Both historical and recent developments of quantitative research in linguistics brought out a great amount of data without a unifying method. The older data have been computed mainly by hand from limited samples of shorter texts, with limited possibilities of data combinations. Newer data based on large corpora offer a great number of quantitative characteristics even in the most different combinations, but they have been mainly extracted from heterogeneous text materials. Statistically, the older data can be considered as less exact. New data, with respect to enormous extent of corpora, can be considered as most exact. Therefore, problems arise not only because of the above mentioned methodological disparities of old and new approaches of computation, but also because of different details studied or because of limited possibilities of direct comparison. Deeper statistical and probabilistic questions arise too, and their discussion should not be ignored.