Definition:Regression discontinuity design (RDD)
📐 Regression discontinuity design (RDD) is a quasi-experimental method that exploits a known cutoff or threshold in an assignment variable to estimate causal effects — a technique increasingly applied in insurance research where policy rules, underwriting guidelines, or regulatory boundaries create sharp eligibility divisions. When an insurer prices motor insurance differently for drivers above versus below age 25, or when a health plan applies a different cost-sharing structure once a deductible is met, the discontinuity at the threshold creates a natural experiment: individuals just on either side of the cutoff are nearly identical in all respects except their treatment status, allowing analysts to attribute outcome differences to the rule itself rather than to underlying risk characteristics.
⚙️ Implementation involves collecting data on the assignment variable (age, credit score, claim amount, policy duration, or any continuous measure used to determine eligibility) and comparing outcomes for observations in a narrow band around the cutoff. A life insurer studying whether a simplified-issue threshold — say, coverage amounts below a certain limit that bypass full medical underwriting — affects mortality experience can compare claims outcomes for applicants who requested amounts just above versus just below that limit. The "sharp" variant of RDD applies when the cutoff perfectly determines treatment; the "fuzzy" variant accommodates cases where the cutoff strongly predicts but does not guarantee treatment, requiring instrumental-variable adjustments. The method's internal validity is high near the threshold, but findings may not generalize far from it — a limitation analysts must acknowledge when extrapolating results to the broader book of business.
💡 RDD appeals to insurers precisely because the industry is built on thresholds: rating-factor breakpoints, coverage-limit tiers, age-banded premiums, regulatory eligibility rules, and surplus-lines filing triggers all create the sharp boundaries the method requires. In jurisdictions where regulators scrutinize whether certain underwriting cutoffs produce fair outcomes — such as the European Union's restrictions on gender-based pricing following the 2011 Test-Achats ruling, or the NAIC's attention to credit-score usage in personal lines — RDD provides a principled way to measure the real-world impact of imposing or removing a threshold. Insurtech data teams also find RDD valuable for evaluating the effectiveness of automated triage rules in claims handling, where cases above a complexity score are routed to senior adjusters. By offering credible causal estimates in settings where randomized controlled trials would be impractical, RDD fills an important gap in the insurance analyst's methodological repertoire.
Related concepts: