Basics
Contrasts
def. of contrast
use of contrast
hybrid yield
comparison
estimate
gmodels
fit.contrast()
estimable()
aov()
model.tables()
confidence interval
hypothesis test
SSw
orthogonal contrast
calculate SSw
default: contr.treatment
Multiple Comparisons
multiple comparisons
error rate
simultaneous error rate
comparisonwise error rate
experimentwise error rate
FDR
SFW
methods
Benferroni
linear.contrast()
Holm
SFW
FDR
FDR
Scheffe
for all contrasts
simultaneous CIs for all contrasts
linear.contrast()
pairwise comparison
methods
Tukey HSD
simultaneous CIs for all possible pairwise comparisons
familywise error rate
studentized range statistics
require equal sample size
aov(); TukeyHSD()
SNK
FDR
studentized range statistics
REGWR
SFW
studentized range statistics
LSD
experimentwise error rate
Duncan's
Dunnet's
compare with control
compare.to.control()
glht()
MCB
compare with best
compare.to.best()
comparison after looking at data
library(cfcdae); pairwise()
General Steps
Check the design type
Factorial Design
step1: set contr.sum
n=1 replicate
half-normal plot to get the significant factors
pseudo p-value
pooling high-order interactions
Complete Block Design
CRB design are balanced
effects are orthogonal
block for variance reduction
Latin Square Design
two block variables
replicating Latin Square
design for residual effect
Graeco Latin Square
three blocking variables
Check the data type
unbalanced data
SS type
contr.sum
contrast
linear.contrast
contr.sum + fit.contrast
multiple comparison
glht() to do TukeyHSD
marginal means
emmeans
interaction plot
emmeans
Fit a model
Check assumptions
normality
qqplot()
constant variance
diagnostics
side-by-side boxplot
residual vs. fitted
diagnostic plots
transformation
power family
boxcox
other stabilizing methods
balanced design preferred
independence
DW test
Investigate effects
interaction effects
interaction plot
pairwise comparison
TukeyHSD
underline diagram
main effects
profile plot
pairwise comparison
TukeyHSD
underline diagram