A manipulative statistical practice where a researcher splits a dataset into very
small subgroups (or isolates a
single case) until no comparison achieves statistical significance, then reports that “no effect was found.” The fragmentation acts as a
kill‑switch: by reducing sample size to near zero, any real effect becomes undetectable. This tactic is often hidden behind legitimate concerns about confounding, but when applied selectively, it ensures that any inconvenient finding can be made to disappear. It’s a favorite of litigation consultants and industry apologists.
Sample Fragmentation (
Kill-Switch) Example: “He divided the adverse event
data by hospital, shift, and
day of week – sample fragmentation, ensuring no group was large enough to show a problem.”