Beruflich Dokumente
Kultur Dokumente
Part 3
Latin Squares Design - Review
Latin Squares design deals with the case
where there are two nuisance factors to control.
It is often used when subject-to-subject
variability is large.
Restriction: both nuisance factors and
treatment must have the same number of
levels.
Only one observation per combination.
Latin Squares Model
yijk = + i + j + k + ijk
nuisance (rows)
nuisance (columns)
treatment
Latin Squares
The formula for sum of squares:
SST = SSRows + SSColumns + SSTreatments + SSE
Degrees of freedom:
SST = p2-1
SSRows = p-1
SSColumns = p-1
SSTreatments = p-1
SSE = (p-2)(p-1)
Latin Squares
The appropriate statistic for testing for no differences in
treatment means is
MSTreatments
F0 =
MSE
SSTreatments 4622
MSTreatments = = = 1541
p 1 4 1
SSRows 986
MSRows = = = 329
p 1 4 1
SSColumns 1468
MSColumns = = = 489
p 1 4 1
SSE 368
MSE = = = 61
( p 2) ( p 1) ( 4 2) ( 4 1)
Latin Squares Example
Treatment Totals (y.j.)
A 1063
B 880
C 967
D 922
MSTreatments 1541
F0 = = = 25.26
MSE 61 > qf(1-0.05, df1=3, df2=6)
4.757063
F , p1,( p2)( p1) = F0.05,41,(42)(41) = F0.05,3,6 = 4.76
6
0.10
4
Sample Quantiles
model1$residuals
0.05
2
0
0.00
-2
-0.05
-4
-6
-0.10
9.2 9.4 9.6 9.8 10.0 10.2 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
y.. j y..i
tij =
1 1
+
p p
Multiple Comparisons
Since we the ANOVA test showed that the means are
different, we can continue to compute multiple
comparisons using the Tukey method with significance
level 0.05.
1
tij > q p,( p1)( p2),
2
1 1 1 > qtukey(1-0.05,4,6)
q p,( p1)( p2), = q4,6,0.05 = (4.90) = 3.46 4.895599
2 2 2