How to evaluate quality of correlation?

[simpleton]

Hi,

I would like to know, how do I evaluate the quality of correlation?

Specifically, I have a set of N datapoints, each represented by k features. I want to know how the k features correlate with each other, and therefore, I created a k by k correlation matrix. I am using R-value, so the values range from -1 to 1. I was told to only look at R-values of at least 0.7, because anything lower than that does not mean much. However, I was wondering, how do I evaluate a correlation value of say, 0.85? If a feature x has a R-value of 0.7 with y and a R-value of 0.9 with z, what quantitative conclusions can I draw and what kind of statements can I make? For example, will I be able to say how much better the correlation between (x,z) is compared to (x,y)?

Thank you very much!

 

[chiro]

Hey simpleton.

The answer to your question is highly subjective.

You need to not only take into account the correlation value, but the nature of the data which include the graphical nature as well as the domain in which the sample is represented.

If there is a non-linear trend with what looks to be a seemingly low-residual, then using the linear correlation method will not really emphasize the true relationship between the variables.

The above points vary depending on what kind of data it is, where it came from, what its used for, and what questions its trying to answer.