Warning: This is an extremely wonky article about polling

An update on our methodology

What do you do when something about your polls seems wrong?

I’m not talking about a surprising result. If your poll comes back at D+5 and the polling average has it at D+1, you shouldn’t panic. That’s in line with a normal sampling error and certainly no reason for alarm. What I mean is this: What should you think when, for several surveys in a row, your sample¹ of Hispanic voters consistently leans well to the left of the actual Hispanic electorate in America?

That’s the question we’ve been thinking through at The Argument. Over the last five surveys, our weighted Hispanic sample backed Kamala Harris by 20 percentage points in 2024 — much higher than what credible postelection studies find.

Many public pollsters consider it best practice to just accept this sort of thing as part of the normal variance and error that comes with polling. After all, one of the top-rated pollsters in American politics has exactly this attitude (and it’s one that many pollsters share): Don’t worry too much about it, because the errors cancel out.