Obviously it's to do with when voters made up their minds. Let's compare this with exit polls of actual voters. The times are not exactly compatible. For the exit polls, for example, I collapsed "just today" with "in the last few days" to resemble Pew's "within a week." It works fairly well, as you'll see below.
|Pew Survey||Exit Polls|
As you can see, the survey by Pew of 1,206 voters matches up damn well with the exit poll data collected from a much larger sample of voters as they left the ballot box last week.
What can we take away from this, other than Pew knows how to run a good survey? Mainly that not many people made up their mind late in the game and those folks tended to split evenly between Obama and Romney -- thus negating the whole Hurricane Sandy hypothesis. I will point out that among those who decided on Election Day, they cut 51-44 to Obama. But those were only 3 percent of the electorate.