Tuesday, October 23, 2012

How I'm Following the Polling for the Last Two Weeks

Take it as a recommendation or not; this is what I'm doing right now.

First: I have tabs open to HuffPollster's national average, HuffPollster's big map, and RCP's national average.   I've been insanely, irrationally hitting refresh on the HuffPollster national average...well, often enough to know that the lead flipped at least four times yesterday. Only once twice so far today. Why those three? I really like HuffPolster's model (from Simon Jackman), especially at this point with so many polls showing up. Also, the page loads nicely, the chart is clear and readable, all the good stuff. The big map works well, too. Why RCP? I don't particularly like their simple average, but it's a good reality check, given that it's the one high-quality aggregator which is thought to lean Republican (that is, hosted by a site or run by people with GOP connections. I'm pretty confident that all of these, and TPM's poll-of-polls, are all straight shooters, but just in case something leaks in, it's nice to have balance).

I try very hard to ignore the individual national polls I see over twitter. I'll admit to one bad habit: I'm one of those foolish souls who checks Gallup every day when the new numbers show up. But it's more of a ritual than it is information gathering. For state polling, Taegan Goddard's Political Wire does a great job of collecting the day's polls -- especially if you don't use twitter he's still tops, and if you do you should of course follow him. It's a good way to get a quick look at a bunch of state polls without getting distracted by any particular one. If I see something that stands out, I'll switch over to the HuffPollster big map and click on that state.

That's pretty much the numbers. I'll look at other things once in a while, and I'm aware of what the poll-based predictors say, but I haven't been checking them regularly. That's pretty much what I'm doing.

On top of the numbers, I mostly read Nate Silver's daily summaries, and whatever Mark Blumenthal  writes over at Pollster, plus I look for Harry Enten and Nate Cohn, try to be aware of what Sean Trende is saying, too. Plus I try to read anything that political scientists who blog about polling and elections write (Drew Linzer and Simon Jackman in particular on the numbers, but of course anything at Monkey Cage, Brendan Nyhan, etc., etc.). There are other good people out there, too, and I'll come across those, but that's my basic run. The idea is to be aware of any additional texture or biases in the numbers we're seeing -- but at the same time it's really important not to get into cherry-picking.

I think that's basically it.

I'll also check Nate Silver or Pollster for Senate races, and I see Senate polls posted on twitter. I wish I had more House info.

The only really strong recommendation I'll give you is the basic one: polling averages, not individual polls. The rest? Look, there's no reason to do any of this; we'll find out soon enough who wins the presidency. So this is strictly about what to do if you're a semi-obsessive political junky...you really, really, really want more info, but you also want an honest evaluation of what's going on.

21 comments:

  1. So HuffPollster has a better model than Silver's? Am I understanding you correctly? Have you posted about this before -- could you direct me to that if so?

    ReplyDelete
    Replies
    1. I believe that Silver doesn't necessarily have a polling average, but a predictive model that uses lots of other data on top of the polls. I think that what JB is looking for is a way to follow the latest polling averages. Silver's model is interesting too and I think he said the other day that it's going to gradually shift over to being only based on polling data in the next few days.

      Delete
  2. I use RCP, but all I do is go there and click on the link to the table with the recent polls. I take great pains to avoid RCP's links, which definitely skew to the right.
    In fact, RCP really bothers me in this regard. The Hotline skewed to the right, but did a pretty good job avoiding that skew away from the front item in the excerpted story part (ie, almost the whole thing). The front item was realiably skewed right, but the content in the body allowed you to see what GOP and Dem papers were saying all at once. (I say "was" because I no longer have Hotline access for the last 3 years, and it bugs me).
    RCP, on the other hand...on the left banner, I see 4 videos getting promoted. They've been the same for at least a week. Chuck Todd saying Romney has momentum; Adam Carolla saying something stupid (I assume), and....wait for it....Coulter and Luntz videos.
    I stay away from HuffPo just because I don't want the cheering section there, either; I'd rather not wake up crestfallen on November 7th. Not that Jackman's model is going to be wrong, but that there's just too much ego gratification available in the links around, and I just don't need (any more of) that.

    ReplyDelete
  3. I'd be really curious to know what JB thinks of Nate Silver's model as it seems like the best balance of national polls, state polls and political science/economic fundamentals.

    ReplyDelete
  4. As well as weighted for historical poll quality.

    ReplyDelete
  5. To everyone:

    I like Nate Silver's stuff very much. No slight intended!

    ReplyDelete
  6. So I guess its now time to pay attention to state polling averages(and by definition the EC), given the time and the closeness of the election, as opposed to just focusing on the national polls?

    ReplyDelete
    Replies
    1. Yeah, I probably should run a full item on this, although I've mentioned it in a few posts. There really does appear to be a national vote/EC split. I'm not totally convinced that it's real, but there's enough of an apparent split that I do think it's worth paying attention to Ohio and a few others.

      I've flipped on this, or perhaps followed the data on this, basically over the last week.

      Delete
    2. "There really does appear to be a national/EC split."

      Nightmarish. Hopefully Obama improves in the national polls enough before Election Day and we don't have to have another Gore/Bush style disaster.

      Delete
    3. Remember, though, most of the stress in 2000 was over counting the vote in Florida, not over the "national/EC split." Of course, now that I think about it, I wouldn't mind avoiding that either.

      Delete
    4. One thing I've wondered about for some time is the effect the 2000 election had on partisan opinion of the EC. There are certainly liberals who support the EC, as well as conservatives who oppose it, and I'm not sure there's anything intrinsic to liberal or conservative thought that would lead to supporting or opposing it. But after the 2000 fiasco, I got the sense that many Republicans came to be defenders of it because they saw attacks on it as attacks on the legitimacy of Bush's presidency. Conversely, many Democrats became ardent opponents of the EC because their candidate in 2000 was a victim to its effects. I wonder whether there used to be as much of a partisan breakdown to views on the EC as there apparently has been since the 2000 election, or even if Republicans were once more likely than Democrats to oppose it. (I do remember Bob Dole in an interview stating he opposed it, but that's just anecdotal.) Of course, before 2000 the idea of electoral/popular splits was simply an item of history trivia to most people, and I'd bet the 2000 election significantly increased public awareness of the existence of the EC.

      Delete
  7. Someone needs to do a polling average that excludes landline-only polls.

    ReplyDelete
  8. Am I the only one who likes to check in on Charlie Cook and Ron Brownstein?

    ReplyDelete
  9. One thing I have noticed in the aftermath of The Great Andrew Sullivan Freakout of 2012 is that even using some of these rather sophisticated tools you can construct a narrative of the election that makes the same mistake that pundits routinely make when talking about elections. Sullivan used a lot of screen pics of predictive models like 538 to make his case that Obama "gave the election away" or was "imploding." And if you look at the lines in a lot of models they do take a giant dive right around the first debate. But then around the time of the Biden/Ryan debate he shifted to showing polling averages from HuffPo or RCP to show the "new normal", that is "Obama blew it and is only barley holding on!" It was a trick on two levels, one if he had kept showing the model lines (like 538) there would have been a big uptick starting around the Biden debate that has changed the "Nowcast" predictions of if Obama will win from a nadir of a 56% to a 72% chance. That's a big change! Secondly if he had just used polling averages, the narrative would have been that Obama started going down before the debate and its now just a very close race. Heck if I was a professional blogger and just switched the visual evidence he was using (polling averages first then switch to models after Biden) I could have made the argument it wasn't that bad at all, and the first debate was one of only a few factors. Anyway, the point I'm trying to make is whatever you follow make sure you are consistent overtime and look at where the race is in terms of where it has been on that particular tool all along. That said I am now going to look at Votamatic.org, because it has been predicting a Obama landslide and makes me feel warm and fuzzy inside.

    ReplyDelete
    Replies
    1. I'm a fan of Sullivan's political analysis, but he is just horrible as a poll analyst. He cherry picks polls to confirm his pre-existing biases. Which I would just chalk up as partisan puffery, except his biases disfavor his preferred candidate!

      For instance, he commented yesterday that he is "not currently emotionally capable of accepting" evidence that Obama is still leading! Just bizarre. For a mentally healthy person, anyway.

      Delete
  10. For what it's worth, if you want more House information, I would read Daily Kos Elections. They used to be Swing State Project and thus not tarred (for better or worse) with the Kos name, and each morning, the editors put out a "daily digest" of everything that happened the day before, which you can see on their "live digest" during the day. At the end of the day, they post a "polling wrap" of all national, state, downballot and House race polling, with some commentary. They do tons of work on redistricting and really get into poll crosstabs and demographics. It's due to their information that I can meet someone from somewhere in California, ask him or her who his or her representative is, and suddenly comment on the state of that House race.

    elections.dailykos.com

    ReplyDelete
  11. Sam Wang over at the Princeton Election Consortium may be the most accurate of all, and it's my understanding he has been quite accurate with elections recently (he and Silver recently did a podcast for NPR, fwiw).

    Anyway, he doesn't take into account national polls for his model, because they have ZERO effect on the outcome of the election, which is entirely based on the Electoral College (state-based, obviously).

    So I guess my question here would be: why look at national polls at all, when it won't affect the outcome? Wouldn't it make more sense to follow only state-polling, which tends to not only be more accurate but actually takes into account polls that affect the outcome? Seems that with national polls you are really only setting yourself up for possible error.

    ReplyDelete
    Replies
    1. The fallacy in your argument is that you're assuming that because the election is decided by electoral votes, therefore national polls must be less reliable. That doesn't necessarily follow. There's a very close connection between the popular vote and the electoral vote, which literally almost always have the same winner. Electoral/popular splits have happened only three times in our history, and in two of them (1876 and 2000), it's debatable whether the divergent electoral results were valid.

      It's true that the advantage Obama holds in Nate Silver's model is almost entirely based on the state polls. But that isn't because Silver believes the national polls are irrelevant to predicting the race; on the contrary, he factors them into the model as well. Nor is it because he sees an electoral/popular split in the works; on the contrary, he estimates there's only about a 7 percent likelihood that Obama will win the electoral college while losing the popular vote.

      Rather, he suspects that the state polls may be providing a more accurate picture of what's happening on the national level than the national polls do. He's considered that he may be wrong about this, and that the opposite may be the case: that Obama's advantage in the state polls may be an illusion, and that Romney's advantage in national polls might be a better indicator of what will happen at the electoral level. Silver explained all this a few days ago.

      Delete
    2. It's all about the size of your sample, gentlemen.

      Delete
  12. I think you are a "national polls only" guy, for the most part. So you should regularly look at your opposite number: Sam Wang. He's a "state polls only" guy, for the most part. With a great record.

    ReplyDelete
    Replies
    1. Yep, he was pretty much exactly right in 2004 and 2008

      Delete

Note: Only a member of this blog may post a comment.