Primary Polls Still Aren’t Very Predictive

Nate Cohn: “You have undoubtedly heard that primary polls aren’t necessarily very predictive far from an election. With just a month to go until the Iowa caucuses, I’m writing to tell you that … it’s still true.”

“This phase of the race — the final stretch before Iowa and New Hampshire — can be the most volatile of the entire campaign, as early-state voters make up their minds, politicians and newspapers make endorsements, and candidates make strategic decisions to invest time and money in particular states.”

“In recent primary campaigns, going back to the 2004 Democratic primary, those candidates who have led in Iowa or New Hampshire polls with just one month to go have lost as often as they have won. On average, candidates’ share of the vote at this stage differed from their final share of the vote by around seven percentage points. With many candidates running, it was not at all uncommon for a candidate to move by more.”

FavoriteLoadingSave to Favorites
  • enfuegobuddha

    According to FiveThirtyEight’s swing-o-matic, if you change the white voter turnout to 100% and leave the Democrat:Republican ratio the same across all races, the Democrats would still win the electoral vote in the general election.

  • Trevor Patrick

    This is starting to become the catch phrase of data based pundits this year, that the polls don’t matter. I think they’ll be saying it in the middle of March, that it’s too early for any polls to matter. While it is true in some cases, I don’t recall the intense focus on how useless polls are back in 2012 when Romney was leading

    • Gumby✓ᴶᵒʰⁿˢᵒⁿ

      Romney didn’t lead the polls much of the pre-primary period. Remember the flavor of the month? Gingrich was leading national at this point in 2012.

      • Trevor Patrick

        he led for a while, it’s true the flavor of the month rose and fell, but romney was a solid number 1 or two throughout. the flavors of the month only led for a month or so before falling, whereas romney slowly gained support. the only candidate this cycle with any similar stability has been trump. http://elections.huffingtonpost.com/pollster/2012-national-gop-primary

        • Gumby✓ᴶᵒʰⁿˢᵒⁿ

          All that is true, but it was also before the mainstream rise of the political statisticians like Silver and Cohn to explain poll predictably. Combined with the FOTM phenomenon, you can see why this wasn’t discussed much in 2012.

    • Some_guy_with_something_to_say

      This particular piece is speaking specifically to primaries, where there are two huge challenges: predicting geographic and demographic shape of the election day voter pool, and the fluidity of voter preference in a multi-candidate, single-party primary. The candidates all have much more in common than they do separating them.

    • enfuegobuddha

      I think the real story here is that polls are becoming increasingly less accurate over the years. Pollsters haven’t quite figured out their craft in the cellphone/internet era.

      • Gumby✓ᴶᵒʰⁿˢᵒⁿ

        I doubt they have ever been terribly predictive this far out. Hell, Jimmy Carter was up 59-34 on Reagan on 1/4/1980

        • enfuegobuddha

          I meant that the accuracy of the polls even close to the election have perhaps gone down, and that will be the bigger story.

      • embo66

        It’s not just a tech / lifestyle problem but an attitudinal one. More people than ever before are refusing to participate in polls. Hatred of politics / general distrust seem the likely culprits here.

        • Gumby✓ᴶᵒʰⁿˢᵒⁿ

          Deleted.

  • eve

    Michele Bachmann — Iowa, 2012

    Sort of related to this article.

  • rssrai

    Romney was leading the Iowa primaries in January. Santorum actually won, but only by .01 percent. Romney actually got the credit for the win in Iowa because the final tally was after NH. I think that primary polls for Iowa in January are pretty predictive. President Obama was ahead by 1.6 percent in early January and won by 7.8 percent. Edwards and Clinton split their votes. O’Malley will not split his votes with Hillary, but could with Senator Sanders since he is trying to run on the same kind of platform.

    • Gumby✓ᴶᵒʰⁿˢᵒⁿ

      The 2012 and 2008 Iowa Caucuses were January 3. They are February 1 this year. I would expect a poll 2 days out to be predictive. Polls one month are less so.

      • rssrai

        DUH!. You are right. I am obviously in la la land. LOL.

  • Delphine

    This will give poor Quin Hillyer heart. He is still boosting for Santorum over at American Spectator, I think I saw this morning.

  • L’Homme Armé

    As I’ve mentioned a couple of times in the past few months, it seems like the second week is when the the shit gets real as far as primaries are concerned. So keep a close eye on the polls in the next couple of weeks, as the movement that happens then is likely more predictive than anything that happened last year.

  • pywaket_1

    “Going back to the 2004 Democratic primary”: considering that 2012 was uncontested, we also call this “2 of them”. Huge sample size you’ve got there, Nate. And you wonder why people think you’re a bad joke of a journalist.

    • Gumby✓ᴶᵒʰⁿˢᵒⁿ

      If you actually read Cohn’s article, you would realize that he was including both GOP and Democratic primaries in his count.

      • pywaket_1

        Terje beat me to it. This not a sample. It’s like saying the Red Sox are the most successful baseball team ever, based on what’s happened since 2004.

        • Gumby✓ᴶᵒʰⁿˢᵒⁿ

          That’s fine, but 4 is still more than 2.

          • still?

          • Gumby✓ᴶᵒʰⁿˢᵒⁿ

            Yea, amazing, isn’t it?

  • PoppyLee

    True, the primaries can be totally unpredictive, but they sure are fun and full of surprises. It’s because the process to vote in some of these states are so unpredictable. Do people really want to spend hours in Iowa caucusing?