Performance of the orthogonal least median squares regression

In this work the orthogonal regression in least median squares, OLEM, is presented. Using a Monte Carlo simulation, the work studies the behaviour of the regression when faced with outliers and lack of normality. The estimated slope and intercept are compared with those provided by the least squares regression, LS, the least median squares regression, LMS, and the orthogonal least squares regression, LSO. OLEM is the most resistant to influential data, but shows greater variability than LSO when the outliers are random. OLEM gives the better estimation of the standard deviation in prediction evaluated as a contribution of the variance of the data on both axes.