Almost exactly four years ago--not coincidentally, during the last Presidential campaign--Kevin Drum wrote this piece about margins of error in polling and how to read them. It's well worth revisiting now and often.
This is a case where the obvious, common-sense reading of the data (a 2% lead in a poll must be a little better than a 1% lead, right?) is more accurate than the common journalistic presentation that purports to correct the obvious reading (a lead only becomes meaningful when it's bigger than the margin of error, and then it is magically superdupermeaningful).
Looking back at Drum's post, I am reminded again how Nate Silver at fivethirtyeight.com has created a new level of statistical political analysis. Whenever I find myself at a public computer with a few minutes to kill, I go to Nate first.