ARRRRG MATEY! According to this graph, HR can clearly infer that we can save the planet and reduce global warming by focusing on our pro-pirate workforce development plans. I’m from Tampa where we have a natural affinity for all things pirate related, so this example of correlation vs. causation cracked me up!
Yes, these are basic statistical concepts, but we need to make sure we don’t make inadvertent leaps in connecting data measures.
Correlation: A relationship simply exists between 2 or more variables. The fewer the pirates the worse global warming gets. They may be correlated but they aren’t necessarily causal. As Steve Boese says: “A strong correlation between two related and relevant data series may not imply or prove causation, but it probably implies something.”
Causation: Cause and effect. A relationship exists between two variables or events, where the second is a consequence of the first.
We must be evidence-based HR pirates. It’s important that we aren’t overly simplistic or inadvertently misidentify correlation as causation when using HR data to make business decisions. The real value comes from credible, clear and actionable HR measurements based on causal relationships.
Think that as HR you have decided to give people 10% more days off in a year and saw that employee engagement increased meaningfully in the following year. You really want to believe that your 10% favor caused the increase in engagement, right? This might or might not be the case. To be able to figure out if there is a causal relationship or not you can do two things: 1. Have a control group in you sample where no increase is given (but make sure that both samples represent your population) and compare the engagement of your groups. 2. Control other variables (by keeping them constant or controlling them through statistical analyses) which might be causing employee engagement. Did you give any other incentives? Was there a major change in the environment? Was there a major change in the lives of your employees outside the work life (e.g. a governmental project supporting working parents’ babysitting needs, a subway station built close by to your company decreasing the commuting time)?
There may be times when correlation is sufficient; when it’s useful just to know that a relationship exists. David Bernstein’s article “Big Data for HR – Don’t Forget, The Basics Still Apply” describes how we can capitalize on “simply knowing there is a correlation between data points:
When the intended outcome is either: 1) Preventive – i.e. reduce attrition by finding patterns in your voluntary termination population, or 2) Increases the likelihood of a positive event – i.e. increasing the completion rate of online applications to your job postings.
Proving causation for a particular issue may be the endgame (i.e., what employees will benefit the most from training), identifying correlations as a starting point is certainly beneficial and may be enlightening. As long as we don’t jump to conclusions.