Title: DOE 121b General Factorial Design
1DOE 12-1b General Factorial Design
2DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designs
3DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be right!
4DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
5DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Observation row, column, 3rd factor, repetition
6DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Mean of full population
7DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Effect on observation due to row treatment level
8DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Effect on observation due to column treatment
level
9DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Effect on observation due to 3rd treatment level
10DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Effect due to row-column interaction
11DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Effect due to row-3rd-factor interaction
12DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
Effect due to column-3rd-factor interaction
13DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
and this is only part of the terms that affect
the observation
14DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
a 3-Factor Interaction effect
15DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
the experimental error
16DOE 12-1b General Factorial Design
We might expect that the Two-Factor Factorial
Design could be extended to Three-Factor (and
higher) Factorial Designsand wed be
right! Lets consider the Three-Factor ANOVA case.
and with
17The Sums of Squares needed for the F-statistic
are easily generalized
18The Sums of Squares needed for the F-statistic
are easily generalized
19The Sums of Squares needed for the F-statistic
are easily generalized
The Sums of Squares for the main effects of
treatments
20and
21and
The Sums of Squares for the Interactions are
22and
23and
Finally, the 3-factor interaction term is
24These allow us to find the Sum of Squares of the
Error,
25These allow us to find the Sum of Squares of the
Error,
These Sums of Squares allow us to calculate
F-statistics for all direct and interaction
treatment effectsthere will be one F0 per A, B,
C, AB, BC, AC, and ABC (7 in all!).
26These allow us to find the Sum of Squares of the
Error,
These Sums of Squares allow us to calculate
F-statistics for all direct and interaction
treatment effectsthere will be one F0 per A, B,
C, AB, BC, AC, and ABC (7 in all!). For example,
to test the significance of treatment A,
27For the significance of treatment B,
28For the significance of treatment B,
For the significance of treatment C,
29For the significance of treatment B,
For the significance of treatment C,
For the significance of treatment interaction AB,
30For the significance of treatment interaction AC,
31For the significance of treatment interaction AC,
For the significance of treatment interaction BC,
32For the significance of treatment interaction AC,
For the significance of treatment interaction BC,
For the significance of treatment interaction ABC,
33For a main effect (A, B or C X), treatment X
(given an a value) will have a statistically
significant effect on the observations if
where
and X A, B, or Cwith x a, b, or cas
appropriate. (We have been using A to represent
row treatments, B to represent column treatments,
and C as the symbol for the 3rd treatment.
34For an interaction effect (A, B or C X, Y),
interaction treatment XY (given an a value) will
have a statistically significant effect on the
observations if
where
and X, Y A, B, or Cwith x, y a, b, or cas
appropriate. (We have been using A to represent
row treatments, B to represent column treatments,
and C as the symbol for the 3rd treatment.
35Example A soft drink bottler is interested in
obtaining more uniform fill levels in the bottles
produced by his manufacturing process. The
filling machine theoretically fills each bottle
to the correct target filling height, but in
practice there is variation around this target.
The bottler would like to better understand the
source of this variability so as to eventually be
able to reduce it.
The process engineer can control three variables
during the filling process percent carbonation
(A), operating pressure in the filling machine
(B), and the number of bottles produced per
minuteline speed (C).
36The filler pressure and the line speed are easy
to control but the carbonation isnt.
Carbonation depends on product temperature which
is harder to control.
37The filler pressure and the line speed are easy
to control but the carbonation isnt.
Carbonation depends on product temperature which
is harder to control. For the experiment, the
decision is made to control carbonation at three
levels 10, 12 and 14. Two levels are chosen
for filling pressure 25 psi and 30 psi. Two
levels are selected for line speed 200 bpm and
250 bpm bottles per minute. The experimenter
decides to do two repetitions, per set of
treatment levels. All runs (3222 24) are
done in random order.
38The filler pressure and the line speed are easy
to control but the carbonation isnt.
Carbonation depends on product temperature which
is harder to control. For the experiment, the
decision is made to control carbonation at three
levels 10, 12 and 14. Two levels are chosen
for filling pressure 25 psi and 30 psi. Two
levels are selected for line speed 200 bpm and
250 bpm bottles per minute. The experimenter
decides to do two repetitions, per set of
treatment levels. All runs (3222 24) are
done in random order. The observation made is the
average deviation from the target fill height
observed in a production run of bottles at each
set of conditionsthis is the so-called response
variable.
39The filler pressure and the line speed are easy
to control but the carbonation isnt.
Carbonation depends on product temperature which
is harder to control. For the experiment, the
decision is made to control carbonation at three
levels 10, 12 and 14. Two levels are chosen
for filling pressure 25 psi and 30 psi. Two
levels are selected for line speed 200 bpm and
250 bpm bottles per minute. The experimenter
decides to do two repetitions, per set of
treatment levels. All runs (3222 24) are
done in random order. The observation made is the
average deviation from the target fill height
observed in a production run of bottles at each
set of conditionsthis is the so-called response
variable. Positive deviations indicate fill
heights above the target, negative indicate fill
heights below the fill height target.
40Data Fill Height Deviations (mm)
41Data Fill Height Deviations (mm)
42Finding the Sums of Squares
43This could also be called the SSCarbonation.
44This could also be called the SSPressure.
45This could also be called the SSLine Speed.
46This is the SS for the Carbonation-Operating
Pressure interaction.
47The Sum of Squares for the Carbonation-Line Speed
interaction.
48This is the Sum of Squares for the Pressure-Line
Speed interaction.
49The Sum of Squares for the three-way interaction
of the Carbonation, Operating Pressure, and Line
Speed.
50The Sum of Squares for the Error is thus,
51The Sum of Squares for the Error is thus,
The significance of Carbonation in affecting fill
height deviation depends on
52Carbonation A will be significant in affecting
fill height deviation if this F0 is
53Carbonation A will be significant in affecting
fill height deviation if this F0 is
For a 0.01, using appendix 612 this
inequality becomes 178.412 gt 6.93. It is clear
that Carbonation has a significant effect on
variability in fill height along the bottling
line.
54Carbonation A will be significant in affecting
fill height deviation if this F0 is
For a 0.01, using appendix 612 this
inequality becomes 178.412 gt 6.93. It is clear
that Carbonation has a significant effect on
variability in fill height along the bottling
line. In fact, p lt 0.0001 in this case.
55The significance of filler pressure in affecting
fill height deviation depends on
56The significance of filler pressure in affecting
fill height deviation depends on
Operating pressure in the filler B will be
significant in affecting fill height deviation if
57The significance of filler pressure in affecting
fill height deviation depends on
Operating pressure in the filler B will be
significant in affecting fill height deviation if
If a 0.01 then this becomes 64.059 gt 9.33, so
pressure does cause variations in filling height.
58Continuing to check the significance of
treatments produces the following outcomes (all
for a 0.01)
59- Continuing to check the significance of
treatments produces the following outcomes (all
for a 0.01) - Line Speed is significant as a treatment on fill
height. (So all three direct effectscarbonation
, pressure, and line speedare significant.
60- Continuing to check the significance of
treatments produces the following outcomes (all
for a 0.01) - Line Speed is significant as a treatment on fill
height. (So all three direct effectscarbonation
, pressure, and line speedare significant. - None of the interaction is significant at the a
0.01 level, though the Carbonation-Operating
Pressure would be if a were increased to 0.06.
(None of the other interactions becomes
significant unless a is increased to near 0.25.