The Joy Of Stats: Meaningless and meaningful correlations
Just because two things might appear to be related, it doesn't mean...
Just because two things might appear to be related, it doesn't mean they are - but it's the meaningful correlations that are the treasures of statistical analysis.
- Duration: 5 mins
- Published on: Tuesday 7th December 2010
- Introductory Level
- Posted under: Statistics
WatchYou need the Flash Player (version 7 or higher) to view this clip - download Flash. http://podcast.open.ac.uk/open2media/joyofstats/joy7.flv Copyright The Open University
ListenSave this MP3 file to your computer You need the Flash Player (version 7 or higher) to use our MP3 player - download Flash. Copyright The Open University Download What can I do with this?
Hans Rosling: I’ve got a joke about silly correlations. There was this American who was afraid of a heart attack and he found out that the Japanese ate very little fat and almost didn’t drink wine but they had much less heart attacks than the Americans. But on the other hand he also found out that the French eat as much fat as the Americans and they drink much more wine but they also have less heart attacks. So he concluded that what kills you is speaking English.
Michael Marmot: The standard criticism of correlations is that correlation is not causation. It’s often used as a cheap shot; oh that’s just a correlation, they’re just statistics, you haven’t proved anything.
Advertising jingle: Smoke, smoke, smoke on cigarettes. Puff, puff, puff and if you smoke yourself to death
The time, the pace, the cigarette, Weights Tipped.
Hans Rosling: The best example of a really groundbreaking correlation is the link that was established in the 1950s between smoking and lung cancer. Not long after the Second World War, a British doctor, Richard Doll, investigated lung cancer patients in 20 London hospitals, and he became certain that the only thing they had in common was smoking. So certain that he stopped smoking himself, but other people weren’t so sure.
Michael Marmot: A lot of the discussion of the early data linking smoking to lung cancer said well it’s not the smoking surely, that thing that we’ve done all our lives, that can’t be bad for you. Maybe it’s genes. Maybe people who are genetically predisposed to get lung cancer are also genetically predisposed to smoke. Maybe it’s not the smoking; maybe it’s air pollution, that smokers are somehow more exposed to air pollution than non-smokers. Maybe it’s not smoking; maybe it’s poverty. So now we’ve got three alternate explanations, apart from chance.
Hans Rosling: To verify his correlation did imply cause and effect. Richard Doll created the biggest statistical study of smoking yet. He began tracking the lives of 40,000 British doctors, some of whom smoked and some of whom didn’t, and gathered enough data to correlate the amount the doctors smoked with their likelihood of getting cancer. Eventually, he not only showed a correlation between smoking and lung cancer, but also a correlation between stopping smoking and reducing the risk. This was science at its best.
Michael Marmot: What correlations do not replace is human thought. You could think about what it means. I mean what a good scientist does if he comes up with a correlation is try as hard as she or he possibly can to disprove it, to break it down, to get rid of it, to try and refute it, and if it withstands all those efforts at demolishing it and it’s still standing up then cautiously say we really might have something here.
Advertising jingle: Hell St Peter at the Golden Gate, that you hates to make him wait, you’ve just got to have another cigarette
Hans Rosling: To find out more about the Joy of Stats visit the Open University’s OpenLearn website.
More Joy Of Stats
Have you got a passion for statistics?
Copyright & revisions
Originally published: Tuesday, 7th December 2010
Last updated on: Tuesday, 7th December 2010
- Body text - Copyrighted: The Open University
- Video - Copyright: The Open University
- Audio - Copyright: The Open University
If you enjoyed this, why not follow a feed to find out when we have new things like it? Choose an RSS feed from the list below. (Don't know what to do with RSS feeds?)
Remember, you can also make your own, personal feed by combining tags from around OpenLearn.