Tutorial 8

CH 6, Partially Replicated Full Factorial Design

Data set for analysis:

RUN Temp (T) Conc (C) Moisture (M)

1 -1 -1 -1 78 Regular 23 full factorial

2 +1 -1 -1 52

3 -1 +1 -1 56

4 +1 +1 -1 84

5 -1 -1 +1 62

6 +1 -1 +1 40

7 -1 +1 +1 58

8 +1 +1 +1 72

------------------------------------------------------------------------------------------------------------------------------------

9 0 0 0 52 Three independent replicates at centre point

10 0 0 0 48

11 0 0 0 50

Qu 1: What is a measure of reproducibility for this data set? Hint: If an observation is reproducible, it
should be able to be made by a different team repeating the experiment using the same experimental
data and methods, under the same operating conditions, in the same or a different location, on multiple
trials. Therefore, when we are looking to measure the reproducibility in fact we are looking for some
kind of variance.

Qu 2: Construct the ANOVA table

Qu 3: Compare the SSE of ANOVA table with SSE from the following equation. SSE = SSTOTAL – (sum of
all SS terms for factors and interactions). Are they the same? Why? And how would you solve that.
Revisit ANOVA and do the numerical reconciliation if needed.

Qu 4: What are the significant terms in the revisited ANOVA table? Why we detected LOF?
Solution:

To TA: Please take the time and write down the details on the board, the reason that we have one or
two questions per tutorial is to have the opportunity to explain each parts in details, of course in a pace
that students could ask their questions. If not, we could simply post the solution on UWLearn, similar to
the book solution manual that they might have. For example, here you see if final answers of ANOVA
plus the methodology, let’s be patient and write down the details of methods on the board. Don’t worry
about the time, if needed we can continue on the following tutorial. It’s better to cover one questions in
details, rather than a few ones but writing down something like solution manual for them.

Detailed Solutions:

(To TA: Draw coded table on board, at least a few rows)

T C M TC TM CM TCM Response

1 - - - + + + - 78

T + - - - - + + 52

C - + - - + - + 56

TC + + - + - - - 84

M - - + + - - + 62

TM + - + - + - - 40

CM - + + - - + - 58

TCM + + + + + + + 72

k = 3, 23 full factorial design, but R = 1 (not fully replicated, look at formula cheat sheet handed in the
class)

Total # of runs = 8 (from full factorial) + 3 (from centre point replicates) = 11

divisor = R 2(k-1)

(eff)T or effT = (contrast)T / (divisor)

Mean square = SSeff / df = R 2k-2 (eff)2 / df

(To TA: Please, calculate effect of C, T, C by T for students with the above mentioned procedure on
board, for example make sure to calculate one of the contrasts at least, instead of writing down just (-6),
the same for SS and MSE. If needed extra formula, look at formula sheet and add the equation)

(eff of T) = (-6) / [ 1 2(3-1) ] = -1.5

(eff of C) = 9.5
(eff of T by C interaction) = 22.5

SS of T = 1 2(3-2) (-1.5)2 = 4.5

SS of (T by C by M 3-factor interaction) = 40.5

Qu 1: What is a measure of reproducibility for this data set?

Based on the three centre points, calculate their variance. This will be s2 = MSE = 4, with dfE = (3-1) = 2

Qu 2: Construct the ANOVA table

Factor Effect SS df MS Fobs (=MS/df)/F probe/F 0

T -1.5 4.5 1 4.5 1

C 9.5 180.5 1 180.5 45*

M -9.5 180.5 1 180.5 45*

TxC 22.5 1,012.5 1 1,102.5 253*

TxM -2.5 12.5 1 12.5 3

CxM 4.5 40.5 1 40.5 10

TxCxM -4.5 40.5 1 40.5 10

Error 8 2 4

TOTAL 1,834.182 10

Some remarks on the ANOVA entries above:

SSTOTAL = (Sum of the 11 Y’s squared) – (gt2 / N) N= 8 + 3 = 11

Hence: SSTOTAL = 40,480 – [(6522)/ 11] = 1,834.182 df TOTAL = 11 – 1 =10

(To TA: from many different calculation options of MSE, as shown in tutorial 6, please review at least
one of the options for student and calculate MSE on board)

F crit (alpha = 5%; df (1, 2)) = 18.5

(Incidentally, how did you estimate the SSE above? Any issues that make you suspect that something is
not right in the ANOVA ‘balance’ above?)

SSTOTAL should be equal to the sum SS terms above it, in other words:
SSE = SSTOTAL – (sum of all SS terms for factors and interactions) = 1,834.182 – [4.5 + 180.5+ 180.5 + …+
40.5] = 362.682

But this is different from 8!!! The ANOVA ‘balance’ is not reconciled!

Qu 3: The same? Why? Solution?

We know that SSE = SSPE + SSLOF

SSPE comes from the independent replicates.

Therefore, 362.682 = 8 + SSLOF, hence SSLOF = 354.682

Solution?

Revisit ANOVA and do the numerical reconciliation. See the new entries and breakdown; compare with
the initial ANOVA.

Factor Effect SS df MS Fobs (=MS/df)/F probe/F 0

T -1.5 4.5 1 4.5 1

C 9.5 180.5 1 180.5 45*

M -9.5 180.5 1 180.5 45*

TxC 22.5 1,012.5 1 1,102.5 253*

TxM -2.5 12.5 1 12.5 3

CxM 4.5 40.5 1 40.5 10

TxCxM -4.5 40.5 1 40.5 10

LOF 354.682 1 354.682 88*

PE 8 2 4

Error (PE + LOF) 362.682 3

TOTAL 1,834.182 10

Qu 4: What are the significant terms in the revisited ANOVA table? Why we detected LOF?

Significant terms in ANOVA: C, M, T x C, and LOF!

We detected LOF only because we had independent replicates!

To TA: here remind students with the meaning of LOF which is?
Linear model is not adequate; curvature effects (detection); quadratic terms in model; start with
inclusion of quadratic terms for main factors; (investigate)

For instance, for a two-factor case (X1 and X2), one can write

Y = β 0 + β1 X1 + β2 X2 + β1 2 X1 X2 + β11 X12 + β22 X22 + ε

(You can do analogously for 3 factors, etc.)

To TA: Tell students, here is what would be next in industrial case studies: (Curvature and LOF lead
naturally into studies of response surface methodology (RSM), central composite designs (CCDs) and
process optimization)