Basic ANOVA concepts
Calculating the ANOVA
One-way ANOVA
Two-way ANOVA
N-way ANOVA
With the rats, we are testing two levels of the
factor red blood cell mass.
What if there are more than two factor levels?
Suppose we want to test 5 levels of cotton
weight

Robust Design
Dr. Carol Romanowski
RIT
Always pick the level with the highest S/N
ratio
Four classes of factors
Class
Class
Class
Class
I: Big effect on both S/N and mean
II: Big effect only on S/N
III: Big effect only on mean
IV: No/little effect on eith

Robust design
I.
Modifying orthogonal arrays
Usually not necessary to have more than three levels of numeric factors
But for nominal factors, we may need more levels
Three techniques to change arrays:
1. Downgrading a column (dummy levels)
2. Upgrading a

Session 2
Review: the Z-score
The z-score represents the number of standard deviations that a data
value is from the mean. We find the z-score by subtracting the mean
from the individual data values and dividing this result by the standard
deviation.
Eith

10/10/2013
Perform statistical testing
ANOVA
Refine model
Remove non-significant variables from full model
Analyze residuals
9/21/11
May again refine model if assumptions are
violated or model is inadequate
Interpret results
Graphical analysis
Main ef

10/10/2013
SST = SSTreatments + SS Error
SSTreatments (large reflects large differences
large)
large
in treatment means
SSTreatments (small likely indicates no
small)
small
differences in treatment means
Usually calculate SSTotal and SSTreatments first,
t

Optimize around a function, not a number
Optimize over a range of values, not just one
Dynamic factor chosen before the
experiment, using engineering knowledge
System
Golfing
Skiing
Car
Car
Automatic Transmission
Mill
Controller
Injection molding
Metal p

Descriptive statistics review
Basic statistics
In order to understand data, we need to have a method to
describe them mathematically. We call that statistics.
Descriptive statistics do exactly what they say they
mathematically describe the data. These sta

Blocking in RBCD and Latin
Squares
Blocking and nuisance factors
Definition of a nuisance factor
Three types:
K&C
K&U
U&U
Can you think of examples?
How do we compensate for NFs?
Experimental error reflects random error
and variability between tes