Abstract The point of the present paper is an elementary distinction, i. e. between definite and modifiable statistical units. This classification will prove to throw light on the paradoxon that correlation analysis, while of fundamental importance in many sciences, in others is severely criticized and disclaimed of quantitative significance. In fact, only when referring to a definite unit a correlation index is unconditionally quantitative. A. moment's consideration will reveal that a correlation index referring to a modifiable unit is not directly commensurable to the corresponding index of another unit. For quantitative analysis of modifiable units detailed hypotheses regarding the effect of the unit size are necessary. Some methods for such analysis are indicated and examplified below. We reserve, however, for an ulterior paper the explicit formulae and a more exhaustive investigation of modifiable units.