[R] correlating values

[Sundar Dorai-Raj]

Ahamarshan jn wrote:
>   I did a correlation for the values 
> 
> -   EX4577  EX4599      EX4566	          EX4522
> WL917	2.53528 0.79077 0.21499  -0.01084
> WP429S	-0.192723715
> WP819	-1.016997552
> WP977	1.378674	-0.07071	0.625089	0.4728363
> WI205S	-0.24443	-1.789526	0.648923	-0.775867
> 
> 
> by using
> round(cor(t(person.data),use="pairwise.complete.obs"))
> 
> i got the result as
> 
> - WL917	WP429S	WP819	WP977	WI205S
> WL917	1	NA	NA	0.344	-0.11424
> WP429S	NA	NA	NA	NA       NA
> WP819	NA	NA	NA	NA       NA
> WP977	0.34461	NA	NA	1       .23294
> WI205S	-0.11424NA	NA	0.23294	1
> 
> i notice that 
> for correlation between WP429S x wp429S the value is
> given as NA where as it should be 1
> same is the case with WP819 x WP819 
> 
> Can someone please help me reason out why this is
> happening? and any corrective measure that I need to
> take so that I can get the true value of 1 for them.
> 
> Thank you
> - ash#
> 

Hi, Ash,

It's difficult to tell with the data you provided (maybe just poor email 
formatting), but it looks like both WP429S and WP819 have only one 
value. You cannot obtain a correlation with only one value (see any 
intro stats book for the definition or correlation or just Google it). Try:

x <- rnorm(1)
y <- rnorm(1)
cor(x, y)
cor(matrix(x, 1, 1))

This is because x and y have zero variance. Forcing the "NA" to "1" as a 
"corrective measure" would be incorrect and misleading.

HTH,

--sundar