UA MATH571B Things to know about statistical model of experimental design
Experimental with one treatment factor
Simple Experiment (Randomized Complete Design)
Balanced
Statistical Model is
yij=μ+τi+ϵij,i=1,⋯,a,j=1,⋯,n
μ is the grand mean, τi is the treatment effect of factor level i.
Assumptions:
- Fixed Effect: ∑i=1aτi=0
- Homogenous normality: ϵij∼iidN(0,σ2)
ANOVA table is (N=na)
Hypothesises of ANOVA F test is
H0:τ1=⋯=τa=0Ha:at least one is not zero
And the rejection rule is F0>F1−α,a−1,N−a.
OLS estimators for the parameters are
Unbalanced
Statistical Model is
yij=μ+τi+ϵij,i=1,⋯,a,j=1,⋯,ni
μ is the grand mean, τi is the treatment effect of factor level i.
Assumptions:
- Fixed Effect: ∑i=1aτi=0
- Homogenous normality: ϵij∼iidN(0,σ2)
RCBD
yij=μ+τi+βj+ϵijϵij∼iidN(0,σ2);i=1,⋯,a;j=1,⋯,b
where τi means treatment effect and βj means blocking effect.
Assumptions:
- Fixed Effect: ∑i=1aτi=0,∑i=1bβj=0
- Homogenous normality: ϵij∼iidN(0,σ2)
ANOVA table is
Source |
SS |
df |
MS |
F |
Treatment |
SSM |
a-1 |
MSM=dfMSSM |
FM=MSM/MSE |
Blocking |
SSB |
b-1 |
MSB=dfBSSB |
FB=MSB/MSE |
Residuals |
SSE |
(a-1)(b-1) |
MSE=dfESSE |
|
Total |
SST |
N-1 |
MST=dfTSST |
|
i=1∑aj=1∑b(yij+yˉ..−yˉi.−yˉ.j)2=i=1∑aj=1∑beij2=SSEi=1∑aj=1∑b(yˉ.j−yˉ..)2=i=1∑aj=1∑bβ^j2=SSBi=1∑aj=1∑b(yˉi.−yˉ..)2=i=1∑aj=1∑bτ^i2=SSMSST=SSM+SSB+SSE
Parameter estimation:
yi.=j=1∑byij,yˉi.=byi.y.j=i=1∑ayij,yˉ.j=ay.jy..=i=1∑ayi.=j=1∑by.j,yˉ..=Ny..,N=abτ^i=yˉi.−yˉ..,β^j=yˉ.j−yˉ.j,eij=yij−yˉi.−yˉ.j+yˉ..
Latin Square Design
Graeco-Latin Square Design
BIBD
Experimental with multiple treatment factors
Factorial design with fixed factors
Factorial design with random factors
Factorial design with mixed factors
Nested design
Split-plot design