Outliers and correlation

[lora]

Hi all

Im working on small size sample (17) . and need to do correlation for alot of prameters my questions is :

-is removing the outliers is the only thing I should do to prepare the my data for doing correlation ? What about the transformation ? when I should use it ?

- What is the best procedure to do correlation if my sample size is (17) and trimming the outliers may affect ?

Many Thanks

 

[JesperHP]

correlation is a measure of the strength of association between variables so technically you cannot do "correlation for alot of prameters" as you suggest. If you truly need help you probably have to rephrase the question. Is the problem that you have a lot of different variables in a sample of size 17?

How do you know they are outliers?

 

[hlsmith]

I believe the posters just wants to run a bunch of pairwise correlations for all of their variables. The correlation procedure will depend on the formatting and/or normaliity of your data.

 

[lora]

yes I have so many variables and need to test if any correlate with another variable ..

and Im confused with which is the best way to prepare the data for doing correlation when I have outliers !! that appear from the plots and skewness ..

when I did read in Andy feild book there is many ways like transformatiom , trimming , etc

so for example when I do trimming the data I still can see from histogram and skewness the data not normal distributed yet !! so shall I do transformation also ?!

so confused !!!
thats why Im asking about the best procedure to prepare the data for correlation if you have some outliers or not normal !!

 

[Lazar]

Do scatter plots. With a small n you will get a much better idea by visual inspection than simply running correlations alone.

 

[lora]

one more thing also I faced when using the Spss for removing the outliers ...
when I need to delete outlier shall I just replace it with (-9 ) as missing data ? or I should just delete the value ? if I delete the value I should delete all other variable raw related to this outlier right ?!?!?!

for example

x y
2 4
3 7 <---- outlier (so shall I remove the whole raw 3 and 7 ?? )


sorry my question might be silly for some of you ;(
1 2

 

[lora]

Do scatter plots. With a small n you will get a much better idea by visual inspection than simply running correlations alone.

thanks for the sugestion
I used to do this to figure the outliers but recently I found that I should follow the quarters Q3 and Q1 methods
see
http://www.youtube.com/watch?v=bRdC1u9veg8


what do u think ?!

 

[lora]

I believe the posters just wants to run a bunch of pairwise correlations for all of their variables. The correlation procedure will depend on the formatting and/or normaliity of your data.

is this mean only remove the outliers ? or needto check the shape of the distribution and the skewness ?

 

[Lazar]

thanks for the sugestion
I used to do this to figure the outliers but recently I found that I should follow the quarters Q3 and Q1 methods
see
http://www.youtube.com/watch?v=bRdC1u9veg8


what do u think ?!

Well I was not really thinking only about outliers. I just think with so few cases you will be better at detecting associations and yes outliers by using your eyes.