|
Abstract:
|
We consider the ramifications of restricting regression analyses so the error terms satisfy the conditions necessary for valid inference (homoskedasticity, normality, independence, and fit). Specifically, we assess the difficulty and time required to solve the resulting constrained nonlinear optimization problem. We also consider measures of divergence of the resulting regression from the unrestricted regression and how to use this divergence to measure how well the unrestricted regression satisfies these conditions, as well as the validity of standard summary statistics and hypothesis tests for the restricted error regression.
|
