(h(g) - h)(h(g-s) - h)'; the variance of the vector h is
where q denotes the value at which the autocorrelation function of the Gibbs sample tapers off (Chib, 1995, p. 316). Using the delta method, an estimate of the numerical standard error of the marginal likelihood on the logarithmic scale is where h 1 = (h11,hr21,h^1 ,hA1)'. In short, a complete set of composed-error model estimates is available from combining ideas in Gelfand and Smith (1990), Koop et al. (2001) and Chib (1995).
Was this article helpful?