Since the seminal contribution of Teece et al. (1994), the strength, scope and quality of corporate diversification is often detected comparing the observed value of some statistics derived from the diversification patterns of a sample of firms, with its expected value. The latter is obtained under a null hypothesis which assumes some random assignment procedure of sectors to firms. The approaches generally adopted in the literature present two problems. First, being based on the observed value of a statistic, these methods could lead, depending on the nature of the sample, to noisy and non-homogeneous estimates. Second, the benchmark value used to identify the presence and strength of deterministic patterns are obtained under specific and privileged null hypothesis. Both effects could lead to the erroneous classification of spurious random effects as deterministic. This paper shows that the adoption of p-scores as measure of relatedness strongly alleviate the first problem, leading to cleaner and more homogeneous estimates. We design and implement a null hypothesis which rules out random artifacts and effectively identify new features in firm diversification pattern. Using the NBER database on patents, we apply our results to the study of the relationship between the coherence and the scope of corporate patent portfolios.