6 Transformation & weighting to correct model inadequacies
The residual analysis of Section 4 diagnoses two common failures: a variance that is not constant, and a relationship that is not a straight line. This section gives the two standard remedies. The first is to transform the response, a regressor, or both, so that on the new scale the assumptions hold. The second, for non-constant variance whose form is known, is to keep the original scale and fit by weighted least squares, giving each observation a say in proportion to its precision.
6.1 Variance-stabilizing transformations
6.2 Transformations to linearize the model
6.3 The Box–Cox method
6.4 Weighted least squares
6.5 Practice problems
6.2 Transformations to linearize the model
6.3 The Box–Cox method
6.4 Weighted least squares
6.5 Practice problems
Questions on this section
Stuck on something here? Ask below and it stays attached to this topic.