Section: New Results

A variable clustering approach for the typology of units: a survey on farming and environment

Participants : Jérôme Saracco, Marie Chavent.

A survey on farming and environment dealing with the current transformations of the farmer job is considered. We propose to replace the usual data mining strategy which consists of applying Multiple Correspondence Analysis by a variable clustering approach. Clustering of variables aims at lumping together variables which are strongly related to each other and thus bring the same information. The ClustOfVar approach used in this paper provides at the same time groups of variables and their associated synthetic variables. In this algorithm, the homogeneity criterion of a cluster is defined by the squared Pearson correlation for the quantitative variables and by the correlation ratio for the qualitative variables. The step of variable clustering enables to get synthetic variables that can be read as a gradient. In our case study, values correspond to some relevant groupings of categories. This enables to interpret and name easily the synthetic variables. Trends in the opinion of farmers are thus highlighted with the variable clustering approach. Then we clarify these first results by applying a clustering method on the scores of the individuals measured by the synthetic variables. At the sociological level, the supply provided by the synthetic variables to interpret the clusters of farmers is obvious.

These results have been obtained in collaboration with Vanessa Kuentz from Irstea, UR ADBX.

They have been published in Journal de la Société Française de Statistique [31] .