The Data Analyst in You: Using Context for Causation

Should lifeguards be concerned by a jump in ice cream sales?

The first time I remember questioning causation vs. correlation was playing hide-and-seek on the school playground (I was born to analyze).  Straight out of The Little Rascals, my classmate’s hair sprung out from behind a ledge, ruining an otherwise perfect hiding spot. He argued, “you could not have possibly seen me because I could not see you.”

This argument is silly due to experience.  This is where you, the non-statistician, are a data analyst.  Although in 99.99% of the situations my classmate would be correct, we have proven that this is merely correlation, not causation, through the simplest of experiments (such as standing directly behind someone).

Unfortunately, not all questions of correlation and causation are this easy to settle.  There are often strong relationships between two variables, but it can be difficult to identify the cause and the effect.  In academia, there is a lot of work and due diligence (meaning time and money) to prove scientific causation. In business, you need an answer yesterday. Expertise and experience are often your only tools. 

There is value in these factors and relationships even if they are not exact cause-and-effect matches. That value comes from using your own experience to identify the variable’s underlying meaning. When presented with a relationship between two variables (calculated by your statistician), ask yourself three questions:  

1. Does it make sense for X to cause Y? 
2. What could X really mean?  
3. Do we have a variable that might better represent that meaning?    

By answering these questions, you will have completed a high level data analysis with no math or arithmetic required!

The classic example is the strong correlation throughout the calendar year between an increase in ice cream sales and an increase in shark attacks.  So if you were a lifeguard, would you be concerned by additional ice cream sales?  

1. Does it make sense for ice cream sales to cause shark attacks?  Sharks might like ice cream if given the chance.  I emphasize that this is unlikely.
2. What could ice cream sales really mean? It is more likely that ice cream sales are correlated with being hot outside. The hotter it is, more people will swim at the beach and the water will be warmer for sharks.
3. Do we have a variable that might better represent being hotter outside? We could try using outside temperature or water temperature and see if that predicts the number of shark attacks.

I recommend checking out these other fun examples.  Are they completely coincidence? Or is there something that ties them together?

For more information on how everyone has a role in data analyses, check out previous articles of mine.