| Its not stereotyping, but coming up with Bayesian estimates of conditional probabilities. According to the data they collected, P(person is female | person is a shooter) is just very low. Similarly, P( person is a college student AND person is of color | person is a shooter) > P(person is a college student AND person is not of color | person is a shooter). Note that "white males" did not always appear. In one case race wasn't a distinguishing factor (workplace violence), in another color was a distinguishing factor (college violence), and in the rest, being white was a distinguishing factor. This is pretty balanced overall. The data needs to speak for itself. It's okay to notice that in some cases, white males are more likely to commit mass shootings. As okay as noticing that in other cases, non-white males are more likely to commit mass shootings. This means that we can be zoom in better on the issues that lead to such crimes being committed, and get people the help they need, before they irreversibly hurt themselves and others. They aren't talking about P(person is a shooter | person is a white male). That would be stereotyping, but also easily dismissed, because as you noted, P(person is a shooter | person is a white male) is also very low. |
The problem of stereotyping is that people take a complex problem and attempts to reduce it by measuring people based on one or a few bits of information.
Crime is a bit like rain. The more one attempt to zoom closer to the atom the less we understand it and the poorer the prediction becomes. It get even worse when attempting to zoom in order to understand rare events.
If we look at the specific crime of rape in Sweden we see that P(person is Muslim | person is a rapist) > P(Person is a Christian | person is a rapist), by around 300%. The political reaction to people noticing the data is a bit volatile to say the least.
It is possible to still use the data, but its best to zoom out rather than in. A common one for crime is the acknowledgement that high risk groups tend share a trait of low social economic status. Thus a popular general prediction method is to measure social economic status when determining risk. What we then get is a more general P(person is socially isolated AND low income AND low education | person commit a crime) that we can compare to other prediction models. People then take each of those classifications and zooms out even further by addressing them independently and outside of crime prevention as improving them has value in itself.