I just read a few diaries here and here about Obama's dip in Nate Silver's new forecast. The alarmism from both Nate and the Kossacks that follow him is surprising, as the dip can be easily explained, at least in part. Unfortunately, the explanation indicates an even more suspicious and alarming tactic coming from the Right. A few quick thoughts below the fold.
First, Nate's numbers. One of the things Nate's model takes into account is polls. I am not sure what Nate's methodology is for screen models, but I would be surprised if he didn't use the typical indiscriminate approach of just dumping the entire polling universe into his database. He has other data points that makes his model more accurate than simple poll aggregation, but on poll inclusion, I doubt he takes the time to verify the validity and integrity of ever poll, instead just dumping them all into the mix.
Like many of you, I read Steve Singiser's wonderful polling wrap everyday just to get monitor the race and get a sense of what numbers will be driving the media for the day. I have noticed a trend in the polling since the debate besides the usual narrative of just the numbers: the sources of the polls have taken a large Right turn.
Using Steve Singiser's polling wrap, I compiled a count of the pollsters from conservative firms pre- and post- debate, as shown in the table above. Any firms I did not know well enough to judge, I added them to the "Rest of Pollsters" column. As you can see by the counts, the overwhelming majority of pre-debate polls were conducted by either reputable firms (PPP, various universities, NBC/Marist, Selzer, etc.) or smaller firms on occasion that I was not familiar with (these were much less frequent). However, after the debate, there has been a notable and undeniable shift in the sources of the polls. Openly and undeniably conservative hack firms, such as Gravis, WAA, and ARG, have conducted the vast majority of post-debate state polling. In addition to simply providing the majority of the data post-debate, these firms have increased their output significantly even compared to their own pre-debate output.
As an example of how this can effect the data Nate (and other aggregators and forecasters) use in their models, let's look at the polling from October 11th (it was a particularly heavy polling day). In Ohio, NBC/Marist had Obama up 6 among both likely and registered voters and at or above 50% among both, numbers consistent with his pre-debate numbers. Against that one poll, conservative outfits put out three. Those three had Obama +1, Obama +1 and Romney +1. In Virginia, CBS/Quinnipiac had Obama +4 and NBC/Marist had Romney +1 LV and even among RV. Against those, conservative outfits had Romney even and +7. Pennsylvania was even more egregious. Philadelphia Inquirer had Obama +8 and at 50%, while a conservative firm had Romney within 2 (a huge, unreasonable shift). Same thing in Michigan (large Obama from a news outfit poll, +2 Obama from a conservative outfit). In Florida, we see Obama +1 for NBC/Marist and Romney +7 with Mason Dixon (conservative outfit).
From this we can see a few things. First, judging from the numbers from non-conservative outfits, Romney's debate bump was ephemeral in the states, and is receding in the national polls. The electoral college has remained essentially unchanged and Romney has very few paths to victory. Second, we see that the Republican polling outfits appear to have used the debate as a cover for a sustained effort at artificially changing the narrative (more on this in a moment). Finally, we also see how these polls, when dumped into an aggregator or forecasting model indiscriminately, serve to narrow the race in ways that simply don't gel with reality.
Now, it seems evident to me that the conservative strategy has been to piggy-back the media narrative that Romney won the debate (a narrative I have repeatedly pointed out is bull, and have looked upon Kossack repetition of it with sorrow) with a wave of misleading (to put it politely) polls to keep the narrative alive and provide the media with a perception that the debate actually changed the trajectory of the race. Beyond simply raising (admittedly mild) self-fulfilling prophecy concerns, this should raise alarms for anyone who remembers the 2000 election or has been paying attention to open attempts to steal the election through manipulation of the electoral system (hello voter ID laws and voter purges). Why? Well, should the conservative pollsters provide enough crap data for the race to appear close not only nationally, but in states like Michigan (R governor), Ohio (R governor), Florida (R governor), Pennsylvania (R governor) and Virginia (R governor), all states that have had voter purge attempts and/or voter ID laws, then the voter suppression, voter fraud, and vote manipulation (all things Republicans are not above doing, as Nathan Sproul unintentionally informed the nation) will be more difficult to detect. Against cries of foul play, Republicans can point to an overwhelming number of Republican polls from conservative hack firms that show a "close race" for plausible deniability. In addition, if the public has been getting stories from news outlets and forecaster predicting a tight race, there may be less public outcry and more public indifference towards charges of vote manipulation and electoral malfeasance. Now I don't tend to get into conspiracy theory crap at all, but there is a tipping point where all things begin to point to the same conclusion. As a great adventurer once said,