Robust Analysis of Covariance

The simple analysis of covariance situation with two groups and one concomitant variable is considered. The parameters of this model with outliers present are estimated by the methods of at least squares and M-estimation. By use of simulation, several forms of M-estimators are compared with the least squares method. In terms of their efficiencies the tests on the equality of slopes and of adjusted means in the presence of outliers are examined under the null hypothesis by studying the behavior of t-like statistics based on the least squares and M-estimates. An example illustrates the techniques discussed.