Approximating the Bayesian estimate of the standard deviation in a linear model