Even though the data here is objective and indisputable, there are all kinds of different ways that you can read it if you want to measure the performance of millennials. The takes that I've seen, for the most part, largely differ based on two questions:
1. Do we care about candidate support or candidate preference? This is the difference between allowing a "neither one" vote or forcing a choice between two options. A "support" measurement cares about only about whether one actively and affirmatively backs a particular candidate; a "preference" measurement also cares about whether one in some sense opposes a particular candidate, even if one does not affirmatively back the other. There are all kinds of analytical and philosophical reasons why one might use one measure rather than the other. If for instance you're interested in latent dispositions that people don't express through their actual voting behavior, you might care more about preference; but if you want to measure things like apathy, you might care more about support.
2. Do we want to make a static analysis or a comparative historical analysis? This is the difference between just looking at this year's numbers or placing them in the context of previous elections. A static analysis, for example, would conclude that Clinton's numbers among black voters were extraordinarily high - but a historical analysis would conclude that they were actually low compared to the past few elections. That sort of approach might be useful if, say, you suspect that there are structural / systematic issues that guarantee a baseline of demographic support for a party regardless of candidate, and are more interested in how candidates vary from that baseline; a static analysis is more useful if you reject that sort of assumption.Crucially, millennial performance in 2016 will look quite different depending on what we decide to measure:
Reading the numbers in a way that flatters or blames millennials is really just a matter of picking the right analytical lens. If you want to attack millennials, do a 2016 support analysis, or a 2016 vs 2012 preference analysis; if you want to praise them, do a 2016 preference analysis, or a 2016 vs 2012 support analysis. These approaches are all perfectly rigorous and legitimate, as far as they go, even though they can give you dramatically different outcomes.
Obviously, coming up with some kind of objective and dispositive conclusion about millennial performance in 2016 is less a matter of (fairly straightforward) number-crunching, and more a matter of defending analytical methodology. This means grappling with all kinds of extremely thorny philosophical and poli-sci controversies about agency, culpability, structural determination, and so on - the kinds of controversies in play when we asked the two basic questions above.
That the overwhelming majority of election analysis doesn't even pretend to care about such issues says everything you need to know about how serious one should take them. In reality, most of our election punditry is just a matter of deciding on a conclusion and then back-filling the corresponding analytical approach, with zero attention to the decisive methodological questions at hand.