[SciPy-User] robust fit

[Sturla Molden]

Den 31.05.2011 17:57, skrev Charles R Harris:
>
>
> I've also had good results with Tukey's biweight

That is the one I prefer too, particularly with a robust estimate of the 
standard error (MAD/0.6745). If the residuals are normally distributed, 
there is hardly any difference from the least squares fit.

Sturla