Comment from Josh Hammond
Time: July 27, 2008, 5:28 pm

So much for the kill-joy polls that pronounced NO bounce for Obama from his trip overseas BEFORE he finished the trip. I believe when he left Obama was about 2 points ahead. A 5 point bounce ain’t bad. But a caution here on all polls. This one is based on registered voters and therefore it is not as “realiable” as those based on likely voters. Soon we should start seeing polls based on likely voters.

Comment from Pat
Time: July 28, 2008, 8:28 am

Pollster.com has a chart that shows that states with 284 electoral votes - 14 more than the 270 needed to win, lean to or are firmly supporting Obama, while states with 147 electoral votes lean towards McCain. The 10 states with 107 electoral votes are tossups. In other words, they claim that Obama doesn’t need to win any of the tossup states in order to win the election.

Comment from timr
Time: July 28, 2008, 10:50 am

John what do you think about Pew Research? Have they been accurate in the past? Just how does any organization poll those who are 30 and under who don’t have land lines?

Comment from John McCreery
Time: July 28, 2008, 11:50 am

Timr, the honest answer is that I don’t really know. I’ve read stuff that suggests combinations of different polling methodologies, e.g., landline + Internet + Mobile messaging, but I’m not up on the details.

The striking thing about Gallup is that Gallup sampling has tended to be biased to the right, so that if Gallup shows Obama up three days in a row (now four, with Obama 48%, McCain 40% today) that’s a conservative estimate of the difference.

Comment from Josh Hammond
Time: July 28, 2008, 3:32 pm

Timr, remember that all polls, if properly drawn (random representative sample) and the proper size, are “accurate” at that point in time. The problem is that any delay in publishing the results, any delay, changes the “accuracy”. That’s one reason the Gallup tracking survey is more reliable than most.

To exaggerate this point, I recall working with companies who wanted to survey customers or employees. When the survey was done, the survey company would provide an analysis a few days later. Then the company would study the results for days, weeks, even months, yes, even months. Then the company would decide internally what the best spin would be on the findings. Then the PR people would…well, you get the idea. The amazing thing, though, was the companies felt they were on to something with the survey findings only to find out that employees and customers had moved on. Political surveys aren’t that bad, but politics is more volatile than someone’s attitude about Buick or Coca-Cola or thier boss.

Pew, as I mentioned before, is a non-profit organization that does its own studies. They don’t have clients like Gallup, Hart, etc. If someone is paying for a survey, they have something they want from it and they look for it in the results. This is where Pew can do different things and why their findings are more reliable.

Comment from Sasha
Time: July 28, 2008, 3:59 pm

pollster.com has done a series of investigative pieces on which polls are done the best and which are the most accurate over time. While it is a lot to wade through, anybody with a serious interest in prediction should take a look at the lot. Here are a few to get you started:

Rating the IA Polls
Telling Good Polls from Bad
Survey USA’s Poll Report Card where you will find Gallup near the bottom in accuracy
Some methodological notes on same that you probably don’t care about

And, in you want a moment of joy, check out their electoral map.