It's a NEW Day!!

October 23, 2004 11:15 AM

Hey! Long time no talk! How are you? I'm pretty good. Lots to tell you. But I'm going to save that. I wanted to post this as a kind of public service message. I know the closeness of this presidential election is driving me batty with suspense, and I've been obsessively watching poll numbers, which are driving me even crazier with their lack of substance.

If you, like me, are desperately looking for some well-reasoned 2004 Presidential election predictions, I've found some great sites that may make you feel like you are actually getting some solid information and analysis instead of just more SPIN SPIN SPIN.

The first is electoral-vote.com. This guy subscribes to most premium poll services, and so he has access to daily state-by-state data from many sources. Why chase down 20 different polls when he's already doing the homework for you? Every morning, he updates his map with the latest data, and then calculates the bottom line . . . the electoral college votes. So much more helpful than those useless nationwide polls that tell you how many people are leaning toward each candidate. Because as we learned in 2000, the popular vote doesn't really matter. It's all about the college.

My favorite part about this site is that the data is transparent. He uses all available polls, without bias, except those from polling agencies caught doing "push polls." With a wave of your mouse, you can see exactly where his data for each state is coming from.

But while he uses all polls, he provides some interesting data on individual pollsters to help you, the reader, form your own opinion about whether certain polling agencies might be "cooking the books." He also shows you how accurate the pollsters were in the 2000 election.

Finally, he has a cool animated map that gives some great perspective on how things have evolved over the last few months . . . and he shares an interesting piece of trivia that I hadn't heard before: For the last 50 years, voters who are undecided at this point in the election process have ended up voting two-to-one in favor of the challenger. (This is apparently an accepted truth among pundits, and here is some hard backup which seems to actually prove it.) So if you are looking at a poll that says "47% Bush, 45% Kerry, 6% Undecided," recognize that the poll is probably actually pointing toward a final number of 49.7% Bush, 50.3% Kerry.

Another site I've found worth noting and using as a point of comparison is electionprojection.com. This site is not as "user friendly" as electoral-vote.com. The public site is only updated weekly, so the polls he uses are quickly outdated. The sources of his state poll numbers are not as transparent, and his methodology is fairly arcane, which always makes me suspect some kind of shell game. He takes recent state poll numbers and weights them with a factor based on 2000 election results and current national polls. The result is not that much different than if he had just posted the state poll results with no modification. But the weighting does have the effect of making swing states appear a little more likely to vote the way they did in 2000, and a little more likely to vote for Bush if his national job approval rating is good. These assumptions seem reasonable, if not entirely "scientific."

For example, the website states "Job Approval is the one statistic that most closely correlates to actual vote totals for an incumbent president." I can't find any evidence for this correlation. I did find two academic papers that suggest a national job approval percentage of 50% or more in July of an election year is a very strong predictor of electoral success (Bush was at around 47% in July). But neither of these papers support the idea that there is any reliable statistical connection between job approval numbers right before the election and actual votes cast for the incumbent.

(pdf files for paper 1 and paper 2)

Also, electionprojection.com ignores the "undecided" poll results, which is equivalent to assuming that undecideds will ultimately split between Bush and Kerry in the same proportion as the decideds. In fact, the hard evidence cited above suggests that they are likely to cut substantially in Kerry's favor.

Despite these differences in methodology, though, the two sites are fairly consistent in their conclusions, which is a good sign. The authors of both websites have personal partisan allegiances, which are freely stated, but they seem to have made every good faith attempt to keep their number-crunching systematic and impartial. But if there is any such thing as "unconscious bias," these two analysts are clearly "biased" in opposite directions.

If two reasonable methods, devised by different people with different approaches and opposite personal convictions, come up with the same answer, you can bet that you're as close to "true" as you are likely to get. The few places where the two sites differ are the ones that bear the most attention and reflection. For example, on 10/17, the last time electionprojection.com was updated, the only significant difference between the two sites was Florida . . . electoral-vote.com called it completely split and didn't assign it to either candidate, making the race a toss-up. Electionprojection.com was counting it for Bush by 1%, predicting an overall Bush victory.

Since then, new poll results and a drop in Bush's approval rating have resulted in both sites assigning Florida to Kerry. Electoral-vote.com has also shifted Ohio over to Kerry. If Kerry gets Ohio and Florida, Bush is pretty much down for the count.

Are these two prophets showing us the earliest solid indicators of a Kerry victory? Tomorrow, electionprojection.com will post its weekly public results. We'll all get to see how they dovetail with the electoral-vote.com numbers, and what that means for the election.

I can't wait! I'll bring the popcorn if you bring the pretzels.

TRACKBACK (57)


	October 23, 2004 11:15 AM Hey! Long time no talk! How are you? I'm pretty good. Lots to tell you. But I'm going to save that. I wanted to post this as a kind of public service message. I know the closeness of this presidential election is driving me batty with suspense, and I've been obsessively watching poll numbers, which are driving me even crazier with their lack of substance. If you, like me, are desperately looking for some well-reasoned 2004 Presidential election predictions, I've found some great sites that may make you feel like you are actually getting some solid information and analysis instead of just more SPIN SPIN SPIN. The first is electoral-vote.com. This guy subscribes to most premium poll services, and so he has access to daily state-by-state data from many sources. Why chase down 20 different polls when he's already doing the homework for you? Every morning, he updates his map with the latest data, and then calculates the bottom line . . . the electoral college votes. So much more helpful than those useless nationwide polls that tell you how many people are leaning toward each candidate. Because as we learned in 2000, the popular vote doesn't really matter. It's all about the college. My favorite part about this site is that the data is transparent. He uses all available polls, without bias, except those from polling agencies caught doing "push polls." With a wave of your mouse, you can see exactly where his data for each state is coming from. But while he uses all polls, he provides some interesting data on individual pollsters to help you, the reader, form your own opinion about whether certain polling agencies might be "cooking the books." He also shows you how accurate the pollsters were in the 2000 election. Finally, he has a cool animated map that gives some great perspective on how things have evolved over the last few months . . . and he shares an interesting piece of trivia that I hadn't heard before: For the last 50 years, voters who are undecided at this point in the election process have ended up voting two-to-one in favor of the challenger. (This is apparently an accepted truth among pundits, and here is some hard backup which seems to actually prove it.) So if you are looking at a poll that says "47% Bush, 45% Kerry, 6% Undecided," recognize that the poll is probably actually pointing toward a final number of 49.7% Bush, 50.3% Kerry. Another site I've found worth noting and using as a point of comparison is electionprojection.com. This site is not as "user friendly" as electoral-vote.com. The public site is only updated weekly, so the polls he uses are quickly outdated. The sources of his state poll numbers are not as transparent, and his methodology is fairly arcane, which always makes me suspect some kind of shell game. He takes recent state poll numbers and weights them with a factor based on 2000 election results and current national polls. The result is not that much different than if he had just posted the state poll results with no modification. But the weighting does have the effect of making swing states appear a little more likely to vote the way they did in 2000, and a little more likely to vote for Bush if his national job approval rating is good. These assumptions seem reasonable, if not entirely "scientific." For example, the website states "Job Approval is the one statistic that most closely correlates to actual vote totals for an incumbent president." I can't find any evidence for this correlation. I did find two academic papers that suggest a national job approval percentage of 50% or more in July of an election year is a very strong predictor of electoral success (Bush was at around 47% in July). But neither of these papers support the idea that there is any reliable statistical connection between job approval numbers right before the election and actual votes cast for the incumbent. (pdf files for paper 1 and paper 2) Also, electionprojection.com ignores the "undecided" poll results, which is equivalent to assuming that undecideds will ultimately split between Bush and Kerry in the same proportion as the decideds. In fact, the hard evidence cited above suggests that they are likely to cut substantially in Kerry's favor. Despite these differences in methodology, though, the two sites are fairly consistent in their conclusions, which is a good sign. The authors of both websites have personal partisan allegiances, which are freely stated, but they seem to have made every good faith attempt to keep their number-crunching systematic and impartial. But if there is any such thing as "unconscious bias," these two analysts are clearly "biased" in opposite directions. If two reasonable methods, devised by different people with different approaches and opposite personal convictions, come up with the same answer, you can bet that you're as close to "true" as you are likely to get. The few places where the two sites differ are the ones that bear the most attention and reflection. For example, on 10/17, the last time electionprojection.com was updated, the only significant difference between the two sites was Florida . . . electoral-vote.com called it completely split and didn't assign it to either candidate, making the race a toss-up. Electionprojection.com was counting it for Bush by 1%, predicting an overall Bush victory. Since then, new poll results and a drop in Bush's approval rating have resulted in both sites assigning Florida to Kerry. Electoral-vote.com has also shifted Ohio over to Kerry. If Kerry gets Ohio and Florida, Bush is pretty much down for the count. Are these two prophets showing us the earliest solid indicators of a Kerry victory? Tomorrow, electionprojection.com will post its weekly public results. We'll all get to see how they dovetail with the electoral-vote.com numbers, and what that means for the election. I can't wait! I'll bring the popcorn if you bring the pretzels. TRACKBACK (57)		Your Words