GLMs Interaction term

[r_v.s]

In the core reading, it says
Note that each factor has a base level that should not be included in the model,and for interactions, each base level row and column of the interaction
parameter matrix should be removed


would you please explain this in the light of the example below it? everything has been included in it, so what is being removed?


Also, where Deviance is defined, it says "This compares the observed value Yi to the fitted value mu_i with allowance for the weights and assigning higher importance to errors where the variance should be small." Would you pls explain the rationale behind higher importance to lower variance errors?

 

[manish.rex]

In the core reading, it says
Note that each factor has a base level that should not be included in the model,and for interactions, each base level row and column of the interaction
parameter matrix should be removed


would you please explain this in the light of the example below it? everything has been included in it, so what is being removed?


When it says that the a single combination of levels of factors coming from interacting terms shouldnot be included in the model, it means that it should be held constant in the model-kind of baseline estimate, and then for other possible interaction terms, parameters should be added.

It does not mean that the row last row having all values as one and first column having all values as 1 have to be removed. It only means that this particular row-column combination should be always there in the model.You can think of it as the intercept term in linear regression.