For the past couple of weeks I've been following The RAND Continuous 2012 Presidential Election Poll. This poll is mentioned less in the media, and I'm not sure whether the various poll-averaging outfits include it in their forecast. However, IMHO it is the best place to gauge the effect of specific events on public opinion. And boy, what an effect the RAND poll has been showing this past 3 days!
The poll is unique in that it tracks The same ~3,000 registered voters. In fact, it seems to have tracked them for several years, but 2012 elections have been polled among the sample only since July. The downside of using the same sample (participants developing a "guinea pig mentality", etc.) are matched by some very good upsides, such as a high response rate (I believe participants log their answers online every 1-2 days). Most importantly, because RAND tracks the same people, day-to-day changes can be more meaningful than when you ask a different set of people on different days. The numbers shown in their plots are 7-day rolling averages. The poll also over-samples swing voters - and then re-weights them back to known demographic composition, as done by most other serious polling outfits.
And now the results: until this week, the RAND poll has shown similar patterns from other polls, such as an overall modest Obama lead, and a modest (but existing) GOP convention bump beaten by a larger Democratic bump.
However, since Monday the 17th when the '47% video' was released, Obama has been simply running away from Romney. The lead has grown from about 3%, to 5.4% last night. This is the largest lead recorded in this poll, it has burst out of the 95% confidence band (shaded gray area) - and it is now increasing by about 1% per day with no end in sight. Remember, these are 7-day rolling averages, and they only include 2.5 days of post-video data! This means that each new day since Monday is about 7% better for Obama than the same day last week.
The other tabs show what this lead is composed of:
the shifts between candidates tab spiked on Monday night, and continues to display a healthy pro-Obama gap. These are 7-day rolling
sums of shifts rather than averages, so the overall number of people changing their support every day is only a fraction of a percent - but still, the persistent pro-Obama gap accumulates to a lead increase over time.
Then, there is the intention-to-vote tab. Since this intention is incorporated in the poll's bottom line, the main poll is a likely voter poll and not a registered voter one. The intention-to-vote shows the familiar and sad pro-GOP intensity gap. However, since Monday the gap has been shrinking - more so due to a drop in Romney numbers, but also due to an increase in Obama numbers. The gap is now 3%.
At this point we should note another unique character of the RAND poll: participants indicate both their level of support for each candidate, and their intention to vote, as numbers between 0 and 100% rather than a 2-level or 4-level response. So the final election prediction shown in the graph above, is the average of all the personal (% support) x (% intent-to-vote) products. While this might seem strange at first, as statisticians we actually much prefer continuous numerical outcomes, over answers on a categorical scale (such as "Certain Obama", "leaning Obama", "leaning Romney", "certain Romney"). Numerical answers are much more precise and sensitive to changes - yet another reason why the RAND poll might be the most sensitive publicly-available seismograph of voter sentiment.
Another interesting tab on this site shows a different cross-tab every day (in tracking graph format). Today it is the ethnic-group voting breakdown, which suggests that the recent pro-Obama surge is mostly driven by a closing of the anti-Obama gap among white voters.
Update (7:45 AM PDT): image added. h/t Scarce. I was too rushed to post one at first.
Another Update (9:20 AM PDT): a diary posted in the wee hours by JMcDonald presented these news. The RAND poll is updated nightly at 1 AM PDT with the previous night's data. h/t Andrew F. Cockburn. Also a little added verbiage regarding the continuous nature of the polling response itself.