Why correlation is not causation??

Added: 2026-05-28

[00:00:01]Did you know that the amount of mozzarella cheese people eat tracks almost perfectly with the number of civil engineering doctorates being awarded?

[00:00:10]The statistical correlation is nearly 96%.

[00:00:14]So, if we have to have safer bridges or or even better infrastructure, do we just need to force civil engineering students to have more pizza?

[00:00:26]Of course not.

[00:00:27]But your brain looks at this graph and immediately tries to connect the dots.

[00:00:32]Welcome to Get That Stat. I'm Dr. Anjushree Krishnan, and today we are breaking down the most dangerous traps in data science, why correlation is not causation.

[00:00:46]Let us first define our terms.

[00:00:49]Correlation means two variables moving together.

[00:00:52]If variable A goes up, accordingly variable B goes up or down. They are just like dancing partners.

[00:00:59]Whereas causation means if variable A directly forces variable B to happen.

[00:01:06]To see how easy it is to confuse between the two, let us have a look at the ocean.

[00:01:10]In the summers, you know what skyrockets?

[00:01:14]It's the ice cream sales and at the same time shark attacks.

[00:01:18]So, does eating ice cream make you more delicious to the sharks?

[00:01:23]No.

[00:01:24]It's in fact the third factor or sometimes the hidden factor which pulls the strings.

[00:01:29]It's the summer heat.

[00:01:31]The heat causes more and more people to buy more ice creams.

[00:01:34]And the same heat also urges people to go and swim in the ocean.

[00:01:40]So, ice cream sales and shark attacks are correlated, but neither causes the other.

[00:01:46]Now, you might think, "Anju, this is just a funny textbook example."

[00:01:50]But in the field of data science, artificial intelligence, and machine learning, this mistake can be catastrophic.

[00:01:58]See, AI algorithms are essentially massive correlation engines.

[00:02:04]They look at billions of data points and find matching patterns.

[00:02:09]But algorithms are not naturally trained to understand the data.

[00:02:15]Imagine there is an AI model trained to predict patient health in a hospital.

[00:02:20]The data shows a massive correlation between patients wearing hospital gowns and patients being severely ill.

[00:02:28]If the AI model takes this correlation for causation, its brilliant medical advice is going to be to cure the patients, just take away their hospital gowns.

[00:02:39]This is exactly why we need explainable AI.

[00:02:42]We can't just trust the model because the numbers match. We need to interrogate the why behind the numbers. If you want to move past the textbook formula and truly understand the data driving your world, hit that and subscribe button. Let's learn together and let's get that stat.