In a linear regression $Y = X\beta + \epsilon$ with unknown error variance $\sigma^2$, which distribution is used for the test statistic when testing a hypothesis about a single coefficient $\beta_j$?