Definition:Controlled direct effect (CDE)
📋 Controlled direct effect (CDE) is a causal quantity that measures the impact of a treatment or exposure on an outcome when a mediating variable is held fixed at a specified level, rather than allowed to vary naturally. In the insurance context, CDE is particularly relevant when analysts want to understand how a pricing decision, underwriting rule, or policyholder characteristic affects claims outcomes through a specific pathway — net of any indirect channels that operate through an intermediate factor. For example, an insurer might ask how a change in deductible level directly affects claims frequency when the policyholder's risk-taking behavior (a mediator) is held constant at a particular value.
⚙️ Estimating the CDE requires an analyst to specify a directed acyclic graph or similar structural model that maps out how variables relate to one another, then intervene — conceptually or statistically — to fix the mediator at a chosen level. In practice, this might involve stratifying the analysis by the mediator's value or using regression adjustment under carefully stated assumptions about confounding. Consider an insurtech firm investigating whether a telematics-based premium discount causally reduces accident severity (the outcome). Driving behavior improvement is a natural mediator: discounts might encourage safer driving, which in turn reduces severity. The CDE isolates the portion of the discount's effect that operates through channels other than driving behavior — perhaps through the type of vehicle a policyholder chooses to insure, or through changes in mileage — by fixing driving behavior at a particular level.
💡 Understanding direct versus indirect causal pathways is far from academic for insurers navigating an increasingly regulated environment around algorithmic fairness and discrimination. Regulators in the European Union, the United Kingdom, and several U.S. states are scrutinizing whether rating factors such as credit scores or geographic location affect premiums through legitimate risk pathways or through channels that proxy for protected characteristics like race or ethnicity. The CDE gives actuarial teams a formal tool to decompose a variable's total effect into the portion that flows directly to the outcome versus the portion mediated by potentially problematic intermediaries. This decomposition can strengthen rate filing justifications, support compliance with anti-discrimination mandates, and inform product design decisions that are both profitable and equitable.
Related concepts: