Imposing constraints on the between-studies covariance matrix