c © 2009, Indian Statistical Institute Trimmed ML Estimation of Contaminated Mixtures

We establish a mixture model with “spurious ” outliers and derive its maxi-mum likelihood estimator, the maximum trimmed likelihood estimator MTLE. It may be computed with a trimmed version of the EM algorithm which we call the EMT algorithm. We analyze its properties and compute various breakdown values of the estimator for normal mixtures thereby proving ro-bustness of the method. AMS (2000) subject classification. Primary 62H12; Secondary 62F35