Вы находитесь на странице: 1из 11

60902 The Scientific Method

NAME: Solution Student Number: FULL MARK 40

Lab 5 Nested and Split-plot Designs

Q1. How much variation is there in the routine analysis of fibre content of soya cotton cakes? The data
are presented by the lab that did the test and all the three technicians within the lab that did the analysis.
(Data from Clarke and Kempson)
To perform an appropriate analysis, you will need 3 columns, one for the responses (fibre), one indicating
the lab and one indicating the technician. When specifying your model, remember that technicians are
nested within labs (so you could number the technicians from 1 to 9 if you wanted).

Use Stat>ANOVA>General Linear Model> Fit General Linear Model


Responses: Q1fibre
Factors: Q1lab Q1technician
Random/Nest: Q1technician Nested in Q1lab
Graphs: Residuals versus fits
Storage: Residuals

Lab A B C
Technician 1 2 3 1 2 3 1 2 3
12.44 12.38 12.90 12.80 12.88 12.95 12.94 12.83 12.77
12.52 12.28 12.92 13.30 12.75 12.80 12.86 12.74 12.83
12.72 12.48 12.83 13.00 12.98 12.65 12.72 13.08 12.81
12.62 12.36 12.99 13.05 12.80 12.70 12.83 12.94 12.83

Now report on the results of the analysis.

i. Do the assumptions for ANOVA appear to have been satisfied?

The residuals appear to have constant variance. 1 mark

Spring 2016 Page 1 of 11


H0: The residuals are normally distributed H1: The residuals are not normally distributed
Test statistic: AD = 0.276, p-value = 0.631. 1 mark
As the p-value (0.631) > (0.05) we do not reject H0.
The residuals are normally distributed 1 mark

Yes, the assumptions for ANOVA appear to have been satisfied. 1 mark

ii. What were the results of the ANOVA table?

General Linear Model: Q1fibre versus Q1lab, Q1technician


Method
Factor coding (-1, 0, +1)

Factor Information
Factor Type Levels Values
Q1lab Fixed 3 1, 2, 3
Q1technician(Q1lab) Fixed 9 1(1), 2(1), 3(1), 1(2), 2(2),
3(2), 1(3), 2(3), 3(3)

Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Q1lab 2 0.5030 0.25148 17.92 0.000
Q1technician(Q1lab) 6 0.7461 0.12436 8.86 0.000
Error 27 0.3790 0.01404
Total 35 1.6281

Model Summary
S R-sq R-sq(adj) R-sq(pred)
0.118478 76.72% 69.82% 58.62%

Spring 2016 Page 2 of 11


Ans:
For the lab:
H0: The mean fibre content in each lab is the same
H1: The mean fibre content in at least one of the lab is different. 1 mark
Test statistic: F = 17.92 p-value: 0 1 mark
Decision: Since p-value (0) < (0.05), we reject H0. 1 mark
Conclusion: The mean fibre content in at least one of the lab is different. 1 mark

For the nested technician:


H0: The mean fibre content for each technician within a lab is the same
H1: In at least one lab, the mean fibre content for at least one technician is different. 1 mark
Test statistic: F=8.86 p-value: 0 1 mark
Decision: Since p-value (0) < (0.05), we reject H0. 1 mark
Conclusion: In at least one lab, the mean fibre content for at least one technician is different. 1 mark

Use Stat>ANOVA>General Linear Model> Comparison


Choose Q1lab and press C= Compare levels for this item

iii. After performing the Tukey comparison, report the final conclusion regarding the fibre content
as tested by the three different labs.

Tukey Pairwise Comparisons: Response = Q1fibre, Term = Q1lab


Grouping Information Using the Tukey Method and 95% Confidence

Q1lab N Mean Grouping


2 12 12.8883 A
3 12 12.8483 A
1 12 12.6200 B

Means that do not share a letter are significantly different.

Ans:
2 = 3 < 1. 1 mark
The mean fibre content reading of lab 1 is lower than the other two labs. 1 mark

Spring 2016 Page 3 of 11


Q2. Commercial respirators include filters to protect against particulate matter. There are two
manufacturers of such filters and three filters from each manufacturer. Each filter was tested three times.
The measurements are the precent penetration by a standard type of aerosol. (from Kuehl, including
data)

Manufacturer A B
Filter 1 2 3 1 2 3
1.12 0.16 0.15 0.91 0.66 2.17
1.10 0.11 0.12 0.83 0.83 1.52
1.12 0.26 0.12 0.95 0.61 1.58

Determine the appropriate linear model and perform the analysis similar to Q1 using Minitab.

i. Assume the residuals have constant variance and are normally distributed. What were the
results of the ANOVA table?

General Linear Model: Q2aerosol versus Q2manufacturer, Q2filter


Method
Factor coding (-1, 0, +1)

Factor Information
Factor Type Levels Values
Q2manufacturer Fixed 2 1, 2
Q2filter(Q2manufacturer) Fixed 6 1(1), 2(1), 3(1), 1(2), 2(2),
3(2)

Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Q2manufacturer 1 1.8689 1.86889 73.61 0.000
Q2filter(Q2manufacturer) 4 3.7413 0.93532 36.84 0.000
Error 12 0.3047 0.02539
Total 17 5.9148

Model Summary
S R-sq R-sq(adj) R-sq(pred)
0.159339 94.85% 92.70% 88.41%

Ans:
For the manufacturer:
H0: The mean percent penetration for each manufacturer is the same
H1: The mean percent penetration for at least one manufacturer is different 1 mark
Test statistic: F = 73.61 p-value: 0 1 mark
Decision: Since p-value (0) < (0.05), we reject H0. 1 mark
Conclusion: The mean fibre content in at least one of the lab is different. 1 mark

For the nested filter:


H0: The mean percent penetration for each filter within a manufacturer is the same
H1: In at least one manufacturer, the mean percent penetration for at least one filter is different. 1 mark
Test statistic: F=36.84 p-value: 0 1 mark
Decision: Since p-value (0) < (0.05), we reject H0. 1 mark
Conclusion: In at least one manufacturer, the mean percent penetration for at least one filter is different.
1 mark
Spring 2016 Page 4 of 11
ii. Perform the Tukey comparison and report the final conclusion regarding the mean penetration
of filters as tested by the two different manufacturers.

Tukey Pairwise Comparisons: Response = Q2aerosol, Term = Q2manufacturer


Grouping Information Using the Tukey Method and 95% Confidence

Q2manufacturer N Mean Grouping


2 9 1.11778 A
1 9 0.47333 B

Means that do not share a letter are significantly different.

Ans:
2 > 1. 1 mark
The mean precent penetration of manufacturer 1 filters is lower. 1 mark

iii. Do the assumptions for ANOVA appear to have been satisfied?

The residuals do not appear to have constant variance 1 mark

Spring 2016 Page 5 of 11


H0: The residuals are normally distributed. H1 The residuals are not normally distributed
Test stat: AD = 1.018, p-value = 0.008. 1 mark
Since p-value (0.008) < (0.05), we reject the null hypothesis.
The residuals are not normally distributed. 1 mark

No, the assumption of ANOVA are not satisfied. 1 mark

iv. As the assumption of normality is not satisfied, replace the responses by their ranks. Data>Rank,
Rank data in: Q2aerodol. You can create any label you like for Store ranks in: (this will be the
column heading of the calculated rank). Repeat the analysis on the ranks. Have your conclusions
from parts (i) and (ii) changed?

General Linear Model: Rank versus Q2manufacturer, Q2filter


Method
Factor coding (-1, 0, +1)

Factor Information
Factor Type Levels Values
Q2manufacturer Fixed 2 1, 2
Q2filter(Q2manufacturer) Fixed 6 1(1), 2(1), 3(1), 1(2), 2(2),
3(2)

Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Q2manufacturer 1 112.50 112.500 53.29 0.000
Q2filter(Q2manufacturer) 4 345.17 86.292 40.88 0.000
Error 12 25.33 2.111
Total 17 483.00

Spring 2016 Page 6 of 11


Model Summary
S R-sq R-sq(adj) R-sq(pred)
1.45297 94.76% 92.57% 88.20%

Tukey Pairwise Comparisons: Response = Rank, Term = Q2manufacturer


Grouping Information Using the Tukey Method and 95% Confidence

Q2manufacturer N Mean Grouping


2 9 12 A
1 9 7 B

Means that do not share a letter are significantly different.

Ans:
The results are similar to that of i and ii 1 mark
except all the tests are on Medians instead of means 1 mark

Spring 2016 Page 7 of 11


Q3. Montgomery (2007) describes an experiment to study pigment dispersion in paint. Four mixes of a
pigment are used. Each day all four mixes are prepared and then applied to three panels with each panel
receiving paint by one of three application methods (brush, roller, spray gun). The experiment is repeated
on a second day.

Mix
Day Method 1 2 3 4
1 64.5 66.3 74.1 66.5
1 2 68.3 69.5 73.8 70.0
3 70.3 73.1 78.0 72.3
1 65.2 65.0 73.8 64.8
2 2 69.2 70.3 74.5 68.3
3 71.2 72.8 79.1 71.5

Use Stat>ANOVA>General Linear Model> Fit General Linear Model


Responses: Q3dispersion
Factors: Q3day Q3mix Q3method
Graphs: Residuals versus fits
Storage: Residuals
Model:
select both Q3day and Q3mix and click Add,
select both Q3mix and Q3method and click Add.
select Q3day*Q3mix and click

Terms in the model:


Q3day
Q3mix
Q3day*Q3mix
Q3method
Q3mix* Q3method

Random/Nest: Q3day change Fixed to Random

If the interaction effect is significant, obtain the interaction plot:


Use Stat>ANOVA> Interaction Plot
Responses: Q3dispersion
Factors: Q3mix Q3method

Spring 2016 Page 8 of 11


Now report on the results of the analysis.

i. Do the assumptions for ANOVA appear to have been satisfied?

The residuals appear to have constant variance 1 mark

H0: The residuals are normally distributed. H1 The residuals are not normally distributed
Test stat: AD = 0.252, p-value = 0.707. 1 mark
Since p-value (0.707) > (0.05), we do not reject the null hypothesis.
The residuals are normally distributed. 1 mark

Yes, the assumption of ANOVA appear to be satisfied. 1 mark

Spring 2016 Page 9 of 11


ii. What were the results of the ANOVA table and the appropriate plot?

General Linear Model: Q3dispersion versus Q3day, Q3mix, Q3method


Method
Factor coding (-1, 0, +1)

Factor Information
Factor Type Levels Values
Q3day Random 2 1, 2
Q3mix Fixed 4 1, 2, 3, 4
Q3method Fixed 3 1, 2, 3

Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Q3day 1 0.042 0.0417 0.03 0.877
Q3mix 3 208.450 69.4833 47.14 0.005
Q3day*Q3mix 3 4.422 1.4739 6.18 0.018
Q3method 2 144.611 72.3054 303.38 0.000
Q3mix*Q3method 6 11.462 1.9104 8.02 0.005
Error 8 1.907 0.2383
Total 23 370.893

Model Summary

S R-sq R-sq(adj) R-sq(pred)


0.488194 99.49% 98.52% 95.37%

For the interaction:


H0: No interaction between mix and method
H1: Some interaction between mix and method 1 mark
Test statistic: F = 8.02 p-value: 0.005 1 mark
Decision: Since p-value (0.005) < (0.05), we reject H0. 1 mark
Conclusion: There is a significant interaction between mix and method. 1 mark

Since the interaction is significant, we cannot interpret the main effects.

Spring 2016 Page 10 of 11


Method 3 and mix 3 give the highest pigment dispersion. 1 mark
Method 1 and mix 1 give the lowest pigment dispersion. 1 mark

Spring 2016 Page 11 of 11

Вам также может понравиться