This happens because null models may vary a lot in their restrictiveness; that is, in how much a null matrix preserves features of the original matrix 6.

The first is a Bernoulli random null model in which the null matrix has the same total number of interactions as the original matrix, albeit randomly positioned.

The Null Ideal of a Matrix: 11083 William P. Wardlaw and Jean-Pierre Grivaux The American Mathematical Monthly , p. 371 (1 page) ...