The lead article in the February issue is "A New Statistic for Influence in Linear Regression," by Daniel Pena. The author defines a new way to measure the influence of an observation in linear regression based on how this observation is being influenced by the rest of the data. More precisely, the proposed statistic is defined as the squared norm of the vector of changes of the forecast of one observation when each of the sample points is deleted one by one. The new statistic is asymptotically normal and is able to detect a group of high-leverage similar outliers that will be undetected by Cook's statistic, and is useful for detecting heterogeneity in regression models in large high-dimensional datasets.
展开▼