摘要：There are two aspects to the method used by quantitative sociology, building up the data and analysing it, which are separated by an “ airlock ” (a databank). Yet the very process of making these procedures explicit is itself strained by the use of two very different languages : the language of measurement, arising from the natural sciences, and the language of conventional codes, inspired by law, political science and cognitive sciences. The ambiguity is examined here through an historical analysis, linking the issues raised by the data-building process to those resulting from the diversity of analytical tools. A comparison between “logistical regression” (logit models) and “ French data analysis ” shows that they differ mainly in terms of the characters they are presenting : “ variables ” in one case, and “ social groups ” in the other.
