Michael Robert Dunn <[EMAIL PROTECTED]> wrote in message news:<aqhivn$h31$[EMAIL PROTECTED]>... > Hello a friend of mine needs help with a problem
Ah, the old "I have this friend, who has this problem" ruse ;) > --- > "Suppose that a plotting of the residuals of a regression with respect to > time indicates a significant outlier in the residuals. (Be careful here, > this is not an outlier in the original data but is an outlier in the > residuals of a regression). > 1.. How could such an outlier occur? What does it mean? Well, you might check how close this observation was to being an outlier in the original data -- if you were using a certain cutoff, say t=2.0, you might find it was 1.95 in the original data, and 2.05 here. This would mean your problem is still most likely in the original data. If this is not the case, then there is some unusual case in your independent variables that is pushing this observation into outlier territory. As with any outlier, typingg erorrz on somebody's part are the most likely cause. A related, but separate, class is the changing of the scale of an independent variable, e.g. from francs to euros at a particular time. Or, a mismatch of scaling definition: if my dependent variable is "school meals eaten per month" and my independent variable is "number of students in the school", this will fail when I get to June: I will have just as many students, but fewer meals because there are fewer school days in June. Another likely cause is some unusual disturbance like hurricanes or strikes, which may be affecting the independent variables but not the dependent variable. > 2.. Is the Durbin-Watson d-statistic applicable in the presence of such > an outlier. The DW stat is a measure of autocorrelation, and will be affected by this. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
