1: \begin{abstract}

2:   We investigate sets of random variables that can be arranged

3:   sequentially such that a given variable only depends conditionally

4:   on its immediate predecessor. For such sets, we show that the full

5:   joint probability distribution may be expressed exclusively in terms

6:   of uni- and bivariate marginals. Under the assumption that the CMB

7:   power spectrum likelihood only exhibits correlations within a banded

8:   multipole range, $\Delta\ell$, we apply this expression to two

9:   outstanding problems in CMB likelihood analysis. First, we derive a

10:   statistically well-defined hybrid likelihood estimator, merging two

11:   independent (e.g., low- and high-$\ell$) likelihoods into a single

12:   expression that properly accounts for correlations between the

13:   two. Applying this expression to the WMAP likelihood, we verify that

14:   the effect of correlations on cosmological parameters in the

15:   transition region is negligible in terms of cosmological parameters

16:   for WMAP; the largest relative shift seen for any parameter is

17:   $0.06\sigma$. However, because this may not hold for other

18:   experimental setups (e.g., for different instrumental noise

19:   properties or analysis masks), but must rather be verified on a

20:   case-by-case basis, we recommend our new hybridization scheme for

21:   future experiments for statistical self-consistency reasons. Second,

22:   we use the same expression to improve the convergence rate of the

23:   Blackwell-Rao likelihood estimator, reducing the required number of

24:   Monte Carlo samples by several orders of magnitude, and thereby

25:   extend it to high-$\ell$ applications.

26: \end{abstract}

27: