Robust linear regression
Sujan recommended me to use the robust regression models (RLM) instead of the standard OLS (ordinary least squares). What are the differences?
Robust: less affected by outliers
A drawback of OLS is that the resulted regression line can be significantly altered by some outliers. As the OLS try to find a best line of the minimum SS, the optimum would likely to have more focus on outliers which have large SS.
RLM stands for all approaches to reduce the effect of outliers. Typically, based on a threshold residual, RLM uses absolute errors instead of SS for points with larger residual than the threshold.
Another way is to use different weight for each point based on the residual. There are many functions to define the weight.