15 October 2008

More Polling Porn

Pure horse-race stuff, via Nate at 538.com:
Yikes. There's a lot to like there. Early voting has started in a few states, and absent some historic event in tonight's debate, Obama can drop into the prevent defense for the next three weeks and sit on this lead. Of course, that's not what he's going to do; he'll push hard right to the end. The margin keeps growing, and McCain is running out of time to make it up.

But I continually remind myself that the Chicago Cubs were the best team in the National League this year. That didn't work out too well, did it?

Three more weeks.


Update: another cool thing about 538.com is that on the right side bar, down a bit, EVERY state poll from EVERY state is listed. (There are 45 from Ohio alone!) Nate weights them, pulls out party bias, trends and does a regression model to try to predict the current state of the state:


Here's his summary for Ohio:
I like the numbers, but as a politics nerd, I love the data!

11 comments:

  1. I don't understand because the liberal medial outlet CNN indicates that Colorado is a toss up. Same with Ohio. Weird.

    Oh, BTW, I saw an Obama ad. In Arizona! That's just showing off.

    ReplyDelete
  2. The Ohio poll shown on CNN or MSNBC (not sure which was on) has different results. It shows a slight lean toward McCain, 48-46.

    Call me stupid (I've been called worse), but I don't understand why there are such broad differences between the national polls, anywhere from +5 to +13 Obama. Even the larger polls, where my naive mind thinks there might be a greater confidence interval, shows a difference of +5 vs. +10.

    To my mind, all this says is that Obama appears to be leading by some margin that McCain may or may not be able to close, and that neither the young cell-phone-only voters nor the won't-admit-I'm-a-racist voters can be counted. IOW, it still seems to me the race could go either way.

    If I'm wrong, what am I missing, or what is my critical thinking error?

    ReplyDelete
  3. 8:05 anon:

    The poll spread is two factors. One, each poll has a margin of error between 3% and 5%. Thus, some results are just outliers.

    More importantly, the end numbers can be "gamed" by the pollster. If you look at the polls' internal "cross tabs," you can see the sampling of D's v. R's v. I's. If you want a different end result, you just change how many D's v. R's or I's you input into the poll.

    That is the pollsters' real job, after all...to read the tea leaves and predict which side is going to win the GOTV.

    That said, it's pretty clear that Obama has a lead...how big of a lead is anyone's guess right now. But even a 2% national lead will equate to an impressive electoral win...a 5-6% national lead is akin to a landslide.

    ReplyDelete
  4. anon 805 - um, Anon 838 is mostly right.

    Anon 838,

    The only quibble I would say is that most pollsters (at least the public ones) have a vested interest in accuracy, so I would not say most "game" their numbers. There is indisputably a house effect as they try to figure out how to weight party IDs in this time of shifting party affiliation, and how they account for the cell phone effect, and also the "likely voter" screen. So Zogby has an indisputable pro-GOP bias and Research 2000 pro-dem. But they're not gaming them to mislead -- they honestly believe their model is better, and they stake their credibility on it. (Zogby has very little credibility any more, sadly.)

    Note: 538.com is awesome because you can pull up *all* the polls from a state, and also because in their larger modeling, 538 pulls out the "house effect" of all the polls to try to get a consensus result.

    ReplyDelete
  5. I know I'm not supposed to interrupt your premature celebration with mere reality, but you may want to check out the TIPP poll. TIPP was the most accurate presidential poll in 2004, and it shows the Obama lead at three measly points. That means he's actually behind, my man.

    ReplyDelete
  6. You're damn right it's too early to celebrate. Polls don't give you EVs.

    Having said that, the TIPP poll doesn't exactly inspire terror in my heart either. One data point among many.

    I remember in 2004 at this time when I was desperately grasping at straws that this or that poll showed that Kerry was actually ahead. I think you're doing the same thing, my friend.

    ReplyDelete
  7. I was desperately grasping at straws ... I think you're doing the same thing, my friend.

    You may be right. My record for political prognostication is not good. In fact, when my wife & kids heard me predict Obama's defeat recently, the concensus was that there was no better guarantee of his victory.

    ReplyDelete
  8. TIPP was the most accurate presidential poll in 2004, and it shows the Obama lead at three measly points. That means he's actually behind, my man.
    ----------

    Ahh...I love intellectual dishonesty....

    Note to Catron: it is hard to make an effective point when the second part of your sentence directly contradicts the first part of your sentence, which you state forcefully to lend support to your position.

    ReplyDelete
  9. Er ... you didn't quite get the point, Anon. I was referring to the Bradley Effect.

    It is assumed (by most credible analysts) that the B.E. is causing Obama to overpoll by at least 5 points.

    In fact, Prof. Charles Henry (the man who coined the term) thinks Obama is overpolling by double digits.

    Thus a 3 point lead would put him behind by 2 points (7 points in Henry's judgement).

    ReplyDelete
  10. Catron,

    It'll be interesting to see if there is a Bradley Effect this year (and I did get what you meant, though you were a bit oblique). Nate Silver at 538 posits that there won't be one. His analysis of the primaries showed that, across the board, Obama outperformed the pollster composite by 3 points on average. Of course, these were democratic voters, so it's not clear whether that will apply to the general election. The other argument against there being a BE this year is that, ostensibly, the BE voters tend to show in polls as "undecided," which is to say they claim to be undecided but there's no way they would really vote for a black guy. But there really are very few undecided voters out there. In many polls, you could give *all* the undecideds to McCain and Obama would still win. That may be tightening, I think. Still, there's just no precedent for a black candidate in a national election.

    I hope there isn't. I would hate to see this election decided in such a fashion. It would be depressing, not just for the outcome, but for the evidence of persistent endemic racism in America.

    On that note, did you see the SNL the other day? They address this very question, hysterically.

    ReplyDelete
  11. Yep. Saw the SNL thing, and it was indeed funny.

    As to Nate, I think he's whistling past the graveyard. Here are 4 (recent) examples of African American candidates who have experienced the BE in general elections:

    Kirk (TX Senate race, 2002) overpoll avg = 13.

    McCall (NY Gov race, 2002) overpoll avg = 5.

    Steele (MD Gov race, 2006) overpoll avg = 6

    Patrick (MA Gov race, 2006) overpoll avg = 5

    Note that these were all statewide races. Thus, they are valid precedents for this race because the presidential election is just a series of statewide races.

    Also, the Bradley effect isn't necessarily a measure of racism. It's a measure of social desirability bias. A respondent may dislike Obama simply because of policies, but still be reluctant to admit he's going to vote against him for fear of appearing racist.

    I think it's pretty hard to escape the reality of the BE. The only real question is how significant it will be. In my not-so-humble opinion, Obama won't win unless the polls in the battleground states show him ahead by at least 5 points the day before the election.

    ReplyDelete