In research, we have what are called "outliers". These are persons who give answers in any kind of project that are outside even the fringes of normal.
In presenting music tests and perceptual research, I used the example of a shipping box. We want to put a bunch of stuff in a box. There is one item that protrudes and keeps the box from closing. We have a decision to make: use a much bigger box, with lots of empty space and extra packing material, or ship the oversize item alone in its own box of a different shape.
The radio parallel is expanding the playlist to be much broader, taking the focus away from the core and moving into areas that are isolated. The alternative is to give the odd-man-out songs to some other station but not ours to retain our focus and specialization.
When a station does a music test, recruiting has to both cover all the core types of partisans, but exclude the very deviant outliers who will send out outside our... and our listener's... comfort zone. But occasionally we get an outlier anyway. When we look at the test results we can see, particularly if we do cluster/factor analysis graphically, if we end up with a "cluster of one" who is a person totally unlike the rest of the sample. We then remove that respondent from the results and reprocess.
There is very definitely a mathematical way of determining this beyond "I don't wanna' play this song on my station".