It is quite possible to drive yourself crazy following National Polling. In particular, the two pollsters that conduct daily tracking are both among the least reliable. Most here are familiar with Rasmussen. In 2000 when they went under the name Portrait of America they were over 8 points off the final result. Their results in 2004 – 2008 were actually pretty good, but there were reasons to be suspicious about some of their numbers in 2010. Interestingly, though, you can argue Gallup is worse. In 2000 their polling showed wild swings that no one else saw. In 2008 they were the only pollster that showed McCain up by more than 3.
In 1996 I tracked national and state polling. What I found at the time was curious: the state polling was not finding leads as large as those for Clinton (National Polling in 1996 overstated Clinton’s margin significantly). Since then I have found that averages of state polling are less susceptible to wild swings and outliers for a simple reason: they are more of them and there are more pollsters doing state surveys.
In 2008 at Openleft I updated my state polling summary about once a week. I will be doing the same here in 2012 with one difference: I will be using those numbers to predict the probability of the Democrats winning a state. This analysis is based on work done by Chris Bowers in 2010 in his state forcast (I have updated it). A detailed description is here. In addition, I will be running 20,000 election simulations to use these probabilities to predict the results in the Electoral College. I will also be comparing these results against two other race raters: Charlie Cook and MSNBC.
Here are the results using current polling (February – April) against Cook and MSNBC. Note: my polling analysis assumes the race is held today.
The simple truth is the polling averages have, since 2000, simply kicked people like Cook’s butt. A 4 point lead in Florida, for example, doesn’t sound like much. But empirically it is enormous. This is why the results are so dissimilar. Here are the states that at least one rates rates as competitive. Note the enormous difference between the polling in these states and how they are rating.
In defense of both Cook and MSNBC, in many of these states Obama is well below 50. In addition, his job approval rating in these states is negative. There is truth in this. Do I really think that Obama has an 89% chance of carrying Florida? No. But the data from 2004 really create a problem for both Cook and MSNBC. In 2004 Bush ‘s approval rating was below 50 for much of their year. Nor was there any evidence that the old conventional wisdom that the undecided break against the incumbent was true in 2004. So Cook and MSNBC may be proven right. But the data suggests otherwise.
The truth is at this moment Obama is in a strong position, stronger than most realize.
Romney needs something to change. The most likely thing to change is the economy. I have posted this before, but this shows the relationship between consumer sentiment and Presidential Approval. The correlation is strong. Obama is clearly above the trend line, largely because people do not blame him for the economy. Nevertheless, the lesson is clear. Economic weakness would likely change these numbers very quickly.
You can see that in this set of numbers from last September to now. As the economy began to improve, Obama’s position got stronger. A reversal in the economy would almost certainly change these numbers significantly.
What about Rasmussen’s state polls? The following table compares the results from the different pollsters. I want to draw your attention to two sets of numbers. Note that PPP and Rasmussen are actually pretty close with respect to Obama’s approval. The difference is in the margin that each finds. When you look at the crosstabs for PPP, you find about 3% of the electorate is made up of liberals who disapprove of Obama but will vote for him. This makes sense: some here probably agree with that. I have trouble, though, understanding why the case would be reversed. In this sense I don’t think Ramsussen’s numbers are internally consistent
