In my textbook, it says, we can see with some algebraic manipulation that: $\beta= r* \frac{s_y}{s_x} $
However, the text does not actually show this manipulation. Can anyone show it or the intuition behind this formula?
In my textbook, it says, we can see with some algebraic manipulation that: $\beta= r* \frac{s_y}{s_x} $
However, the text does not actually show this manipulation. Can anyone show it or the intuition behind this formula?
Let $c$ be the covariance between $y$ and $x$. The corresponding correlation is defined as $r = c/(s_xs_y)$. So then $$ \beta = \frac{c}{s_xs_x} = \frac{c}{s_xs_x}\Big(\frac{s_y}{s_y}\Big) = \frac{c}{s_xs_y}\Big(\frac{s_y}{s_x}\Big) = r\Big(\frac{s_y}{s_x}\Big). $$
As for intuition, one way to think about why this makes sense is to think about the units involved. The correlation coefficient is dimensionless. The slope is in $y$-units per $x$-units -- for example, if we regress income in dollars on age in years, the slope of that regression has units of dollars-per-year. Now note that the ratio $s_y/s_x$ is also in $y$-units per $x$-units, since $s_y$ is in units of $y$ and $s_x$ is in units of $x$ So multiplying the dimensionless correlation coefficient by $s_y/s_x$ results in something that is in the units that we know the slope has. Obviously this is not a full intuition for why the right-hand expression equals the slope, but it's at least an intuition for why it might be similar.
This is the slope of a simple OLS regression with intercept. The full derivation without linear algebra can be found here.
These are the two parameter estimates in the case of a linear regression model of the form: $\large Y\,=\,\beta_o \,+\,\beta_1\,X:$
$$\ \hat\beta_1\,=\, \underbrace{\text{cor}(Y, X)}_{r}\,\frac{\text{SD}(Y)}{\text{SD}(X)}\,=\,\frac{\text{cov} (Y,X)}{\text{var}(X)}\small $$
remembering that $ \text{cor} (X,Y) = \frac{\text{cov}(X,Y)}{\text{SD}(X) \text{SD}(Y)}.$
is the slope, while the intercept can be calculated knowing that the regression line passes through the means:
The intercept
$$\large \hat\beta_o\,=\,\bar Y\,-\,\hat\beta_1\,\bar X $$
To be clear, I believe that your formula uses $r$ as the Pearson correlation coefficient, and $S_y= \text{SD}(Y)$ (and $S_x= \text{SD}(X)$.