I am going to state the obvious: Rasmussen has had more favorable horse race numbers for Republicans this cycle than other pollsters. Why? It could have to do with its likely voter screen, interactive voice response technology, or perhaps, as some have suggested, a more sinister motive to "shape the debate". But something funny has happened in the past four days, Rasmussen's numbers have come back in line in three states.
Indeed, Rasmussen has released polling data that seems to show a Democratic uprising in Rasmussen's polling. In the state of Kentucky, Jack Conway has, in less than two weeks, cut Rand Paul's post-primary lead from 25 (on the prior Rasmussen poll) to 9. In Connecticut, Dick Blumenthal has seen his lead over Linda McMahon grow from 3 to 23 on Rasmussen polling in less than two weeks. Finally, in Missouri, Roy Blunt's 5+ point lead seen in every Rasmussen poll since February has dropped to a statistically insignificant point.
In the case of the first two contests, the shifts are exaggerated because the previous poll done by Rasmussen was fielded in one night less than 24 hours after two very important events: the first New York Times story on Blumenthal's war record on May 17 and Rand Paul winning the Kentucky primary on May 18. However, a Greenberg Quinlan Rosner poll showed a 15% Blumenthal lead over almost the identical field period as Rasmussen's 3 point Blumenthal advantage poll. While Rasmussen polls in Connecticut and Kentucky taken before these outliers were not as "out there", there was still a Republican house effect in these states and Missouri .
Interestingly, these more friendly Democratic numbers remind me of something that happened during the 2008 election. My friend David Shor, who I am currently working on projects with, documented that Rasmussen had a "summer" reversal of its Republican house effect in 2008. Looking at all polling from presidential, senatorial, gubernatorial, and house races, David found "Rasmussen polls have a statistically significant Pro-Republican house-effect that appears during primary season in the beginning of the year, disappears during the summer, and then very rapidly appears right before the Republican National Convention". His chart catalogs this well, as seen by the higher p-value in the summer months (150 to 70 days). For the less mathematically inclined, the p-value indicates the probability that a difference occurred by chance alone. So p values closer to 0 indicate a very high likelihood that the difference between Rasmussen and other pollsters was real, while a p-value closer to 1 means that any differences are most likely to to random chance.
It is too early to know whether we will see a similar shift in other Rasmussen polling. These are only three polls in three contests and for all I know the next round of polls will show other pollsters moving towards Rasmussen.
What I will say is that it does look familiar.