Ipsos projects that Obama holds an edge in the most hotly contested states, including Florida, Virginia and Ohio, and is likely to win by a relatively comfortable margin of 332 electoral votes to 206 electoral votes.
The accuracy of Reuters/Ipsos online polls are measured using a credibility interval. The survey of 1,030 likely voters has a credibility interval of 3.5 percentage points.
If you have just over 1,000 voters, the number of voters by state is much smaller, and what Ipsos calls its credibility interval is much, much larger.
I trust Nate Silver much, much more, and while I think (as I have said before) that Obama will win, I think it less than 50-50 that he tops 300 electoral votes. Now, for you math geeks out there, I will say that currently, 332 electoral votes (the total shown above) is the mode, or most frequent, result in Silver’s Monte Carlo Simulations. But Silver has that pegged at less than a 1-in-6 chance (see below).
Granted, there’s a roughly 9% chance that Obama hits 347 electoral votes (that’s the Obama surge scenario, where he takes North Carolina), which means there’s approximately a 1-in-4 chance Obama gets to 332 or more. Given Silver’s forecast that Romney has a 30% chance of winning, that means there’s approximately a 45% probability of outcomes between 270 and 331 electoral votes.
I go through all this not just because I’m a nerd, but because I’m reading The Signal and the Noise, Silver’s book about the (frequent) failure of prediction. And the Ipsos/Reuters conclusions are a perfect example of extrapolation from limited data. They may be right, but it’s not a solid prediction.
Which, for the record, doesn’t mean I hope they’re wrong.