Tomorrow’s Republican primary in South Carolina could provide some big surprises, if you believe some of the latest polls. The most dramatic is a new poll from Opinion Savvy showing that Marco Rubio might be in striking distance of longtime frontrunner Donald Trump.

Of course, you should take any poll with a grain of salt, and there’s a flurry of them telling us different things. As Dan Spencer noted this morning, the latest Wall Street Journal/NBC News/Marist poll had Trump 28, Cruz 23, Rubio 15, Jeb 13, Kasich and Carson at 9 each – a much tighter race than the double-digit leads and polling in the mid/high 30s than we had seen from Trump previously. The OpinionSavvy poll, taken the last two days, has Trump 27, Rubio 24, Cruz 19, Jeb 11, Carson 8, Kasich 7, and 4% undecided, but warns that “Trump has managed to improve his ranking within the youngest demographic, and he maintains the highest commitment from voters: 97% of Trump supporters indicated that they do not foresee changing their vote.” The two polls thus paint a similar picture of softening Trump support, but very different pictures of the race between Rubio and Cruz for second and/or an outside shot at beating Trump.

The other two most recent polls, by ARG and Emerson, both have Trump still riding high at 34 and 36, respectively. ARG has Rubio charging, up over the last four two-day periods from 14% to 16% to 20% to 22%. Consistent with ARG’s tendency to sample Kasich-friendly voters, it also has Kasich ahead of Cruz 14-13, holding mostly steady the last four days, and Jeb in single digits. Emerson has Rubio and Cruz all but tied (Rubio 19, Cruz 18) with Jeb and Kasich at 10% each and Carson at 6. There’s also a Clemson poll with Trump at 28%, followed by Cruz 19, Rubio 15, Jeb 10, Kasich 9, Carson 6.

Who to believe? Henry Olsen argues that the polls that show Trump in the 20s are projecting an electorate with too many conservatives and too few new voters. The poll averages provide some efforts at balancing them out – RCP at this writing has Trump 31.8, Cruz 18.3, Rubio 17.6, Jeb 10.5, Kasich 9.6, Carson 6.9; Huffpost Pollster, which averages a bit differently, has Trump 34.1, Cruz 19.3, Rubio 15.7, Jeb 9.8, Kasich 9.0, and Carson 6.0.

How reliable is any of this? Compare the results to the final RCP average in the last two South Carolina primaries (poll averages in parentheses):

2012: Gingrich 40.4 (33.5), Romney 27.9 (28.5), Santorum 17.0 (11.8), Ron Paul 13.0 (13.8), Perry 0.4 (2.5). Newt led Romney 37-21 less than three weeks out and collapsed to third place after Santorum won Iowa, but recovered his lead in the last two days and overperformed his polls by 7 points, in part because Perry dropped out and threw him his support. Santorum also beat his polls by 6 points.

2008: McCain 33.2 (26.9), Huckabee 29.9 (25.9), Romney 15.1 (14.7), Fred Thompson 15.7 (14.6), Ron Paul 3.7 (4.4), Giuliani 2.1 (3.4). Huck led 32.3-19.7 ten days out and only lost his poll lead four days out, so there was a lot of late movement to McCain, plus below the top a lot of late movement from Romney to Fred. McCain, incidentally, is the only winner of the South Carolina Republican primary since 1980 who did not win by a double-digit margin, which explains why the state has been so decisive so often:

SC Margins

It’s hard to say how much of that is bad polling and how much is just the wild nature of these races right on the heels of Iowa and New Hampshire and with the field narrowing rapidly and endorsments coming late in the game. Closely related is the fact that turnout can be really volatile in South Carolina – look at the vote totals for the last four contested Republican primaries:

1996: 276,819
2000: 573,101
2008: 445,499
2012: 605,623

In other words, turnout in the GOP primary since 2000 has ranged from 13.45% of all eligible voters in 2008 to 20.07% in 2000. That’s a lot higher than the Iowa caucus, where turnout this year surged to 8.17% of eligible voters, but lower than New Hampshire, where it has ranged the last three cycles from 23.67% to 27.43%. And because South Carolina doesn’t have the steady high turnout of New Hampshire, it’s harder to project who will show up.

South Carolina has an open primary, and turnout was massively up on the Democratic side in 2008, which may partly explain the drain out of the GOP race, as well as a general lack of enthusiasm for the options (McCain got 147,686 votes compared to 305,998 for George W. Bush in 2000 and 244,065 for Newt in 2012 and more in line with the 124,801 votes for Bob Dole in 1996 – both McCain and Dole appealed to South Carolina’s veterans but not more broadly across the state). If Iowa and New Hampshire are any indication, expect Saturday’s turnout to blow past that 2012 figure, especially with the Democrats not voting for another week.

The Rubio camp in particular has to be feeling pretty good about some of the signs of momentum here, which look a lot like his late charge in Iowa. But there are also plenty of tea leaves here for the Cruz campaign to be optimistic about, and at the same time still polls showing Jeb and/or Kasich right on the heels of Rubio and/or Cruz. So hold on to your hats: South Carolina could surprise us all.