How can I evaluate the result of robust regression?

Hello ALL!

I used nlinfit() to do the regression, but some of my input data is abnormal.Then I switch option.Robust to 'On'. I seems the result quite good.But how can I evaluate the result without R-square(the definition of R-square in Robust regression seems inapplicability[http://www.mathworks.cn/matlabcentral/newsreader/view_thread/317948])?

If there is a INDEX can prove the result of robust regression is better than the ordinary regression, that maybe the best! Thanks for your reading!

Answer 1

[Roger Wohlwend]

It is true, you cannot calculation the R-square of the robust regression, but you can do something similar: I would use the weights the algorithm assigns to each data point to exclude the outliers - just define a threshold, say 0.1, if the weight is smaller, ignore the data point - and calculate the R-square of the remaining data points.