Introduced on day 5
Collider
A variable caused by two others on a path; conditioning on it (filtering or adjusting) creates a spurious association between its causes.
In detail
Example: suppose fares were set at random, unrelated to demand. Selling out needs strong demand or a low fare, so among sold-out flights the weak-demand ones are mostly low-fare ones - fare and demand now look related although they were not. Selecting on an outcome is collider bias.