Relationship between R squared and Pearson correlation coefficient
THE DERIVATION IN THIS POST IS INCORRECT, PLEASE SEE AN UPDATED ONE INSTEAD.
Coefficient of determination (aka. $R^2$)
Consider the ordinary least square (OLS) model:
\[\begin{equation} y = \mathbf{X} \beta + \epsilon \label{eq:OLS} \end{equation}\]After fitting the model to the data ($\mathbf{X}$, $y$), let
\[\hat y = \mathbf{X} \beta\]We would like to understand the relationship between the variance of $y$ and that of $\hat y$.
\[\begin{align*} \mathbb{V}[y] &= \mathbb{V}[\hat y + \epsilon] \\ & = \mathbb{V}[\hat y] + \mathbb{V}[\epsilon] + 2\text{Cov}(\hat y, \epsilon) \\ &= \mathbb{V}[\hat y] + \mathbb{V}[\epsilon] \end{align*}\]See proof for the second equality in the Supplemental. The third equation is because $\hat y$ and $\epsilon$ are independent random variables, thus their covariance is 0.
Define
\[\begin{align*} R^2 &= \frac{\mathbb{V}[\hat y]}{\mathbb{V}[y]} = \frac{\sum_i^n (\hat y_i - \bar{\hat y})^2}{\sum_i^n (y_i - \bar y)^2} \\ \end{align*}\]Therefore, $R^2$ measures the ratio of $\mathbb{V}[\hat y]$ over $\mathbb{V}[y]$, and it’s commonly interpreted as the the amount of variance in $y$ that can be explained by the OLS model.
Pearson correlation coefficient
\begin{equation} \rho = \frac{\text{Cov}(y, \hat y)}{\sigma_y \sigma_{\hat y}} \label{eq:pearson_corr} \end{equation}
See definition on Wikipedia.
Relationship between $\rho$ and $R^2$
Now we’ve defined both coefficient of determination and Pearson correlation coefficient, let’s see their relationship.
Note
\[\begin{align*} \text{Cov}(y, \hat y) &= \text{Cov}((\hat y + \epsilon), \hat y) \\ &= \text{Cov}(\hat y, \hat y) + \text{Cov}(\epsilon, \hat y) \\ &= \mathbb{V}(\hat y) \end{align*}\]See proof for the second equality in the Supplemental. So
\[\begin{align*} \rho^2 &= \frac{\mathbb{V}[\hat y]^2}{\mathbb{V}[y] \mathbb{V}[\hat y]} = \frac{\mathbb{V}[\hat y]}{\mathbb{V}[y]} = R^2 \end{align*}\]Therefore, $\rho^2 = R^2$, neat!
Here is a notebook I wrote demonstrating this result.
Supplemental
Note it’s straightforward to prove
\[\begin{align} \text{Cov}(X, Y) &= \mathbb{E}[(X - \mathbb{E}[X])(Y - \mathbb{E}[Y])] \nonumber \\ &= \mathbb{E}[XY]) - \mathbb{E}[X]\mathbb{E}[Y] \label{eq:cov2exp} \end{align}\]This fact will be used in both of the two proofs below.
Proof for $\mathbb{V}[X + Y] = \mathbb{V}[X] + \mathbb{V}[Y] + 2\text{Cov}(X, Y)$
\[\begin{align*} \mathbb{V}[X + Y] &= \mathbb{E}[(X + Y)^2] - \mathbb{E}[X + Y]^2 \\ &= \mathbb{E}[(X + Y)^2] - (\mathbb{E}[X] + \mathbb{E}[Y])^2 \\ &= \mathbb{E}[X^2] + \mathbb{E}[Y^2] + \mathbb{E}[2XY] - \mathbb{E}[X]^2 - \mathbb{E}[Y]^2 - 2\mathbb{E}[X]\mathbb{E}[Y] \\ &= \left( \mathbb{E}[X^2] - \mathbb{E}[X]^2 \right) + \left(\mathbb{E}[Y^2] - \mathbb{E}[Y]^2 \right ) + \left ( 2(\mathbb{E}[XY]) - \mathbb{E}[X]\mathbb{E}[Y] \right)\\ &= \mathbb{V}[X] + \mathbb{V}[Y] + 2\text{Cov}(X, Y) \end{align*}\]The 4th equality used Eq. $\eqref{eq:cov2exp}$.
Proof for $\text{Cov}(X, (Y + Z)) = \text{Cov}(X, Y) + \text{Cov}(X, Z)$
\[\begin{align*} \text{Cov}(X, (Y + Z)) &= \mathbb{E}[X(Y + Z)] - \mathbb{E}[X] \mathbb{E}[Y + Z] \\ &= \mathbb{E}[XY] + \mathbb{E}[XZ] - \mathbb{E}[X] \mathbb{E}[Y] - \mathbb{E}[X] \mathbb{E}[Z] \\ &= \big( \mathbb{E}[XY] - \mathbb{E}[X] \mathbb{E}[Y] \big) + \big( \mathbb{E}[XZ] - \mathbb{E}[X] \mathbb{E}[Z] \big) \\ &= \text{Cov}(X, Y) + \text{Cov}(X, Z) \end{align*}\]The 1st and 4th equalities used Eq. $\eqref{eq:cov2exp}$.
References:
- https://economictheoryblog.com/2014/11/05/the-coefficient-of-determination-latex-r2/
- https://economictheoryblog.com/2014/11/05/proof/