It is really like the difference between Tour de France and the Formula One world championships. In Tour de France, the winner is the one who uses the least time overall to finish the course. Thus, winning by a huge margin is going to affect the overall result very much. In formula one, you get points according to how you finish. So the winner gets a certain amount of points, number two gets a certain amount and so on. Thus, I doesn't really matter if your winning margin in a race is 1 second og 10 minutes. So my general rule is: "Pearsons is like tour de France, Spearman is like formula one"
Yes, one aspect is whether the variables are normally distributed (if not, you might prefer Spearman). Another aspect is the scale: Pearson treats the data at interval scale, Spearman first transforms to ranks, so Spearman makes fewer assumptions, but loses some information. Also check whether outliers are valid data points or could maybe hint to errors in data collection. Generally, you should explore your data before applying statistical methods.
When i run the spearman correlation test I get a missing output. the " level" under the table does not appear and so I can't get the scatterplot any solutions please?
Good point of view. But, of course, you should remove those outliers before any calculation and thus pearson and spearman correlation coefficients shouldn't differ so much in real life.
Agreed. In most cases, they differ only marginally. Just like to add that outliers can be valid, meaningful values. It's good to be aware of their influence.
It is really like the difference between Tour de France and the Formula One world championships. In Tour de France, the winner is the one who uses the least time overall to finish the course. Thus, winning by a huge margin is going to affect the overall result very much. In formula one, you get points according to how you finish. So the winner gets a certain amount of points, number two gets a certain amount and so on. Thus, I doesn't really matter if your winning margin in a race is 1 second og 10 minutes. So my general rule is: "Pearsons is like tour de France, Spearman is like formula one"
I like that comparison very much. Spot on and easy to understand. Thank you!
@olavjonas What about the Kendall's Correlation? Can you put it in that interpretation? :D
This example really helped me to understand the differences between Spearman's and Pearson!
Thanks, glad to read that!
Thanks for giving such a nice example and description.
nice video, thank you! I have learnt smth. new:)
Please can you clarify how we choose between Pearson and Spearman? Do here conduct normality test?
Yes, one aspect is whether the variables are normally distributed (if not, you might prefer Spearman). Another aspect is the scale: Pearson treats the data at interval scale, Spearman first transforms to ranks, so Spearman makes fewer assumptions, but loses some information. Also check whether outliers are valid data points or could maybe hint to errors in data collection. Generally, you should explore your data before applying statistical methods.
When i run the spearman correlation test I get a missing output. the " level" under the table does not appear and so I can't get the scatterplot any solutions please?
+Sana La Morena Have you got valid cases in both variables?Is there variation? Maybe a perfect correlation of +1 or -1?
It would be better if the voice was loud and clear. Sound has a great impact on the vídeo.
Sorry and thanks for your feedback!
Welcome! ☺
Good point of view. But, of course, you should remove those outliers before any calculation and thus pearson and spearman correlation coefficients shouldn't differ so much in real life.
Agreed. In most cases, they differ only marginally. Just like to add that outliers can be valid, meaningful values. It's good to be aware of their influence.
Yes, of course. Regards.
Hello Moderator, you are doing a great job. Please I need a MATLAB code to determine KROCC/SROCC/PLCC and RMSE . Thank you.
Hi Felix,
thanks for your comment! Sorry I don't work with MATLAB.
How do you feel about Pearson vs Kendalls Tau?
+BoB Very similar, since Tau is also non-parametric, and outliers won't affect it the same way as Pearson correlation is affected.
How scatterplot were produced?
These are simple SPSS plots. I don't find them very appealing. Nowadays I would use R / ggplot2. Looks much better.
There is a German version as well (different screens, but same idea):
ua-cam.com/video/Y7-ZT1dGOxM/v-deo.html
Good explanation. TQ
excellent
U sound like James Charles