DATA SCIENCE

The Most "Ridiculous" Concept That A Data Scientist Must Know

Explaining the P-value in the most "ridiculous" way

Naser Tamimi
4 min readJan 5, 2022

--

Photo by Pietro De Grandi on Unsplash

P-value is not an easy and intuitive concept at first (especially if you read its definition on Statistics textbooks or Wiki pages). But, here, I try to explain it simply using a powerful word called "Ridiculous." You must ask why I use a word like ridiculous several times in the last few sentences (I apologize, but it helps). The reason for that is I believe "ridiculous" is a keyword to learn and remember the P-value concept. How? Let me explain it to you.

Disclaimer: Many statisticians and data scientists before me tried to explain the P-value using “ridiculous”. I am not the only person (even the first person) who came up with this explanation of the P-value. This particular word helped me and many other data scientists (and of course statisticians) to have a better understanding of P-value. It may help you too.

How Ridiculous?!!

How many times did you make an assumption and you found out that you were wrong? I assume pretty often. In those situations, you made some hypotheses after observing some pieces of evidence and finally judged or acted based on that. When you found out that you were wrong, you might have reviewed your thinking process and told yourself: "How ridiculous I was when based on the evidence, I made believed my hypothesis!!!".

This is exactly what the P-value is measuring. P-value measures the ridiculousness of your null hypothesis. What is the "Null Hypothesis"? Read the next section.

Null Hypothesis?!!

You collect data (for example, after adding a new feature to your app), and voila! There is a significant difference between your data before adding the feature (we call it population data) and after data (let's call it a sample or test data). It is a super exciting moment! You knew this feature would increase user engagement. But wait !!!

You must always try to control your excitement when testing a hypothesis. Suppose you add a new feature to your app that seems to enhance the user experience (for example, it increases user engagement). In that case, you must control…

--

--