I am learning hypothesis testing for Pearson Correlation Coefficient. The source did not explain why the test statistic $$\frac {r\sqrt{n-2}}{\sqrt{1-r^2}}$$ satisfy T distribution with $n-2$ degree of freedom.

Could anyone show me the assumption and proof?


1 Answer 1


(providing an answer to the question)

When the residuals in a linear regression are normally distributed, the least squares parameter $\hat{\beta}$ is normally distributed. Of course, when the variance of the residuals must be estimated from the sample, the exact distribution of $\hat{\beta}$ under the null hypothesis is $t$ with $n-p$ degrees of freedom ($p$ the dimension of the model, usually two for a slope and intercept).

Per @Dason's link, the $t$ for the Pearson Correlation Coefficient can be shown to be mathematically equivalent to the $t$ test statistic for the least squares regression parameter by:

$$t = \frac{\hat{\beta}}{\sqrt{\frac{\text{MSE}}{\sum (X_i - \bar{X})^2}}}= \frac{r (S_y / S_x)}{\sqrt{\frac{(n-1)(1-r^2)S_y^2}{(n-2)(n-1)S_x^2}}}=\frac{r\sqrt{n-2}}{\sqrt{1-r^2}}$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.