The most recent New York Times/CBS News poll, which showed a tied race between Mr. Trump and Mrs. Clinton heading into the Republican convention, found that 33 percent said they had voted for Mr. Obama and 25 percent for Mr. Romney. The results were similar in May: 41 percent for Obama and 32 percent for Mr. Romney (the numbers are higher because it was asked only of registered voters in May). The most recent NBC/WSJ poll, conducted after the Democratic convention, showed an even larger 46-to-31 split among registered voters.

This is not a new phenomenon. Back in 2012, Pew’s surveys showed Mr. Obama ahead by 34 to 25 among voters from 2008. If you have a really long memory, you might even remember controversy about polls that showed people recalled voting for George W. Bush over Al Gore in 2000 by a comfortable margin in 2004 polls, even though Mr. Gore won the popular vote. But it’s not perfectly consistent, either: With Mr. Bush’s popularity flagging in 2007 and 2008, more polls started showing that voters recalled voting for John Kerry in 2004.

With these figures in mind, the U.S.C./LAT poll’s decision to weight its sample to 25 percent for Mr. Obama and 23 percent for Mr. Romney is quite risky. If the panelists, like those in other surveys, are likelier to recall voting for the winner (Mr. Obama), then the poll is unintentionally giving extra weight to Republican voters. Or you can imagine a counterfactual: If the poll were weighted to 33 percent for Obama and 25 percent for Romney (per the NYT/CBS numbers), then Mrs. Clinton would hold a more comfortable lead.