How can programmers often find unit tests for features that are deleted?

I am quite puzzled by the following survey:

The survey shows that 26% of programmers find test for features that does not exist anymore regularly or frequently, 2% find them all the time, and only 18% never find them.

If the test is testing a particular behavior of a feature and that feature is deleted, should the test not fail? So how come people can find them later on?