The polling news for Trump is already not great. He is consistently behind in national polls by a margin of anywhere from 5-13 points. The statewide polls, especially in battleground states, are even worse. But there’s a huge difference in the national picture if you believe Trump is really only behind 2-4 points (as some pollsters like LA Times/USC and Rasmussen say) or if you believe he is behind by closer to 10-13 points.

The issue, for me, is that I still don’t think the polling is capturing the demographics of the coming electorate accurately. I’m not trying to unskew polls here, I’m trying to account for something I genuinely don’t think pollsters know how to accurately predict, and are totally guessing about: the size of the Hispanic vote in this upcoming election. And what you think about how that is going to look like has a huge impact on how you see the state of the race.

It is a well known and observed political fact that (eligible) Hispanics vote at a much, much lower rate than whites or blacks. In 2012, they voted at a rate of about 48%, as compared with 64% for whites and 66% for blacks. The end result is that Hispanics represented about 8% of the U.S. voting population in 2012, despite comprising over 17% of the total U.S. Population as a whole.

Now, Hispanic voting rates have been slowly and gradually trending upwards, and pollsters have been doing a reasonably good job of accounting for this fact over the years, in spite of the fact that the Hispanic community is notoriously difficult to poll or predict.

The polling models seem to assume a relatively modest continued growth of the Hispanic vote as an overall total of the population, and more or less a constant continuation of the voting split between Rs and Ds from 2012. YouGov’s model shows the Hispanic vote at 9% of total voting population, splitting 52-20 in favor of Clinton. Pew has the Hispanic vote at 8% of the total population, with roughly the same split. I’m not paying to get crosstabs from Rasmussen’s garbage polling, but I would bet it’s the same. The LA Times/USC poll, which has been the most consistently pro-Trump poll in the country, likewise does not release demographic crosstabs.

But look what happens if you assume (as I do and some of the other polling companies do) that Hillary Clinton’s efforts to turn out the Hispanic vote at an unprecedented level will be even marginally successful. Monmouth shows the Hispanic vote comprising 11% of the overall total and breaking much more sharply for Clinton. The end result is a poll that shows Trump facing a 13 point deficit instead of a 4-7 point deficit. McClatchy likewise shows a 12% Hispanic vote total, and a 15% overall lead for Clinton, and so on.

If you look at the way party loyalty is more or less locked in throughout the country, it’s difficult to run up a lead of more than 4-7% assuming the continuation of recent voting trends (although Trump is certainly trying). However, the one huge remaining variable out there remains the extent to which Trump’s rhetoric and Hillary’s turnout operation will successfully target and encourage Hispanics to vote.

Again, this is not an attempt to unskew any polls; I think the polls more or less have the state of the race correct, given standard assumptions. What I am saying is that there may be an unaccounted for factor that might cause the race on election day to be worse for Trump than even the polls are predicting. Then again, it might not. But it’s something for Trump and his supporters to keep in the back of their minds.