Correlation: Pearson vs. Spearman

How do correlation analyses work? What is the difference between the parametric Pearson correlation and the nonparametric Spearman's Rank correlation? In a practical example the two analyses yield opposite results: One is significantly positive, the other significantly negative. This surprising result is explained graphically and by looking at the data set, but without using any mathematical formulas. After two influential outliers are excluded, the two coefficients arrive at the same result.
Contact me, e. g. to discuss (online) R workshops / trainings / webinars:
LinkedIn: / wolfriepl
Twitter: / statistikindd
Xing: www.xing.com/profile/Wolf_Riepl
Facebook: / statistikdresden
statistik-dresden.de/kontakt
R Workshops: statistik-dresden.de/r-schulu...
Blog (German, translate option): statistik-dresden.de/statisti...
Playlist: Music chart history
• Music Chart History

[@Brickkzz]

Great example - I selected the right analysis for my study!

[@brendali5803]

Thanks for sharing. Very impressive example - the outliers really mess up Pearson's correlation, and Spearman's correlation is not sensitive with outliers.

[@njdalanon417]

Yes that is true, but just to add, as long as the assumptions for Pearson's correlation are met, it is still more powerful than Spearman's. :) Great video.

[@brendali5803]

@@njdalanon417 I see. Thanks for your sharing! :-)

[@BrandonSLockey]

Bang on, this comment should be pinned

[@julesdrums6167]

Thank you! Very helpful for my understanding.

[@kashif3173]

Great explanation

[@XpennyX07]

thanks so much very helpful!

[@missyuu9348]

I don't get it

[@StatistikinDD]

Use Pearson if the data are on an interval scale and at least approximately normally distributed. Use Spearman if the data are on an ordinal scale or if there are extreme values (influential outliers). Note that there may not be a simple answer in the sense of one being correct, the other one wrong. You can calculate both and compare results. If they are the same, it doesn't matter much which one you choose. If results differ, the video tells you why this could be the case.

[@taoyang563]

Fantastic

[@joepvanheesch2865]

Thanks a lot!

[@krauseyuan430]

THANks! BUT how to check the discrete missing value automatically in a statistical method by SPASS if there is a large amount of data ?

[@StatistikinDD]

That's an old video I re-uploaded ... Have moved on to R / RStudio some time ago.
In SPSS you can check descriptive statistics on the variables to find out extreme values, looking at maximums, or at how far away values are from mean +/- standard deviation.
A graphical method would be to use boxplots (which show outliers), or scatterplots. You can also create linear models and check model assumptions, e. g. about influential data points, which could be extreme values.
So, not one single best method, but several approaches ...
All the best

[@krauseyuan430]

@@StatistikinDD thans soo much for ur explanation .I'm a beginner and try to understand statistical analysis methods in experimental medical research paper .do u have any recommendations for online videos?

[@StatistikinDD]

@@krauseyuan430 Sorry, no ... Depends on exactly what you are looking for. I'd also have to use search terms ...

[@kavanpuranik98]

Can we do a correlation between two different types of variables? Like linear vs ordinal?

[@StatistikinDD]

You're safe assuming the lower level scale. So for linear vs. ordinal you can use Spearman correlation.

[@zkomins]

What computer software was used?

[@StatistikinDD]

This was SPSS. It's an old video that I re-uploaded recently. Currently I use R / RStudio.

[@billytbsiang]

I still do not understand. When I should use Pearson and when I should use Spearman?

[@StatistikinDD]

Pearson if the data are on an interval scale and at least approximately normally distributed. Spearman if the data are on an ordinal scale or if there are extreme values (influential outliers). Note that there may not be a simple answer in the sense of one being correct, the other one wrong. You can calculate both and compare results. If they are the same, it doesn't matter much which one you choose. If results differ, the video tells you why this could be the case.

[@MatroidX]

@@StatistikinDD Thanks for this great video and comment. Helped my intuition and understanding a great deal!

[@lovescript7436]

How if my data is interval but not normal distributed, should I use Spearman or Pearson?

[@jennirose1116]

He sounds a bit like max verstappen lol. Good video tho helped me understand.