But this new one does.
The poll, from the Los Angeles Times and USC Dornsife, is not a normal poll. It's a tracking poll, which tallies a week's average of results for its daily figures. Since early July, it has consistently shown Trump doing at least 2 points better versus Hillary Clinton than he is in the average of polls from RealClearPolitics. At the moment, Trump is leading by a point according to the Times/USC -- but trails by more than 5 in the polling average. The RCP average also includes the Times/USC poll, which means that its Trump-friendly numbers help pull the average in a Trump-friendly direction.
So why does the Times/USC poll consistently show Trump doing better than other surveys? Earlier this month, the New York Times' Nate Cohn pointed out a unique part of the poll's methodology. The poll is a panel of 3,200 people, who the Times and USC continually poll over the course of the election. Those 3,000 people (only about a seventh of whom are actually polled each day) were asked who they voted for in 2012, and the results are weighted to reflect that.
The possible problem here is that people often misremember how they voted. This was a fairly big issue after the 2008 election when people were more likely to say they voted for Obama than polling suggested. There has been research conducted into why it happens, noting that it's usually the winner who more people remember having backed. Call it Woodstock syndrome: People like to say that they were on board with the popular thing even if they weren't.
In the Times/USC poll, though, that's a potential problem. Say that people who preferred Mitt Romney in 2012 remember having backed Barack Obama. Let's say, too, that those voters are also more likely to back Trump than are people who correctly remember voting for Obama; after all, they voted for the Republican in 2012, so it seems more likely that they would do so again. That means that a percentage of the Obama-portion of the weighted Times/USC poll is made up of people who are more inclined to back Trump than the rest of that population. Because of that weighting, in other words, a group that one would think should be a Clinton base of support may include some inadvertent Romney sympathizers.
The unorthodox methodology of the Times/USC poll extends beyond that, as Cohn notes, so there are other ways in which the survey might capture a different state of the race.
But there's one other big number that is worth noting. Asked who they think will win the race, 54 percent of respondents say that they think Clinton is the most likely winner. That question -- who people think will win -- has been shown to have advantages in predicting the outcome of the election. The idea is that asking who someone will vote for includes only that person's opinion. Asking who they think will win can capture a broader sentiment. People hear the opinions of friends and family every day, and if you yourself back Clinton but everyone around you backs Trump (which, we will note, is unlikely), you'd probably be more likely to think that Trump will win. In a sense, it's surveying more than just the one person included in the poll.
The top-line, though -- the number that Trump loves -- shows him with a lead. It's the only poll that shows that result, and the poll average continues to show Clinton with a big lead. We likely won't know for sure if the Times/USC numbers are the better ones until Nov. 8.