1: \begin{abstract}

2: We consider the set $\blockClass{p}$ of parametric block correlation matrices with $p$ blocks of various (and possibly different) sizes,

3: whose diagonal blocks are compound symmetry (CS) correlation matrices and off-diagonal blocks are constant matrices. 

4: Such matrices appear in probabilistic models on categorical data, when the levels are partitioned in $p$ groups,

5: assuming a constant correlation within a group %(within group correlation) 

6: and a constant correlation for each pair of groups.\\ %(between group correlations).\\

7: We obtain two necessary and sufficient conditions for positive definiteness of elements of $\blockClass{p}$.

8: Firstly we consider the block average map $\phi$, consisting in replacing a block by its mean value.

9: We prove that for any $A \in \blockClass{p}$, $A \pdsymb 0$ if and only if $\avemap(A) \pdsymb 0$.

10: Hence it is equivalent to check the validity of the covariance matrix of group means,

11: which only depends on the number of groups and not on their sizes.

12: This theorem can be extended to a wider set of block matrices.

13: Secondly, we consider the subset $\blockClassI{p}$ of $\blockClass{p}$ 

14: for which the between group correlation is the same for all pairs of groups. 

15: Positive definiteness then comes down to find the positive definite interval of a matrix pencil on $\symMat{p}$.

16: We obtain a simple characterization by localizing the roots of the determinant with within group correlation values.

17: \end{abstract}

18: