Robust statistics for ionosphere data: from Rawer’s nombre de corrélation (1951) to maximum-depth regression technique (1998)*

Abstract. Contrast between “normal" and “disturbed" states of the ionosphere early induced the suggestion to present ionospheric data with the aid of a “robust" (i.e. outlierresistant) statistic, namely, the median instead of the conventional arithmetic mean. K. Rawer, in 1951, defined on this concept a nombre de corrélation, which proved to be well resistant to misleading effects of outlier data when analyzing correlations between ionospheric and related data. Various attempts over many years, however, to extend this idea towards a robust fitting of regression models to outliercontaminated data, remained unsatisfactory. Only a few years ago, a mathematically correct and unambiguous technique for robust regression has been reached by Belgian authors (Rousseeuw and Hubert, 1998), based on the new paradigm of “maximum data depth", which is exemplified here for ionospheric data presentation.