Abstract
A fundamental characteristic that every social scientist requires of his statistical results is that they remain "stable". This concept includes various ideas, each of a different nature. Research demonstrates that statistical results are not resistant to data recoding even when it is a question of a simple recoding of the variables by their modes which is often used in social science research. Thus, there is a first criteria for stability: the results must remain independant of a simple recoding of the basic data.
A second research shows that there are at least two types of randomness: a first type is the introduction of arbitrary data that is distributed in a homogeneous fashion among the subjects. This type of randomness modifies very little the structure of the results. A second type is the introduction of fictitious subjects with descriptive variables that have arbitrary values. This type of randomness can modify the structure of the results. These two notions can be confounded by reseachers on a conceptual level, even if they do distinguish between them on a pratical level in their reseach activity. Thus there is a second criteria for stability: the structure of the results must resist the progressive introduction of arbitrary information or "noise".
A third research, which uses four "classical" methods of multivariate analysis employed in social science (two classificatory types: dynamic clusters and fixed center typology analysis - two factorial types: correspondance analysis and principal component analysis), demonstrates that there exist basic types (which constitute a basic typology) that remain unchanged for these four methods, but they are only revealed by the confrontation and the crossing of the results of several methods. Thus there is a third criteria for stability: the results must not rely on one method or one kind of method. We end with a description of other caracteristics that clearly define a basic typology.
Get full access to this article
View all access options for this article.