Robust Cost Modeling

The purpose of this note is to contrast the usual ordinary least squares regression (OLS), which uses a normal error model, against a robust model that uses a log-cosh error model as described in the paper. For any realistic set of observations, the two models generally give different coefficient values, sometimes dramatically so. We argue that the coefficients coming from the robust model are better estimates of the true underlying values, since the robust error model better matches the properties of the data.

Request the full whitepaper by filling out the form below.