As it turns out, my last diary about polls was not my last diary about polls. I have previously posted about the suspicious up tick in polls from conservative outfits after the first debate (here), and how those polls have been affecting the margins of the polling averages to drive the narrative of "Mittmentum" and make the race appear closer than careful analysis would suggest. Previously, I compared the polling averages of the swing states with all polls and without the polls from ideological organizations with either no history or a history of biased, inaccurate polling to look at the state of the race without the ideological, hack pollsters (here).
Since we are nearing the election and most firms are releasing their final pre-election polls, I figured I would pull up the polls one more time and compare the numbers to the election results in a couple of days. Such analysis can give some preliminary insights into the effects of the post Citizens United proliferation of ideological interest groups and their polling on pre-election polling aggregates and how these ideological polling firms can be used to drive the media narrative surrounding the race. One thing to be noted in the front-end: since this is the last wave of polls before the election, I anticipate the polls from the ideological firms to be using more rigorous methods to come more in line with other pollsters and protect their mainstream credibility for future elections. Should their last polls more accurately predict the outcomes of the states, the firms can claim to be unbiased and can point to their final polls as proof of credibility and accuracy. Consequently, I expect there to be a much smaller difference between the full polling universe and the non-ideological/non-partisan pollsters alone than there was in my previous post.
The methods are covered in the link above, and I am just repeating those methods here. As a reminder, the list of firms that will be removed is as follows: Rasmussen, Gravis, ARG, Purple Strategies, WAA, Mason Dixon and the smaller Republican outfits, PPP polls done on the behalf of left-leaning groups, Project New America polls, and any polls done by smaller Democratic affiliated firms. This time around, I will not be looking at the "first debate bump," as the notion has been widely debunked (including in my last analysis). The updated analysis below the fold.
IMPORTANT NOTE: Pollster updates their composites to include all samples from a given time-frame once the polls are released. The lag time means that a poll with a sample from time-frame in the charts below released after I have written my analysis will change the charts themselves when they are incorporated. Discrepancies between numbers in my analysis and numbers in the charts are created by this, not my inability to do basic subtraction, and certainly not intentional misrepresentation of the data on my part. I will try to update the analysis as the numbers change.
Since Ohio is President Obama's electoral college safeguard, I will start there. As you can see, the inclusion of all polls from all firms shows Obama up 3.4%. Solid lad and .6% higher than the 2.8% president Obama showed in my last post's polling universe (a week ago).
True to form, dropping the ideologues brings us an Obama lead of 4.1%, showing the ideological firms coming into line with the rest of the polling universe to protect credibility. In my last look at this, Obama led by 5.1% among the polls from non-ideological outfits, which reflected an Obama +2.7 point difference in the margin when the ideological pollsters were dropped.
The current polling universe has Obama up 2.4%.
Removing the ideological pollsters and Obama's lead increases to 3.1%, or tie for all intents and purposes. As I wrote before, this may be because Romney has been campaigning there since the 2008 primary, and while winning every state is the goal, New Hampshire make very little EC difference.
Last time, Iowa showed a solid lead for Obama in the polling universe and it grew to a large lead when the ideological pollsters were removed. Now, Obama is up 2.6% in the averages for the whole polling universe.
Interestingly, now, dropping the ideological firms shows the opposite effect on the polling averages. The race tightens very slightly to an Obama 2.1% lead. Again, the removal of the ideological firms only changes the margin slightly, indicating the ideological firms are coming back in line.
The current state of the race, including all polls, is an Obama lead of 1.6%.
A week ago, removing the ideological pollsters changed the race to an Obama lead of 2.4%. Now, removing ideological pollsters again reflects a race more in line with the full universe of polls showing Obama up 2.3%. While it does reflect a larger Obama lead, the difference with and without the ideological firms is much smaller this time around.
In Virginia, Obama currently leads in the polling universe by 1.9%.
Removing the ideological pollsters shows a larger lead for Obama of 3.6%.
Florida was an interesting case before. A week ago, removing the ideological pollsters actually swung the state from a slight Romney lead (0.7%, essentially a tie) to a 2 point Obama lead. In fact, without the ideological pollsters, the Florida trajectory seemed steadily in Obama's favor by one or two points for the entire cycle. The current polling universe shows Romney leading Florida by 0.5%.
However, now, removing the ideological pollsters changes the margin in Romney's favor. Romney leads by 2.2%. Now, Project New America has been active in the state in addition to the conservative outfits. I am not a believer in the false equivalence that left-leaning firms are equally likely to manipulate the data to drive a narrative (indeed, Nate and others have presented evidence confirming that left leaning firms do not bias as deeply as conservative outfits), but for the sake of argument, I did drop left leaning firms like Project alongside Gravis, ARG, and Ras. Dropping Project polls that may have valid toplines may explain the unique case of Florida. Or, it could be the case that Romney is actually leading Florida by more than we thought (given early voting numbers, however, I doubt it).
The polling universe currently shows a 1.5% Romney lead.
Sans the ideological polling, Romney's lead drops to a .5% lead, essentially a tie.
A More Direct Comparison and Some Concluding Thoughts
The table above shows the final top-lines in the the swing states from my last polling analysis of this kind (from October 26th) and the current top-lines from swing states from the final group of polls (from November 5th). In my last post on this topic, I will add the final vote margins in these states and provide some concluding thoughts comparing the polls to the final votes. For now, some thoughts on what I am seeing. Caveat: there are several ways to read what my analysis is showing, and it's important to bear in mind that until tomorrow, when I look at the vote counts, there's no way to confirm my educated guesses. We will have a better idea Wednesday, but I wanted to lay out my hypothesis in the front end so we can either confirm or reject them.
First, the polls from last week to this week could be expected to narrow. The race always narrows just before election day, as undecideds finally pick a team and base voters from both sides rally for the home stretch. I believe that explains the slight tightening of the non-ideological numbers from last week to this week.
Second, as the race comes to an end, I believe that the ideological firms used more rigorous, industry standard methodology for their final polls to protect their credibility and use as evidence for future races. You can see this in both Gravis and Ras, both of which had difficulty bringing themselves to show Obama leads and both of which who now reflect the general consensus (or close to it) in most state polls. As you can see above, the margins last week made the race appear much closer when the ideological polls were included than when they were dropped. The difference between the full universe of polls and the non-ideological pollsters was substantial in most states, and made the race appear extremely close; however, the difference between the margins in all polls and the margins in non-ideological polls was much smaller this week, in the final round of polling. In four states, the race looked like the margin was less than two points. In pivotal Ohio, Romney was only down by less than three. The appearance in the aggregates drove a narrative that kept Romney's base hopeful, and may have even created some self-fulfilling effects (as Kos correctly points out, people work harder for a perceived winner).
After weathering a media narrative about "Mittmentum" that never really existed, as I showed in my previous post and as Nate more conclusively demonstrated, I think we can all appreciate the importance of examining how Citizens United and the flood of corporate money can be used to even influence the data by which we measure elections and public opinion and the media narrative driven by that data. Whether we like it or not, traditional media still sways public thought and political discourse. We have seen a lot of lazy reporting this cycle, infuriating false-equivalence, and maddeningly simplisitic data analysis built on cherry picked polls. Polling aggregates inform traditional media headlines, and if they can be influenced and manipulated by the proliferation of hack polling firms, we need to identify these affects and absorb them into the way we look at and analyze election data.
The Maps Predicted by Non-Ideological Firms
Before, on October 26th, the map predicted by the polling averages excluding ideological pollsters looked like this:
Including the week of polling since, the projected map from non-ideological polls (and the full polling universe, but with smaller margins) looks like this:
Excluding the ideological pollsters on October 26th showed President Obama logging 328 electoral votes. Now, with the final polls in, the non-ideological firms show President Obama getting 303 electoral votes (basically, Florida flipped from last week).