Comments on: Correlation Coefficient in Excel http://articles.excelyogi.com/correlation-coefficient-in-excel/2008/11/06/ Analyst Resources Mon, 21 Nov 2011 21:44:07 +0000 hourly 1 http://wordpress.org/?v=3.3 By: Joseph Harris http://articles.excelyogi.com/correlation-coefficient-in-excel/2008/11/06/comment-page-1/#comment-18 Joseph Harris Fri, 07 Nov 2008 17:22:33 +0000 http://articles.excelyogi.com/?p=80#comment-18 John... Thanks for the input. I do not fully agree with your assessment. You cannot really paint correlation with a broad brush whether you are using the correlation coefficient or the coefficient of determination. You cannot say that an R² of greater than any specific number is a large amount of correlation; it depends on the context of what you are looking at. For example, if you are saying over 12 years (12 data points) you have an R² of 50% that may or may not mean much, but if you have 4,380 data points (# of days in 12 years) and a R² of 50%, that is most likely a very significant amount of correlation. You are right that most statisticians would use R², but for real world views, and simple correlation (not causation), R is perfectly fine to use. Joe John…
Thanks for the input. I do not fully agree with your assessment. You cannot really paint correlation with a broad brush whether you are using the correlation coefficient or the coefficient of determination.

You cannot say that an R² of greater than any specific number is a large amount of correlation; it depends on the context of what you are looking at. For example, if you are saying over 12 years (12 data points) you have an R² of 50% that may or may not mean much, but if you have 4,380 data points (# of days in 12 years) and a R² of 50%, that is most likely a very significant amount of correlation.

You are right that most statisticians would use R², but for real world views, and simple correlation (not causation), R is perfectly fine to use.

Joe

]]> By: Jon Peltier http://articles.excelyogi.com/correlation-coefficient-in-excel/2008/11/06/comment-page-1/#comment-17 Jon Peltier Fri, 07 Nov 2008 00:42:45 +0000 http://articles.excelyogi.com/?p=80#comment-17 I know those values for small, medium, and large correlations were what Wikipedia presented, but a correlation of ± 0.5 ("medium/high" in the table above) means only 25% of the variation is due to covariance between the variables. 25% is a rather small correlation, actually. In <a href="http://www.mega.nu/ampp/rummel/uc.htm" rel="nofollow">Understanding Correlation</a>, Prof. R.J. Rummel says "[S]quaring should be a healthy corrective to the tendency to consider low correlations, such as .20 and .30, as indicating a meaningful or practical covariation." You should evaluate R², not R, and you should not imagine that your correlation is large unless R² is greater than perhaps 0.75. I know those values for small, medium, and large correlations were what Wikipedia presented, but a correlation of ± 0.5 (“medium/high” in the table above) means only 25% of the variation is due to covariance between the variables. 25% is a rather small correlation, actually.

In Understanding
Correlation, Prof. R.J. Rummel says “[S]quaring should be a healthy corrective to the tendency to consider low correlations, such as .20 and .30, as indicating a meaningful or practical covariation.”

You should evaluate R², not R, and you should not imagine that your correlation is large unless R² is greater than perhaps 0.75.

]]>