1: \begin{abstract}

2:   It is shown that a seemingly harmless reordering of the steps in a

3:   block Gibbs sampler can actually invalidate the algorithm.  In

4:   particular, the Markov chain that is simulated by the

5:   ``out-of-order'' block Gibbs sampler does not have the correct

6:   invariant probability distribution.  However, despite having the

7:   wrong invariant distribution, the Markov chain converges at the same

8:   rate as the original block Gibbs Markov chain.  More specifically,

9:   it is shown that either both Markov chains are geometrically ergodic

10:   (with the same geometric rate of convergence), or neither one is.

11:   These results are important from a practical standpoint because the

12:   (invalid) out-of-order algorithm may be easier to analyze than the

13:   (valid) block Gibbs sampler (see, e.g., \cite{yang:rose:2019}).

14: \end{abstract}