatment or variable on an outcome, including all direct and mediated pathways.

A total effect answers a change-oriented question: if X changes, how does Y change through every causal route from X to Y? It includes a direct route and any routes carried by mediators. This differs from a direct effect, which holds the rest of the system fixed, and from an association created by a confounder, which is not a causal path. section 6.1.2

Start by stating which effect serves the decision. A topic can affect pageviews directly through reader interest and indirectly through the advertising platform selected. The total effect combines both routes. If the question is about the topic apart from platform influence, the target is the direct effect instead; choosing the wrong target can produce a correct answer to the wrong question. section 3.2.4

Graph illustrating the influence of Topic on Pageviews. This graph can help us understand how the topic affects the number of pageviews a page receives. Additionally, we may also explore how the platform affects pageviews.

Use the diagram to trace the directed routes from Topic to Pageviews and distinguish the mediator role of Platform from the treatment role it can take in another question. section 6.1.2

For linear models, total effect = direct effect + indirect effect. Causal influence travels only along directed paths from X to Y; a path such as X ← Z → Y signals confounding rather than a causal route from X to Y. The additive decomposition does not generally carry over to nonlinear models, whose fuller treatment requires counterfactuals. section 6.1.2

Hotel distance under an existing pricing policy

Suppose distance affects customer count directly and also changes price, which then affects customers. With the historical pricing policy, the total distance effect is the sum of the direct effect (−10) and indirect effect: (−2) × (−3) = 6, giving −4. Because no confounder lies between distance and customers, a simple regression of customers on distance estimates this total effect; the reported coefficient is about −3.898. A new pricing policy changes the indirect route, so it changes the relevant total effect. section 6.1.1

Figure 6.1 Graph illustrating the data-generation process for creating synthetic data, as described in the example. The arrows in the graph are labeled with their corresponding linear coefficients, all of which are negative, as this fits the context of the data modeled.

Inspect the synthetic hotel graph to see how distance reaches customers both directly and through price, with coefficients attached to each arrow. section 6.1.1

For a total-effect estimate with linear regression:

• Use the causal graph to identify back-door paths from X to Y. section 7.4.5
• Adjust for a set that closes every back-door path and contains no descendant of X. section 7.4.5
• Do not control for mediators when the goal is the total effect; controlling for them is required for the direct effect instead. section 7.4.5
• Fit Y ~ X + Z, where Z is the valid adjustment set; the coefficient of X is an unbiased total-effect estimator under the stated linear-model conditions. section 7.4.5

The same variable can play different roles as the causal question changes. Platform is a mediator when estimating Topic → Pageviews, but becomes the treatment when estimating Platform → Pageviews. Total effects therefore belong to a specified intervention, outcome, graph, and adjustment strategy—not to a variable in isolation. section 3.2.4 section 6.1.2

How Causal Inference for Data Science covers it

How deep it goes

Defines total, direct, and indirect effects; derives linear decomposition and regression estimation, while deferring nonlinear definitions involving counterfactuals.

Examples

Uses topic, platform, and pageviews, plus synthetic hotel distance, price, and customer data.

What it assumes

Assumes familiarity with causal graphs, confounders, mediators, adjustment, and linear regression while introducing effect definitions explicitly.

What you can do after

Apply it via linear regression by selecting covariates that close back-door paths to estimate a total effect.

How it connects

Connects total effects to direct effects, indirect effects, mediators, confounders, DAGs, back-door paths, and linear regression.

liveBookmanning.com

Read Causal Inference for Data Science on liveBook Get the book at manning.com

Related concepts

• causal pathuses · A directed sequence of factors or variables through which changes can be transmitted from a cause to an outcome.
• causal adjustmentproduced by · Causal adjustment estimates a causal effect by conditioning on variables that block noncausal pathways and averaging over their target-population distribution.
• indirect effecthas part · An indirect effect is the causal influence of one variable on another that operates through one or more intervening variables.
• treatment effectkind of · A treatment effect is the difference in an outcome attributable to receiving an intervention or exposure rather than a comparison condition.
• backdoor criterionconfigured by · The backdoor criterion is a graphical condition for identifying a causal effect by requiring a variable set to block every backdoor path while excluding treatment descendants.
• correlationequivalent to · Correlation is a statistical relationship describing how two or more variables or quantities vary together, including the strength and direction of their association, without by itself implying causation.

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