by Kosha Bramesfeld, PhD
I’m sure you’ve heard it before: “correlation is not causation”. But, what does that mean? A correlation describes the association between two or more variables. For example, you might learn that people with more knowledge about “X” are also the most productive at “Y”. In light of these findings, it can be tempting to think: X causes Y. Let’s implement X, so that we can increase Y at our organization.
Unfortunately, correlational studies only establish one of three conditions of causality: association. To establish that X causes Y, an evaluator must also address two additional conditions: establishing temporal order and ruling out alternative explanations.
· Establishing temporal order. To establish causality, you must be able to identify which variable is the “cause” and which variable is the “effect”. Returning to the earlier example, you would need to establish that increases in productivity (“Y”) occurred only after knowledge about “X” was acquired. If increases in productivity existed before the knowledge was acquired, or if the two variables co-occurred haphazardly, without a clear pattern of X coming before Y, then one cannot establish that X causes Y. The problem with correlational studies is that all of the variables are typically observed or measured at the same time, making it difficult to establish which variable is the “cause” and which variable is the “effect”.
· Ruling out alternative explanations. To establish causality, you must also be able to rule out alternative explanations. Just because X and Y co-occur does not mean that the association is a causal one. The co-occurrence could be coincidental (known as a spurious correlation) or the occurrence of X may be confounded with other factors. For example, if your organization institutes a new training program at the same time that the organization changes its accountability standards, it may be difficult to know if any changes that arise in the outcome are due to the training program or to the changes in accountability. The best way to rule out alternative explanations is to isolate and change only a small number of a controlled factors at a time and to include a “no-intervention” control group that is exposed to all of the same external factors and conditions as the “intervention” group, but is not exposed to the key intervention of interest (in this case the training). Correlational studies, in which all of the variables are observed or measured, and none of the variables include a control condition, cannot be used to rule out alternative explanations.
In summary, “correlation is not causation” because correlational studies establish only one of three conditions of causality: association. The best methods for establishing cause-and-effect not only establish association, they also (a) measure outcomes over time in order to establish temporal order and (b) experimentally create treatment and control groups, so as to more clearly rule out alternative explanations.
Fletcher, J. (2014). Spurious Correlations: Margarine linked to divorce? BBC News. Retrieved from: http://www.bbc.com/news/magazine-27537142
Comic from https://goo.gl/i4hZzj
Find out more about Kosha on LinkedIn https://goo.gl/CNpZxP All copyrights are not ours but with the owners of the photos. All guest post comments and opinions are their own and not reflective of Robin Yap.