 Solution

One fossil-record study is Selectivity of end-Cretaceous marine bivalve extinctions by David Jablonski and others, which compared survival of marine bivalve genera across the end-Cretaceous mass extinction against their pre-event attributes. Geographically widespread genera had lower extinction intensity than narrowly distributed ones. In the same analysis, body size, burrowing versus exposed life position, and shelf bathymetric position were not significant selectors; the apparent feeding-mode contrast was concentrated in particular groups rather than establishing a universal feeding rule.^[ref]

The logic of the test is to define the pre-event genera at risk, classify their geographic ranges using pre-event occurrences, then compare the fractions becoming extinct. Using pre-event rather than surviving ranges avoids measuring the predictor after the outcome. A two-by-two contingency table gives an elementary broad/narrow-range comparison; a logistic regression can instead retain a continuous range measure and include correlated covariates. These are explanations of the inferential design, not a claim that this paper used that exact regression specification.

A persuasive interpretation must address unequal rock exposure, collecting intensity, taxonomic revision, incomplete occurrence records and the difference between local loss and global extinction. The important result is selective survival in a real fossil dataset and the failure of several plausible alternative predictors in that setting. It is observational evidence: geographic range can correlate with unmeasured ecological properties, so the association alone does not prove a unique kill mechanism.