Correlation does not imply causation
There is an apocryphal tale about an inner city education department in the American Mid West.
In their efforts to increase children’s performance they commissioned an extensive study of the socioeconomic environment their students were raised in, to see what impact it had on a child’s educational prospects. The study delivered one simple, strong and statistically valid insight.
Households that contained lots of books had children who did well at school.
This finding gave the board of education cause to think and led to much deliberation.
An audacious plan
They decided to provide a library for every child. For years they sent free books to every student registered within their geographic boundary. Over the 10 or so years that every child attended school they were given their own personal library.
The plan, audacious and expensive as it was, did nothing for the standard of education within the district.
Correlation does not equal causation
The reason children who lived in houses full of books did well at school wasn’t because they had lots of books in the house. It was because they had parents who bought lots of books. It was all to do with the parent and nothing to do with the books.
This is good news for those of us who own a kindle.
Spurious correlations are everywhere
It is easy to jump to conclusions
Pirates protected the world from global warming
Murderers prefer Google Chrome, it keeps them off the streets surfing
Cheese will kill you in your sleep
Correlation and causation are not the same thing
The point(s) of the post is simple:
- To improve the system find the cause of the problem, not simply something that is correlated with it.
- Buying books won’t make you clever — believe me, I’ve tried.
If you enjoyed this post click here to have the next delivered straight to your inbox.
Read another opinion
Image by josefnovak33
James,
This is a point that cannot be repeated often enough. Focus on getting at the root cause. While correlation exposes association (when we look at two variables that actually belong together – unlike cheese and sleep deaths), it doesn’t get to the root of the problem.
Annette :-)
I’ve just read the entire article. I’m fine with it right until the last line: ” It was all to do with the parent and nothing to do with the books.” <— I think this is a stretch; Surely it had *something* to do with the books!?
==================
b) In my high school and in post secondary education, the students who did the best spent their fair share of time with book study. BUT… they also did things like physical exercise (Intramural sports, gym work-outs etc.) they socialized (within limits) they paid attention to their diets and tried to eat good, nutritious foods – as opposed to steady diet of fast foods. // When it came time to pick up the books, these students were ready to learn and absorb the pertinent info.
I suppose Dan that is a little like the argument that guns don’t kill people. Fortunately I have just enough sense not to start that one.
Thanks for your comment
I have found that there are four cause and effect types.
1 – simple – yes we can map and change the effect when we alter the cause
2 – unknown – there is a cause and effect, changing the case changes the effect but we do not know why
3 – complex there is no relationship between them but when one changes the other changes and the reasons will vary too much and change on a regular basis
4 – chaos – changing one causes an effect on another system
Cheers
The number of people bitten by a jellyfish correlates to the number of ice creams eaten by people. In other words: the more people eat ice creams, the more people are bitten by jellyfishes!
Maybe it is the other way round Denis. Maybe the ice cream is to sooth the jelly fish wounds
That is why we always have to be careful with statistics. – Just like any other tool – statistics can be used for important purposes explaining the facts – but it also can be used for manipulation and cheating. “The background” has to be explained, the “data” as well and the way data is collected and treated. Otherwise any statistics may lead to “wrong conclusions” – (…because statistics is part of mathematics). – Nice subject!
I liked the “pirates vs. global warming” correlation. As we’re commemorating the start of WW1, you might try this one, from 1914: how helmets caused head injuries.
Alone among the major combatants, the British army in 1914 did not issue its troops with steel helmets (they would not stop a bullet so they were only extra weight). When the shooting started, many soldiers suffered head wounds from (low-velocity) shell fragments rather than (high-velocity) bullets, so the Army had a rethink.
The casualty returns kept coming in and to the amazement of all concerned the percentage of men with head wounds went up. In fact they were highest in the units that had 100% issue of tin hats. So, clearly, the helmets were making matters worse.
It took a bit longer to realise that sure enough, there were more men coming in with head wounds. . . but there were fewer lying in no-man’s land with their heads split open. Which shows the peril of looking at one set of numbers in isolation.
Lovely story Duncan, thank you
James,
I think they went wrong when they “commissioned an extensive study of the socioeconomic environment”. Extensive often tends to mean the over-application of some dodgy maths and stats by a whole bunch of people locked away in a bunch of rooms who have little experience or understanding of the real issue. There’s a lot to be said for ‘fieldwork’.
Adrian
I think you may have a point Adrian. Thanks for the comment
Great article James and insightful comments from everyone. In my consulting experience I have observed that, although correlation and causation may not go hand in hand between the two defined processes……however correlation itself it will in most cases give furthur insightful process related information
A reverse root cause analysis or reverse CT Tree of the independent process in many cases will reveal the actual process having sharing both correlation and causation with the dependent process. .
For instance in the school case study a root cause analysis or reverse CT Tree would have revealed the actual cause ( attitude of parents ). This could then be further corroborated with a small design of experiment
in.linkedin.com/in/senguptaraja/
A very good point Raj, amazing what the question “why?” will uncover.
Thanks very much for for the comment.
James