| Titre | Nettoyage de fichiers dans le cas de données individuelles : recherche de la cohérence transversale | |
|---|---|---|
| Auteur | Elizabeth Kremp | |
 |
Revue | Economie et prévision |
| Numéro | no 119, 1995/3 | |
| Rubrique / Thématique | Méthodes |
|
| Page | 171-193 | |
| Résumé anglais |
Cleaning Files Containing Individual Data: The Search for Transversal Consistency
by Elizabeth Kremp
This article first defines the notions of aberrant values and extreme values. It then describes the statistical tools and presents different univaried methods for identifying these values. Eight techniques based on these tools and methods are tested on a file of company data for one ratio. One of the conclusions of these tests is that robust statistics need to be used in the methods seeking to identify aberrant points. Three of these techniques are applied to seven ratios for a comparison, evaluation of the role of the choice of ratios and measurement of the cumulative observation elimination phenomena. Two of these techniques produce very similar results. The easiest technique to apply eliminates the observations situated at more than three interquartile intervals from the first and third quartiles. However, if the distribution of the real population for the variable studied differs greatly from a normal distribution, this technique can eliminate too many observations. In this case, a variant that eliminates the observations at more than five interquartile intervals would appear preferable. Source : Éditeur (via Persée) |
|
| Article en ligne | http://www.persee.fr/web/revues/home/prescript/article/ecop_0249-4744_1995_num_119_3_5738 |
