# What does it mean to explain variance?

In particular, I am wondering why we have this concept Multiple R (which I can understand as the correlation between observed and predicted scores in multiple regression), and then a separate concept R-squared which is just the square or R.

I've been informed that R-squared is the percentage variation explained and R is not, but I don't understand the distinction that is being made between correlation and explained variation.

• One of my greatest teachers recounted his PhD thesis defence in which he explained to a group of physical oceanographers that his regression could explain XX% of the variation in some data. A distinguished examiner stood up and proclaimed "With your correlations you have explained exactly nothing; rather you have offered us a suggestion which falls apart 10% of the time" Apr 8 '14 at 15:39
• This is a duplicate of: stats.stackexchange.com/questions/90793/… Mar 31 '15 at 19:12

A main issue here is that the measure of "variation" in regression analysis is related to the squared differences of observed variables from their predicted mean values. This is a useful choice of a measure of variation, both for theoretical analysis and in practical work, because squared differences from the mean are related to the variance of a random variable, and the variance of the sum of two independent random variables is simply the sum of their individual variances.

$R^2$ in multiple regression represents the fraction of "variation" in the observed variable that is accounted for by the regression model when squared differences from predicted means are used as the measure of variation. The Multiple R is simply the square root of $R^2$.

I'm afraid that I've never understood the usefulness of specifying the value of the Multiple R rather than $R^2$. Unlike the correlation coefficient $r$ in a univariate regression, which shows both the direction and strength of the relation between 2 variables, specifying the Multiple R doesn't seem to add much beyond a chance for additional confusion.