Don’t get me wrong, correlations are great. They highlight relationships between two variables and, when used correctly, can be used to support future planning to encourage change. Emphasis on the ‘when used correctly’ part.
There are two possible correlations that two variables can display;
There can also be no correlation between two variables, meaning there is no relationship between them.
Depending on your hypothesis, finding a relationship between two variables can leave you with a warm fuzzy feeling. However, it should also leave you with the instinctive curiosity to find out if this association is just that – two variables sharing a similar trend – or if it is in fact indicative of one variable having an effect on the other. The latter is not something which a correlation alone can convey.
It is essential to understand the difference between two variables sharing a trend and two variables that have an actual impact on one another, and is important that you know which one applies to your correlation. Knowing that you have two variables that do indeed impact one another can help to inform future behaviour change activities, as you know changes to one will have a knock-on effect onto the other. However, if these variables simply share a trend and do not have an actual impact each other, then using this correlation to inform behaviour change activities would be futile. Because of this, the biggest mistake that can be made once a correlation is found is to stop your analysis, proceed with the correlation and use this to make key decisions on future planning and intervention work.
This not only runs the risk of focusing on something which doesn’t actually have an impact, but can also lead to more harm than good. What if a variable actually encourages a different change that directly conflicts with the change you want to see? What if there’s a third variable at play that is actually causing this relationship to occur? Or, what if these two variables are simply related by chance? If so, you would have invested who knows how much resource into an intervention or policy change that actually isn’t going to have much of an impact in bringing the change you want to see at all, and may possibly even make things worse. Not good.
It is not unknown for two variables which don’t actually impact or affect each other to have a correlation. Tyler Vigen’s Spurious Correlations website (and book!) showcases this perfectly with a number of weird correlations which just make no sense. These are US based, and examples include a correlation between the divorce rate in Maine and per capita consumption of margarine, and a correlation between the per capita consumption of cheese and the number of people dying by becoming entangled in their bedsheets. If these were taken at face value, we’d be looking at margarine being banned to save marriages in Maine and cheese being banned to save people from a deadly tangle with their bedsheets – has my point been made yet?
If you’ve gotten as far as conducting a correlation or two with your dataset, then it’s not much further to go to implement and embed further statistical analyses. These further tests will help you understand whether you can accept or reject your primary hypothesis (i.e. your prediction for the dataset). Namely, they will tell you whether one variable does indeed have an impact on the other, and will more appropriately equip you for any future planning decisions than a correlation alone would.
So, what should you take away from this? Not to be repetitive, but it comes back to the classic ‘correlation is not causation’ message which every statistician should have in their heads when using this statistical measure. However, it also highlights the importance of constantly probing your data and findings for more.
You’ve found a relationship between two variables? That’s great. Now take it one step further and consider: what does that mean?
Oh, you’ve found that these two variables do in fact have an effect on one another? Fantastic! What does that mean?
It’s not enough to take data and findings at face value, you need to go deeper, understand what the insight is, what it means for your audience and how it can link into encouraging positive change. Our in-house research team prides themselves on their ability to do this, and have consistently been able to look beyond the surface to find insights and turn these into action, whether through campaigns or actionable recommendations for change. If you want to find out more about our insight finding capabilities, or if you’ve found another weird correlation that you want to share, then get in touch with our research team today.