Note this is a fancy name for "different denominators lead to different averages". If you weight classes as a first-class object, you get the small classes with equal weighting as the large classes. If you weight them by student experience, the small classes count less because there are fewer student-hours in them.
It's a name for an important special case of "different denominators lead to different averages", where the cause of the perhaps-unexpected denominator is that some quantities you're interested in estimating correlate with how likely you are to observe them.
That correlation is a key point here, and any description of the effect that doesn't include it is describing at most part of it.
This post describes situations where the average of an attribute observed by a participant is larger than the average over all elements.