Even with Monmouth University polling's modifier, this is an understatement: "Turnout in the Iowa Democratic caucuses is extremely unpredictable."
It's been enjoyable watching Steve Kornacki, Chris Hayes, Chuck Todd and other analysts cast wide-ranging CYA predictions based entirely on turnout — the higher the better for the "insurgent" Sanders, of course. Sanders' problem is that his winning-turnout numbers are grounded in the youth vote, an infinitely unreliable bloc. Clinton, on the other hand, dominates in the female vote and the over-50 vote, while tying Sanders among men.
The overall predictive result, according to Monmouth (graded "A-" by FiveThirtyEight), is that Hillary is still ahead of Bernie among likely Iowa caucus-goers by at least 5 points. What's more, "Among voters who have a history of voting in state Democratic primaries, Clinton currently leads Sanders by 47% to 39%."
Like Trump's, a Sanders victory in Iowa is predicated on historically less likely voters turning out. I'd give Trump better odds, for his less likely voters are both older and angier, both of which partially offset the unlikelihood of their ever turning out.
Turning to Marist's polling (graded "B+" by 538), the story is essentially the same: "Hillary Clinton, 48%, and Vermont Senator Bernie Sanders, 45% … among likely Democratic caucus-goers in Iowa." Marist tells a vastly "different story in New Hampshire," however, "where Sanders, 57%, has opened up a 19 point lead over Clinton, 38%, among likely Democratic primary voters."
Unremarked by Marist is that a Sanders loss in Iowa could well subtract from his present 19-point lead in N.H.
Were I to predict? Pretty much what's above. I'd say: a narrow Clinton victory in Iowa, a narrowed loss in N.H., and then on to Nevada, where Clinton presently leads by about 20 points, and South Carolina, where she leads by about 30.
And there, in South Carolina, is just about where this will all end.