Identifying outlying growth profiles in the growth of conifers

Our objective in this paper is the detection of atypical growth profiles, which is illustrated in a growth study from 74 families of conifers. Our approach starts by fitting a 2-level linear model where we assign the measurements made on time in each family to the first level of the model, and assign the families to the second level. In order to identify atypical growth profiles we analyze the (multivariate) residuals in the second level of the fitted model. Mahalanobis distances to the origin indicate potential atypical growth profiles, however, Hadi's more sophisticated procedure concludes that there are no outlying residuals, thus avoiding the wrong conclusion that observations with high Mahalanobis distances to the origin are necessarily outliers.