Nouvelle méthode de traitement de séries de données tronquées dans l'étude de la pollution atmosphérique

The statistical treatment of sets of data including some values that are below the detection limit is frequently used in air pollution studies. These values are usually, and conventionally, set at one half of the detection limit. The present paper discusses the validity of this convention and shows on both theoretical and empirical grounds that it is better to set these values at two thirds rather than one half of the detection limit.