Sie sind auf Seite 1von 12

ANOVA and General Factorials Solutions

#1

a.

Boxplot of y vs code
12 11 10 9
y

8 7 6 5 1 2 code 3 4

There seems to be more variability in the observations from code 2. We will probably see a difference in means because of code 1.

b.

H 0 : 1 2 3 4 H A : i m j for at least onei j

c.

Analysis of Variance for y, using Adjusted SS for Tests

Source code Error Total

DF 3 36 39

Seq SS 26.803 56.471 83.274

Adj SS 26.803 56.471

Adj MS 8.934 1.569

F 5.70

P 0.003

Since the p-value is less than alpha, 0.003 0.05 , we reject the null hypothesis and conclude at least one pair of means is different, which is what we expected to see.

d.

Normal Probability Plot


(response is data) 99 3.0

Versus Fits
(response is data)

90

1.5

Percent

Residual

50

0.0

10

-1.5

-3.0 -3.0 -1.5 0.0 Residual 1.5 3.0 7.0 7.5 8.0 8.5 Fitted Value 9.0

Residuals Versus code


(response is data) 3.0

1.5

Residual

0.0

-1.5

-3.0 1 2 code 3 4

The residuals vs. fits and code show no patterns. The normal probability plot shows that all of the points fall close to the lines. These things mean that the assumptions are met.

e.

Tukey Simultaneous Tests Response Variable data All Pairwise Comparisons among Levels of code code = 1 subtracted from: code 2 3 4 Difference of Means 2.130 1.830 1.490 SE of Difference 0.5601 0.5601 0.5601 T-Value 3.803 3.267 2.660 Adjusted P-Value 0.0029 0.0122 0.0539

code = 2 code 3 4

subtracted from: SE of Difference 0.5601 0.5601 T-Value -0.536 -1.143 Adjusted P-Value 0.9498 0.6660

Difference of Means -0.3000 -0.6400

code = 3 code 4

subtracted from: SE of Difference 0.5601 T-Value -0.6070 Adjusted P-Value 0.9292

Difference of Means -0.3400

Based on the above results, 1 seems to be significantly different than 2 and 3, and marginally significantly different than 4.

#2

a)

Boxplot of yield
91.0 90.8 90.6
yield

90.4 90.2 90.0

temp pressure

150

160 200

170

150

160 215

170

150

160 230

170

Based on the boxplots it seems that the relationship for temp is the same at each level of pressure. Also, it seems that there is more variation in the response for the levels of temperature when the pressure is 230. Overall it looks like there is indeed something going for both factors.

b)

Analysis of Variance for yield, using Adjusted SS for Tests Source pressure temp pressure*temp Error Total DF 2 2 4 9 17 Adj SS 0.76778 0.30111 0.06889 0.16000 1.29778 Adj MS 0.38389 0.15056 0.01722 0.01778 F 21.59 8.47 0.97 P 0.000 0.009 0.470

Based on the above, we can remove the interaction. Now the reduced model becomes:

Analysis of Variance for yield, using Adjusted SS for Tests Source pressure temp Error Total DF 2 2 13 17 Seq SS 0.76778 0.30111 0.22889 1.29778 Adj SS 0.76778 0.30111 0.22889 Adj MS 0.38389 0.15056 0.01761 F 21.80 8.55 P 0.000 0.004

Both factors are significant, so this is our final model.

c) Now lets check assumptions:


Normal Probability Plot
(response is yield) 99 0.2

Versus Fits
(response is yield)

90

0.1
Residual

Percent

50

0.0

-0.1 10 -0.2 1

-0.30

-0.15

0.00 Residual

0.15

0.30

90.0

90.2

90.4 Fitted Value

90.6

90.8

Residuals Versus pressure


(response is yield) 0.2 0.2

Residuals Versus temp


(response is yield)

0.1
Residual Residual

0.1

0.0

0.0

-0.1

-0.1

-0.2

-0.2

200

210 pressure

220

230

150

155

160 temp

165

170

Based on these plots the assumptions are met.

d) Now lets look at the main effect plots. Recall that the response is yield, which we are trying to maximize:

Main Effects Plot for yield


Data Means pressure temp

90.7 90.6 90.5 90.4 90.3 90.2 200

Mean

215

230

150

160

170

Based on the plots above, we should set pressure to 215 and temperature to 170 to maximize the yield.

#3 a)

Boxplot of strength
201.0 201.0

Boxplot of strength

199.5
strength strength

199.5

198.0

198.0

196.5

196.5

195.0 3 4

195.0 2 4 8

Cooking Time Boxplot of strength


201.0

%Conc

199.5
strength

198.0

196.5

195.0 400 500 650

Pressure

Based on this, it does seem that each factor has some effect on strength.

b)

Analysis of Variance for strength, using Adjusted SS for Tests Source DF %Conc 2 Pressure 2 Cooking Time 1 %Conc*Pres 4 %Conc*Cook 2 Press*Cook 2 3-way 4 Error 18 Total 35 Adj SS 7.7639 19.3739 20.2500 6.0911 2.0817 2.1950 1.9733 6.5800 66.3089 Adj MS 3.8819 9.6869 20.2500 1.5228 1.0408 1.0975 0.4933 0.3656 F 10.62 26.50 55.40 4.17 2.85 3.00 1.35 P 0.001 0.000 0.000 0.015 0.084 0.075 0.290

Based on the above, we reduce the model:

Analysis of Variance for strength, using Adjusted SS for Tests


Source %Conc Pressure Cooking Time %Conc*Pressure Error Total DF 2 2 1 4 26 35 Adj SS 7.7639 19.3739 20.2500 6.0911 12.8300 66.3089 Adj MS 3.8819 9.6869 20.2500 1.5228 0.4935 F 7.87 19.63 41.04 3.09 P 0.002 0.000 0.000 0.033

Each of the terms in the reduced model is significant.

d) Lets check assumptions:

Residual Plots for strength


Normal Probability Plot
99 90
Residual Percent

Versus Fits
1 0 -1 -2

50 10 1 -2 -1 0 Residual 1

196

198 Fitted Value

200

Histogram
10.0
Frequency Residual

Versus Order
1 0 -1 -2

7.5 5.0 2.5 0.0 -1.6 -0.8 0.0 Residual 0.8 1.6

10

15 20 25 Observation Order

30

35

Based on the normality plot, we may have a potential outlier, but it seems to be still ok. The same goes for the fits plot.
Residuals Versus Cooking Time
(response is strength)

Residuals Versus %Conc


(response is strength)

Residual

Residual

-1

-1

-2 3.00 3.25 3.50 Cooking Time 3.75 4.00

-2 2 4 %Conc 6 8

Residuals Versus Pressure


(response is strength)

Residual

-1

-2 400 500 Pressure 600

These seem to be fine as well. If you are not convinced, you can do the tests with the residuals.

e)

Main Effects Plot for strength


Data Means 199.0

198.5
Mean

198.0

197.5

3 Cooking Time

Interaction Plot for strength


Data Means 200.5 200.0 199.5 199.0
Mean
%Conc 2 4 8

198.5 198.0 197.5 197.0 196.5 400 500 Pressure 650

Based on the main effect plot, we should set cooking time to 4 hrs. Based on the interaction plot, we should set percent concentration of hardwood to 2 and pressure to 650.

Das könnte Ihnen auch gefallen