Matching Bayesian and frequentist coverage probabilities when using an approximate data covariance matrix