importance of samples in regression

Question

In regression analysis it could happen that some samples are more relevant than others (i.e. they are more important in understanding the underlying model). Is there any technique to train a regressor by weighting training samples according to their relevance? A trivial example could be giving 0 weight to outliers.

Answer 1

You could probably add a dichotomous variable as covariate on the more importand samples. See how that affects your model

Or more importantly use weighted least squares. Read about it more here: http://www.itl.nist.gov/div898/handbook/pmd/section1/pmd143.htm