In statistical process control, detection of special causes of variation and the estimation of the time when they occur are two important tasks for process improvement. When dealing with normal independent observations, maximum likelihood estimators for a change-point have been derived, and their structure happens to correspond with a least squared formulation. This formulation facilitates the estimation of a change-point. However, in the presence of outliers, the estimation bias increases. To deal with this problem, regression analysis uses, among other procedures, a technique called least absolute residuals, where the norm L2 is replaced by the norm L1 in the objective function to be minimized. Using this idea, a robust estimator for a change-point in a series of independent normal observations is developed. Preliminary simulation performance results are presented and compared with the corresponding maximum likelihood estimators in the presence of outliers. Additionally, practical implementation is described by using a numerical example. It is expected that this estimator is found useful for practitioners when dealing with suspicious outlying observations.
