This Table is not produced under usual circumstances. The measurement is overparameterized at this point, so response residual computations can be misleading.

 

To examine response residuals after adjusting for bias, model the interaction term as the main effect and do not model the main effects, i.e.,

Usual model: log (P/P) = Bn - Di - Fk
and then compute interaction Bni

Instead model: log(P/P) = Bni - Fk directly