Large differences in Pearson vs Spearman

#1
Dear all,

My name is Mark and I am currently writing my thesis for a Master in Economics. I examine whether year to year changes in certain industry variables have an effect on the likelihood of CFO termination (0/1 variable). I have six X-variables, that are calculated as the absolute pro rata change between an industry year average (of a given variable) and the previous year average (t1-t0/t0). So, the x variable is continuous and the same for all firms in the given industry in the given year.

I made a Spearman and Pearson correlation matrix and what I found was that some of these variables have very large differences (between Spearman and Pearson). The big differences are all connected to one variable named advertising. The correlations between this variable and four other variables all have big differences in Spearman vs Pearson. [ see attachment]

Also, I winsorized all my data at the 99th and 1st percentile, so I already took care of outliers.

Could somebody provide some explanation on why these correlations defer (Pearson vs Spearman). Also, does this indicate that there is something wrong with my data? Do I have to modify the data or can I just conclude things?

So, what is the normal set of actions to take when you run in to this situation?

Thank you in advance,

Kind regards,

Mark
 

gianmarco

TS Contributor
#2
Hello!
Since there is a difference between the two coefficients, maybe that the greatest discrepancies could be due to the fact that some relationship are monotonic but not linear, or viceversa.
I understand that you know the difference between the two coefficients. The Wikipedia page about Spearman rho is a good quick and dirt point to start from.

Have you plotted your data, in order to inspect if some of the variables are not linearly correlated?

Cheers
Gm
 
#3
Hello!

Thank you so much for your help. I believe I already found what I did wrong. I found a sort of simular article, and they explained that they calculated the correlations per year and then took the average. I tried it, and it resolves the problem. I guess my data just did not allow me to do it at once.

Again,

Thank you,

Mark
 

Karabiner

TS Contributor
#4
Unfortunately I only can see one matrix, therefore I don't know
what exactely you mean by saying that the differences are large,
but anyway you could perform some x-y-scatteplots in order to
check whether relationships are perhaps markedly non-linear
(as Gianmarco wrote).

With kind regards

K.
 
#5
The attachment presented the differences (Pearson - Spearman). But as I replied to Gianmarco, I believe that I have found the solution.
Thank you for your help, it is always good to know the answer to these sort of questions. I will still look at it.

Again,

Thank you for your time and help!