10.4Two-Factor Analysis of Variance: Introduction and Parameter Estimation 455
However, if we letμdenote the average value of theμi— that is,
μ=
∑m
i= 1
μi/m
then we can rewrite the model as
E[Xij]=μ+αi
whereαi =μi−μ. With this definition ofαias the deviation ofμifrom the average
mean value, it is easy to see that
∑m
i= 1
αi= 0
Atwo-factoradditivemodelcanalsobeexpressedintermsofrowandcolumndeviations.
If we letμij=E[Xij], then the additive model supposes that for some constantsai,i=
1,...,mandbj,j=1,...,n
μij=ai+bj
Continuing our use of the “dot” (oraveraging) notation, we let
μi.=
∑n
j= 1
μij/n, μ.j=
∑m
i= 1
μij/m, μ..=
∑m
i= 1
∑n
j= 1
μij/nm
Also, we let
a.=
∑m
i= 1
ai/m, b.=
∑n
j= 1
bj/n
Note that
μi.=
∑n
j= 1
(ai+bj)/n=ai+b.
Similarly,
μ.j=a.+bj, μ..=a.+b.
If we now set
μ=μ..=a.+b.
αi=μi.−μ=ai−a.
βj=μ.j−μ=bj−b.
then the model can be written as
μij=E[Xij]=μ+αi+βj