Top 5 Reasons Correlation Does Not Imply Causation

This video is part of the Correlation Does Not Imply Causation, So What Does series, presented by Jim Colton, GraphPad Lead Statistical Consultant.

[abramgaller]

It would be awesome were there to be an internet seminar on the basic statistical fallacies .

[worldaround]

Confounding factors

Someone wants to determine whether going to college makes you rich. We look at a very poor nation where the majority of the people are poor and don’t have a college degree. However, in this same country, you find that there are a few people who are very rich, and the majority of them have a college degree. So doesn’t it mean that going to college is what made them rich? The answer is that it is not that simple because people belonging to rich families are more likely to get a college degree, and those from rich families are very likely to inherit that wealth and become rich. In our example, the confounding factor is belonging to a rich family.

This means that there is definitely a correlation between being rich and going to college. In other words, the majority of rich people do have a college degree, while the majority of poor people do not. However, to say that going to college is the cause of being rich is not straightforward because of the presence of confounding factor. This means that correlation does not guarantee causation; that is why it is said in English that correlation does not imply causation. Correlation doesn't imply causation because of confounding factors, coincidence or small sample size and other things.

The word "confound" originally descends from a Latin word that means confusion, in fact the two words - "confound" and "confusion" are linguistically related. Let us explore a little bit of linguistics.

You will often find words in Hindi and Sanskrit that look very similar and roughly mean the same thing, but the way the words end can help you decide whether a word is from Hindi or Sanskrit.

Similarly, many words look very similar and roughly mean the same thing due to them being linguistically related, and based on their endings, you can determine whether a word is English or Latin. Take words such as "confusion" and "confound"—both have almost identical beginnings; only the endings are different. In fact, "confound" and "confuse" mean the same thing. The presence of confounding factor can lead to correlation and this may confuse you into believing that causation exists. To eliminate a confounding factors, we have to make the study design more complex to account for confounding factors.

Another example, Let’s say you want to find out whether diabetes causes heart attacks, and there is a genetic defect that causes heart attacks but does not cause diabetes and is not related to it. Is that genetic defect a confounding factor? The answer is no, because a confounding factor is something that can confuse you by showing correlation even when causation does not exist. The genetic defect can’t increase the number of people who have both diabetes and heart attacks, simply because although the genetic defect can cause heart attacks, it is not related to diabetes. The formal definition of a confounding factor is that it is related to both the exposure variable and the outcome variable.