Monday, November 5, 2012

Election Eve and Polling Bias

Well it's election eve and Nate Silver is still predicting an Obama win....with the caveat that it is possible that if Romney wins it will mean nearly all state polling might be biased against Republicans.

I don't think he was saying this to be glib, or ruling the possibility out.  He actually goes quite in depth as to where he thinks error could occur.

To me though, this brought up an interesting point.....what do we do if it's true?  If nearly all swing state polls are saying Obama, and they break Republican, we will have to do quite a bit of reworking of our polling system.  But that's not what this post is about.

This post is actually about a rather entertaining comment I saw in a discussion about this.  Why haven't there been more concentrated efforts to skew polls?  Essentially, if you live in a swing state and hate political advertising, why not start a movement to get people in your state to all answer the same candidate to obscure the fact that it was a battleground state and reduce the number of dollars spent there?

This sounds wacky, but how many people would really have to buy in to this to make a difference?

Let's take my home state of New Hampshire.  As of January, there were about 770,000 registered voters.  As of today, polls show they are tied for Obama and Romney.  From what I can find, even the best polls only have a 10% response rate, and many are at 2 to 5%.  The UNH Granite State Poll is widely reported and only surveys 500 people.  It seems it would not take many people making an effort to answer their phones and state they are for a particular candidate to start to skew things.  Even if word got out, it would introduce enough uncertainty in to the polls to confuse the heck out of the political consultants and the media...and wouldn't that at least be entertaining for the rest of us?

It's not like this is was tried with Sanjaya on American Idol and there were rumors about Bristol Palin on Dancing With the Stars.  Those efforts took far more people than it would take to skew the polls in a small state like New Hampshire.  With 58% of adults using Facebook to get political information, it shouldn't be too hard to mobilize people....just like Twitter was used to start chants at the Boston Garden during the playoffs last year.

This is the danger of big data.  While data driven decision making is awesome, it's also hackable.  I'm just curious what the back up plan is if polls don't work any more.


  1. I have heard this idea floated many times, but it never seems to come to pass. I suspect one strong factor is that people really want who they want.

    1. They want to tell you what they really think.
    2. They fear that creating a false optimism or giving too much encouragement to the opposition might each have disastrous consequences.
    3. We all have far more magical thinking, of not giving the bad thing power by saying it out loud, than we would care to admit.
    4. Most people would also suspect it was trick by one side or the other that they didn't want to risk.

    Regarding poll accuracy, it has been messy. Zogby had a few elections where they absolutely nailed the predictions far better than everyone - and then failed spectacularly. Rasmussen was poor in 2000, excellent in 2002, 2004, and 2006, then average in 2008 and poor in 2010. I don't know if this is because polling is so hard to do right, or because turnout actually is fluid up until the end.

    I am grateful that I have not had a single push-poll this year. Of course, I don't accept many survey calls any more, so they may have been out there and I just missed them.

    As for all the people calling us to encourage us to vote for X, I can only think that they know they are annoying the hell out of their friends, but simply can't think of anything else to do that might work.

  2. There's a scene in _Good Omens_ where a demon trapped in an answering machine is released when a telemarketer calls the number. I can dream, can't I?

  3. Called ID must be a nightmare for phone polls. Your mother and I have spent the last two months looking at the phone before answering. We have made an occasional mistake and then simply hang up. Are the folks who actually answer and participate a cross sample? That is a serious question that may be answered clearly today. Or we may be left with the same system. It is always amusing to remember that the pollsters had called the 1948 election for Dewey over Truman based on their telephone polls. It seems that more Republicans had phones then did Democrats.