I’m teaching my statistics students about “the margin of error” – always great to do in a presidential election year. The papers are full of examples. Some give the margin of error to one or two decimal places and report the exact number of people who answered the survey question.
Around this time in many elections, the race is too close to call. The papers use phrases like “statistical dead heat.” For those whose statistics class was long ago, that means that there is no evidence that one is ahead of the other, or that their percentages are not identical.
In spite of this, every newspaper goes on to report the actual percentages – the ones whose difference is probably due to random chance and not to any real difference in the preferences of the electorate. Then they spend several paragraphs analyzing exactly why Romney or Obama is ahead – even though they’ve just explained that the difference is not real.
One article reported on two polls – both “dead heats” but one where Obama randomly had the lead and the other where Romney randomly had the lead. There couldn’t be better evidence that of the randomness of the lead. Instead, the article was twice as long, probing minute differences in the time frame or methods of the two surveys.
Why can they get the details right, but can’t understand the big picture?
- Oct. 7: National Polls Show Signs of Settling (fivethirtyeight.blogs.nytimes.com)
- Romney takes slight lead in new Gallup poll (cbsnews.com)
- Thirty Days Until the Presidential Election: Fundamentals Still Favor Obama (themoderatevoice.com)
- Almost Everyone Misinterprets The Polling On The Presidential Election – Here’s Why (businessinsider.com)
- One Last Encore For the Great “Statistical Tie” Fallacy (motherjones.com)
- More polls: Romney now within three points in Michigan, two in Pennsylvania (hotair.com)
- The Myth Of The Statistical Tie (jonathanturley.org)
- Big Bird Swallowed Mitt Romney’s Jewels (bakulaji.typepad.com)