If you've been in this game for some time, as I have, you may remember the thrill that accompanied econometric models in their early days. These were going to tell us the secret of how advertising worked and what made people buy our brands. The econometric models are very useful, of course, to help marketers interpret the past, but alone they cannot predict the future. And a badly conceived or interpreted model is worse that none at all.
Every time such a model is presented, I find it useful for the agency to re-iterate what correlation actually is, and what it isn't. At college, I had "correlation does not imply causation" drummed into me I don't know how many times. It's like this, if A and B are correlated:
A could cause B
B could cause A
A and B are the consequences of a common cause but one does not cause the other. So, sleeping in your shoes does not cause headaches, even though instances of the two events are correlated. One look at our friend above can tell you the underlying cause.
There is no connection between A and B - the correlation is coincidental.
This is just one of the factors that Tim Harford considers in an excellent FT Magazine article about Big Data and the potential pitfalls in its interpretation. We must beware that the errors we make with more manageable data sets are not simply compounded as the data set gets bigger - for example, via sample bias. And, in the end, the human factor in terms of insightful interpretation remains key. His closing sentences sum this up very nicely:
“Big data” has arrived, but big insights have not. The challenge now is to solve new problems and gain new answers – without making the same old statistical mistakes on a grander scale than ever.
Posts
No comments:
Post a Comment