Poll Dancing: How accurate are opinion polls?

Thursday, May 6th, 2010

Poll Dancing: How accurate are opinion polls?
It’s voting day in the UK General Election today. Here’s a hackerly look at the accuracy of opinion polls for The Guardian Datablog. See if you can spot any patterns.

(They’ve run the image a bit small. Here’s a hi-res on my Flickr)

Here’s the data http://bit.ly/polldancing too. It has a load of extra info going back to 1987.

Happy voting!

 

Tags —


Credits —


DESIGN: David McCandless
RESEARCH: David McCandless, James Key
ADDITIONAL DESIGN: Matt Hancock, Joe Swainson
SOURCES: Guardian/ICM Poll Results, YouGov Trends (PDF), Ipsos MORI Research
DATA: Explore in this Google doc

Related

  • http://andrewsteele.co.uk/ Statto

    It’s easy to attach too much significance to the random motion of polls if you don’t include error bars. Check out my graph of polls in the run-up to this election with uncertainties.

  • Esme

    You might want to fix your legend “TIME BEORE ELECTION” instead of “BEFORE”. Also, I suppose I know which parties the colors represent, but a key would still be nice.

  • Michael

    Agreed on a key being helpful. Would love to know what the different colors mean.

  • G

    Not only the key… but also what country and for what office! I’m assuming UK because of timing but is the graph depicting New Zealand? Wisconsin? Co. Kerry?

  • Jack

    I hope you will update this when we know the final result of the 2010 election? I’d love to know how they did this year as the result seems SO unexpected!

  • http://twitter.com/fatmike182 fatmike182

    Awesome & funny…
    probably you could add even more content if you display the size of the sampled population.
    Great work again!

  • http://www.lxfm.com brighter day gang

    Revised flickr comment: As a layman and only looking at your visuals, it appears to me that polls are becoming more accurate. But in context, through my awareness as flawed as it may be, it seems that the margin of error MUST become smaller and smaller in order to keep up with an increasingly informed population, therefore putting the poll in the place it has always been in. That is, the more accurate these polls become via improved technology, the more accurate they NEED to be as more individuals are becoming better informed and better able to handle changing information on a DAILY/HOURLY/MINUTE/SECOND basis because of technology. More recent data will be telling, I’ve heard they’ve been using facebook/twitter to gather data. When the time is right, the poll will become the vote. Or vice versa. Or none of the above. I like to cover all my bases. win/win. goldman sachs. lizard people.

  • http://www.rhinoplastyguide.com/ Rhinoplasty

    So this is the overall build up of the poll and i wonder how accurate they are and creative as well. This makes the assumption that the procedure is similar enough between many different polls and uses the sample size of each poll to create a polling average.

  • http://www.r4-ds.es/ acekard 2i

    It is great information as I want to know that polls really accurate or not. But after reading the above article I came to know most of all things related to poll. It is really great help and also useful information.

  • Francis

    The graph is fundamentally flawed. Through the grey line in the middle, it assumes that the result of the election would be the same through the time period. This is not the case, cf. “one week in politics is an eternity”, etc. In reality, the pools may have been very accurate on a daily basis and we will never know this unless we hold an election on the same day of each poll. So the only interesting data is that of the actual day of the election. All the rest is irrelevant, at least for this analysis (“which poll is most accurate during the campaign”).