@parados,
It all depends on what you want to know, and how your population is structured. Using averages means nothing if the population you study is pluri-modal rather than normal, for instance. The stats linked up by rev make this error: They assume a normal, gauss like distribution of income among all candidates supporters. This may not be true, and thus their averages may mean nothing whatsoever.
The right way to go if you want to prove that Trump has little support among the working class is to focus on specific income brackets characteristics of the working class, and look at the voters behavior WITHIN each income bracket. Averaging ALL voters incomes including rich, poor and in between will mix apples and oranges.
Not that I expect any of you to get a rather subtle statistical point...