| Abstract: | Most Maximum Likelihood Estimates for the slope in the errors-in-variables models without replications rely on artificial assumptions, e.g., known ratio of variances of the two error terms. When this ratio is known, these estimates will perform well. If the ratio is not known, an alternative estimator has to be used. The Median-Slope estimator, proposed in this paper, is one such alternative. No assumptions are made about the parameters or the underlying distribution, but the predictors, although unknown, are assumed to be fixed and ordered. The Median-Slope estimator has a high breakdown point and is shown, via simulation, to perform well, especially for heavy tailed distributions. Large sample properties are investigated and the estimator is applied to some real data set. |
|---|