PlumBot: Bot: Creating new article from JSON

2026-03-27T06:00:55Z

Bot: Creating new article from JSON

New page

📊 '''Common support''' — also known as the overlap condition — refers to the requirement that, for every combination of observed characteristics in a dataset, there must be a positive probability of receiving both the treatment and the control condition. In insurance analytics, this concept arises whenever analysts attempt to compare two groups — such as policyholders who adopted a [[Definition:Telematics | telematics]] device versus those who did not, or commercial risks that were placed through a [[Definition:Managing general agent (MGA) | managing general agent]] versus those written directly — and need to ensure that the comparison is made between genuinely comparable observations rather than between populations so different that any estimated effect is pure extrapolation.

⚙️ Assessing common support is a practical prerequisite for matching and weighting methods such as [[Definition:Propensity score matching (PSM) | propensity score matching]] and [[Definition:Coarsened exact matching (CEM) | coarsened exact matching]]. Analysts typically examine the distribution of [[Definition:Propensity score | propensity scores]] or key covariates across treated and untreated groups and discard observations that fall outside the region of overlap. In an insurance context, imagine a [[Definition:Workers' compensation insurance | workers' compensation]] insurer evaluating whether a return-to-work program reduces [[Definition:Claims | claims]] duration. If the program was offered only to claimants in low-[[Definition:Severity | severity]] injury categories, there may be no comparable untreated claimants among high-severity cases, meaning the common support condition fails for that segment. Proceeding with the analysis as though these groups are comparable would produce unreliable estimates. The trimming of non-overlapping observations reduces the effective sample size but dramatically improves the credibility of the resulting [[Definition:Causal inference | causal]] estimates.

💡 Violations of common support are particularly treacherous in insurance because the populations being compared often differ systematically by design. [[Definition:Underwriting | Underwriters]] select risks, [[Definition:Policyholder | policyholders]] self-select into optional coverages, and [[Definition:Reinsurance | reinsurers]] choose which portfolios to participate in — all of which create structural gaps in overlap. An [[Definition:Insurtech | insurtech]] developing a usage-based [[Definition:Motor insurance | motor insurance]] product, for example, may find that early adopters skew heavily toward younger urban drivers with newer vehicles, leaving virtually no overlap with the traditional book's rural or older-vehicle segments. Recognizing and transparently reporting the boundaries of common support prevents overgeneralization of findings and keeps decision-makers from applying insights beyond the population for which evidence actually exists. In regulatory contexts, demonstrating that a study respects the overlap condition strengthens the defensibility of any resulting [[Definition:Rating factor | rating factor]] or pricing adjustment.

'''Related concepts:'''
{{Div col|colwidth=20em}}
* [[Definition:Propensity score matching (PSM)]]
* [[Definition:Coarsened exact matching (CEM)]]
* [[Definition:Selection bias]]
* [[Definition:Causal inference]]
* [[Definition:Confounding variable]]
* [[Definition:Conditional average treatment effect (CATE)]]
{{Div col end}}

Definition:Common support - Revision history

PlumBot: Bot: Creating new article from JSON