Definition
Plain language
The maze of small, reasonable judgment calls in any data analysis, where each turn is defensible but the route you take decides the answer.
As stated in the literature
Andrew Gelman's term for the space of defensible analytical choices — which variables, which sample, which model, which controls — in which each fork is individually justifiable yet the chosen path determines the reported result, making bias invisible to review of the single path taken.
Why it matters: It shows how bias can creep into research invisibly, since each individual choice looks justified even though the chosen path quietly determined the outcome.
For example, one researcher drops a few outliers and picks a certain age range and finds a strong result, while another making equally reasonable choices on the same data finds nothing.
Heard on the show
“Statisticians have a name for that maze: the garden of forking paths, Andrew Gelman's phrase.”Episode 196 — AI Agents Reached Opposite Conclusions From the Same Data — and Passed Review