ABSTRACT
We noted in previous chapters that the method of least squares provides us with good estimates of the regression intercept and slope (β0 and β1, respectively) if the random error term ε is assumed to follow a normal distribution having zero mean and constant variance. But what if we are unsure as to the form of the probability density function of ε? Obviously we need a regression procedure that will yield reasonable estimates of β0 and β1 for a whole host of possible (unknown) random error distributions. Such regression procedures are said to be distribution free or nonparametric in nature and, as we shall now see, are, for the most part, based on the ranks of certain values determined from the sample observations. Two rankoriented regression methods will be presented shortly; one uses the ordinary median of a certain set of slope estimates and the other uses the weighted median of these slope estimates.
