Conventional wisdom is that Iowa is a two-way race between Trump and Cruz, with Trump probably holding a small lead. Conventional wisdom is that Rubio and Carson will duke it out for third. Conventional wisdom is that the Democrat race is too close to call between Bernie and Hillary.

The source of all this conventional wisdom, of course, is polling. Polling is fun to discuss, and we do a lot of it here. At the end of the day, however, it’s a lot like the statistics you hear in pre-game shows for NFL football games: it provides data for the talking heads to discuss so that they can provide an interesting show, but at the end of the day, no one wants to admit that they have no idea who is going to win the game.

The problem with the polling data this time around is that it’s flat out telling us that it’s unreliable. The way the polling “science” is supposed to work is that each candidate’s level of support is measured within a “margin of error.” What that margin of error actually says is that, with 95% probability, the given candidate’s support is within that number of points.

You can imagine the clusterfudge this creates with a poll that features 10 candidates. Suppose that you have a poll with a MoE of 4, and it reads Trump 30, Cruz 25, Rubio 15, Carson 10, Christie 8, Bush 6, Paul 5, Fiorina 5, etc. What this means is that Trump’s support has (theoretically) a 95% chance of being between 26 and 34. Likewise, Cruz’s support has a 95% chance of being between 21 and 29, Rubio between 19 and 11, and so on and so forth. The reality is that polling in such a crowded race is a mostly useless exercise in guesswork, especially given the volatility of the polling in Iowa this year (and the historically bad results of Iowa polling).

Probably, in a state like New Hampshire, we can have some additional level of confidence that Trump truly does have a fairly substantial lead.  But Iowa right now is an absolute crapshoot.

The best and most dramatic proof of that, actually, occurs on the Democrat side. Loras College and CNN conducted polls of likely Iowa Democratic caucus goers over a completely overlapping set of days. One of those polls found Clinton leading the race by an overwhelming margin, 59%-30%. The other found Bernie leading by a fairly comfortable margin of %51-43. In fact, the last four polls taken in Iowa – all taken last week – show, in order, Sanders +1, Sanders +8, Clinton +9, Clinton +29. And polling a two person race is infinitely simpler and more reliable than polling a 10 person race.

Polling on the Republican side has been no less volatile, with polling conducted in the last week showing Trump’s support ranging anywhere from 25 to 39%, and showing Cruz’s support ranging from anywhere between 23 and 34%.

Recent poll results in Iowa, on the Republican side in particular, have been notoriously awful. At this exactly point of the race in 2012, the RCP polling average told us Ron Paul was going to win Iowa with 22% of the vote, and Santorum was supposed to finish sixth with around 7%. When the real ballots were cast, Santorum stunned the field by garnering 24% of the vote and Ron Paul finished a disappointing third. Likewise polling at this exact point in the 2008 cycle completely missed the boat for the Democrats as well, predicting a narrow Clinton victory when in fact Obama won relatively comfortably.

The bottom line here is that anyone who tells you with any level of confidence who is going to win Iowa and by how much is not to be trusted, any more than anyone should trust Howie Long to know who is going to win any given football game before it is actually played.

It’s going to be a hell of a week.