I recently wrote a blog post that discussed a similar point of view. Sometimes we are more interested monotonic relationships in general rather than linear relationships in particular.
Great explanation. To put it another way, spearman rank correlation is great for hypothesis testing only. I.e. if you just want to know the strength of correlation. However, it won't tell you what bowling score someone will have if you know their basketball score i.e. estimates are on the ranks, not the actual values.
That which you say is what Linear regression is. They are closely related. The aim of Correlation usually is to assign a numerical value between -1 and 1 showing linearity. Regression is to predict what bowling score someone will have if you know their basketball score, while correlation is to verify how closely related they are. Think of it like a Corr. is used to check if we can use Ln. Reg. on this data.
Nice videos. Please dont ever compromise the mathematical rigour, even for 50k extra views. Can you please consider a video on how you handle multicollinearity, also what you do when you encounter multiple inter-correlated features.
@@ritvikmath If you want a deeper dive, look into Kendall's general correlation coefficient. It generalizes some of these familiar correlation coefficients. Not sure if it is practical enough to mention for your channel, but you might enjoy seeing the pieces come together.
Simple thought: it can be nice to see spearman correlation if let's say the ranking on the bowling was out of complete match with the basketball, i.e. 13245....