When something does happen to suddenly and discontinuously shift the electorate--a good performance in a debate, say--a seven-day moving-average poll dribbles the effect of that shift out into its estimates in little pieces over the next seven days.
This creates the illusion of momentum: the line goes up, and keeps going up. But it actually is not momentum: it's simply that the share of the poll results that are pre-event and thus stale drops only gradually.
There are much better ways to handle estimating the current level when a process is potentially subject to discontinuous shocks and you know when they are than a seven-day moving average. But the RAND organization--and the Gallup organization, and others--don't use them. Intellectual malpractice.
Now 3/7 of the RAND panel is post-third debate. Obama now has a lead of 5.92%. Back when 0 of the RAND panel was post-third debate, Obama had a lead of 1.90%.
The point estimate of what Obama's RAND panel lead will be when the entire RAND panel is post-third debate is now… 11.28%