As I mentioned below, the NYT has a very sophisticated realtime prediction widget that shifts its prediction as results come in. It had been bouncing around dead even with a mild advantage either way. Then it swung hard in Jones’ favor. I saw a number of people speculate that there had been some sort of reporting error. Here’s a tweet just now from a member of the team that runs it.

I would treat this as a tentative read, subject to change. What it points to is dramatic enough that it could be a reporting error or just a quirk in the prediction model. I am, however, seeing some indications of something we saw in Virginia: both candidates getting solid percentages from their strong regions but with Moore getting slightly weaker turnout. As always we need more data. But for the moment things look better for Jones than for Moore.

9:29 PM: This is starting to come into focus: differential turnout. Moore’s winning the red areas, and often with the percentages you’d expect. And vice versa. But turnout is a bit weaker in the red areas and a bit stronger in the blue areas. Frankly, you can’t say that’s a totally unpredictable outcome with motivated Democratic base and a Republican candidate who has been credibly accused of preying on teenager girls.