Cette page appartient aux archives web de l'EPFL et n'est plus tenue à jour.
This page belongs to EPFL's web archive and is no longer updated.

Practical 2 : Still some influential observations after cleaning


In part e), after fitting the model we chose, we noticed 4 observations outliers + leverage points, so we decided to remove them. But after refitting the model without thoses observations a new outlier + leverage point observation appears. 

Is that normal ? If so, should we refit the model again without that observations ?

Thanks in advance,

Posted by Jeremy Gotteland on Tuesday 10 December 2013 at 11:07
There will indeed be influential observations when you do regression diagnostics for the final model. In this practical, it is enough to note their existence and comment that one should look at the situation in more detail. There's no need to start refitting the model without these observations. The reason is that, as you (hopefully) noted in your report for practical 1, removing influential observations is not always the most appropriate way to proceed as your example for instance shows. Instead, a lot more reasonable approach would be to consider robust regression as done for example in Serie 11, Exercise 1. But we do not require that in this practical.
Posted by Mikael Kuusela on Tuesday 10 December 2013 at 11:41