We study gravitational clustering of mass points in three dimensions with random initial positions and periodic boundary conditions (no expansion) by numerical simulations. Correlation properties are well defined in the system and a sort of thermodynamic limit can be defined for the transient regime of cluste ring. Structure formation proceeds along two paths: (i) fluid-like evolution of density perturbations at large scales and (ii) shift of the granular (non fluid) properties from small to large scales. The latter mechanism finally dominates at all scales and it is responsible for the self-similar characteristics of the clustering.

We report the correlation analysis of various redshift surveys which shows that the available data are consistent with each other and manifest fractal correlations (with dimension $D \simeq 2$) up to the present observational limits ($\approx 150 \hmp$) without any tendency towards homogenization. This result points to a new interpretation of the number counts that represents the main subject of this letter. We show that an analysis of the small scale fluctuations allows us to reconcile the correlation analysis and the number counts in a new perspective which has a number of important implications.

We analyze subsamples of Abell and ACO cluster catalogs, in order to study the spatial properties of the large scale matter distribution. The subsamples analyzed are estimated to be nearly complete and are the standard ones used in the correlation analysis by various authors. The statistical analysis of cluster correlations is performed without any assumptions on the nature of the distribution itself. The cluster samples show fractal correlations up to sample limits ($\approx 70 h^{-1} Mpc$) with fractal dimension $D \approx 2$, without any tendency towards homogenization. Our analysis shows that the standard correlation methods are incorrect, since they give a finite correlation length for a distribution that does not possess one. In particular, the so called correlation length $r_{0}$ is shown to be simply a fraction of the sample size. Moreover we conclude that galaxies and clusters are two different representations of the same self similar structure and that the correlations of clusters are the continuation of those of galaxies to larger scales.