Professional Documents
Culture Documents
CHAPTER 6
Note to Instructor: For computer exercises, the procedure Regression under Stat in Minitab can be used for the
regression analysis except for computing confidence intervals on the regressor variables.
Sections 6-2
Analysis of Variance
Source DF SS MS F P
Regression 1 0.000094464 0.000094464 275.84 0.000
Residual Error 4 0.000001370 0.000000342
Total 5 0.000095833
y 0.0249 0.129x
b) 0.0005747
-0.0007088
0.0001486
-0.0004799
-0.0000644
0.0005298
e) SST = 0.000095833
SSR = 0.000094464, SSE = 0.000001370, and SSR + SSE = 0.000095834
Therefore, SST = SSR + SSE
f) R2 = 98.6%. This is interpreted as 98.6% of the total variability in thermal conductivity can be explained by the
fitted regression model.
h) See the Minitab output in part a). Based on the analysis of variance, we can reject the null hypothesis and conclude
that the regression is significant because p-value is 0.000 less than = 0.05.
1
Engineering Statistics 5th edition October 01, 2010
0.0005
Residual
0.0000
-0.0005
Density
0.0005
Residual
0.0000
-0.0005
Fitted Value
2
Engineering Statistics 5th edition October 01, 2010
1
Normal Score
-1
Residual
k) r = 0.993, P-value = 0. Therefore, there is a significant correlation between density and conductivity.
Because the slope is significantly different from zero based on the conclusions from parts (g) and (h), the correlation
coefficient is also significantly different from zero.
6-2. a)
y 8.51 9.25x
3
Engineering Statistics 5th edition October 01, 2010
b)
c) SSE = 84 V 2 = 8
e) SST = 252383
SSR = 252299, SSE = 38, and SSR + SSE = 280621
SST = SSR + SSE
f) R2 = 100%. This is interpreted as 100% of the total variability in Usage can be explained by the fitted regression
model.
h) See the Minitab output in part a). Based on the analysis of variance, we can reject the null hypothesis and conclude
that the regression is significant because The P-value = 0.000 < 0.05.
4
Engineering Statistics 5th edition October 01, 2010
5
Engineering Statistics 5th edition October 01, 2010
k) r = 1.000, P-value = 0, Therefore, we conclude there is a significant correlation between temperature and usage.
Because the slope is significantly different from zero based on the conclusions from parts (g) and (h), the correlation
coefficient is also significantly different from zero.
Analysis of Variance
Source DF SS MS F P
Regression 1 0.0013727 0.0013727 32.76 0.000
Residual Error 18 0.0007542 0.0000419
Total 19 0.0021270
y 0.393 0.00333 x
6
Engineering Statistics 5th edition October 01, 2010
b) -0.0054488
0.0072519
0.0065614
-0.0127685
0.0069249
-0.0004269
-0.0027627
-0.0044372
0.0002431
-0.0074196
0.0015643
0.0078738
0.0035833
-0.0077656
-0.0011131
-0.0024386
0.0105570
-0.0054342
-0.0014357
0.0068913
c) SSE = 0.0007542 V 2 = 0.0000419
d) se( E1 ) = 0.0005815, se( E 0 ) = 0.04258
e) SST = 0.0020402
SSR = 0.0013149, SSE = 0.0007253, and SSR + SSE = 0.0020402
SST = SSR + SSE
f) R2 = 64.5%. This is interpreted as 64.5% of the total variability in deflection can be explained by the fitted
egression model.
g) See the Minitab output given in part a).
Based on the t-tests, we conclude that the slope and intercept are nonzero.
Based on P-values, both the intercept and slope have P-values = 0.000 < = 0.05. We conclude that the intercept and
slope are significantly different from zero.
h) See the Minitab output in part a). Based on the analysis of variance, we can reject the null hypothesis and conclude
that the regression is significant because the p-value is 0.000 less than = 0.05.
i) E0: 0.39346 r 2.101(0.04258); 0.304, 0.483
E1: 0.0033285 r 2.101(0.0005815); 0.00211, 0.00455
Zeros are not included in the CIs, so both intercept and slope are significantly different from zero. The conclusions
from part (g), (h) and (i) are the same.
j) Residual plots appear reasonable, so the model provides an adequate fit.
7
Engineering Statistics 5th edition October 01, 2010
0.01
Residual
0.00
-0.01
68 73 78
Temp
0.01
Residual
0.00
-0.01
Fitted Value
8
Engineering Statistics 5th edition October 01, 2010
1
Normal Score
-1
-2
-0.01 0.00 0.01
Residual
k) r = 0.803, P-value = 0, Therefore, we conclude there is a significant correlation between temperature and deflection.
Because the slope is significantly different from zero based on the conclusions from parts (g) and (h), the correlation
coefficient is also significantly different from zero.
6-4. a)
9
Engineering Statistics 5th edition October 01, 2010
b)
j) The normal probability plot of residuals appears reasonable. The plots of residuals against yi and xi seem to have
funnel pattern, so the model does not provide an adequate fit.
10
Engineering Statistics 5th edition October 01, 2010
k) r = 0.607, P-value = 0.021. Based on this test there is a significant correlation between temperature and turbidity.
However, the residual plots indicate the model might not be valid and then these tests are invalid. Because the slope
is significantly different from zero based on the conclusions from parts (g) and (h), the correlation coefficient is also
significantly different from zero.
11
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 1 90.759 90.759 84.28 0.000
Residual Error 13 13.999 1.077
Total 14 104.757
y 40.6 2.12 x
b) -0.97179
0.00066
1.56517
0.35430
0.21735
0.99417
0.79921
0.83762
0.24199
-1.22252
-0.49351
-0.38411
-0.74715
-2.15947
0.96808
12
Engineering Statistics 5th edition October 01, 2010
1
Residual
-1
-2
1 2 3 4 5
strength
1
Residual
-1
-2
29 30 31 32 33 34 35 36 37 38
Fitted Value
13
Engineering Statistics 5th edition October 01, 2010
1
Normal Score
-1
-2
-2 -1 0 1 2
Residual
k) r = -0.931, P-value = 0; Therefore, we conclude there is a significant correlation between strength and
temperature. Because the slope is significantly different from zero based on the conclusions from parts (g)
and (h), the correlation coefficient is also significantly different from zero.
6-6. a) The plot below implies that a simple linear regression seems reasonable in this situation.
14
Engineering Statistics 5th edition October 01, 2010
9
8
7
6
5
y
4
3
2
1
0
60 70 80 90 100
x
Analysis of Variance
Source DF SS MS F P
Regression 1 92.934 92.934 53.50 0.000
Residual Error 18 31.266 1.737
Total 19 124.200
An estimate of V 2 = 1.737
c) y 10.1 0.174(85) 4.69. The predicted mean rise in blood pressure level associated with a sound pressure level
of 85 decibels is 4.69 millimeters of mercury.
6-7. a) 0.055215
b) (0.05304, 0.05738)
c) (0.052505, 0.057925)
d) The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
6-8. a) 472.499
b) (471.183, 473.816)
c) (467.975, 477.024)
d) The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
15
Engineering Statistics 5th edition October 01, 2010
c) (0.6256, 0.6537)
d) The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
6-11. a) 36.095
b) (35.059, 37.131)
c) (32.802, 39.388)
d) The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
6-12. a) 4.683
b) (4.055, 5.312)
c) (1.844, 7.523)
d) The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
6-13. a)
The regression equation is
Deflection = 32.0 - 0.277 Stress level
Analysis of Variance
Source DF SS MS F P
Regression 1 45.154 45.154 40.38 0.000
Residual Error 7 7.827 1.118
Total 8 52.981
V 2 1.118
75
70
Stress level
65
60
55
50
11 12 13 14 15 16 17 18 19
Deflection
16
Engineering Statistics 5th edition October 01, 2010
1
d) 3.61
0.277
e) y 32.05 0.277(78) 10.444 . There are no residuals because there are no observations at x = 78%
c) 0.178(4) = 0.712
e) Fitted yi :
0.83585
1.01391
1.37002
1.72613
2.08225
2.43836
2.79447
3.15058
3.50670
3.86281
4.21892
Scatterplot of y-hat vs y
4.5
4.0
3.5
3.0
y-hat
2.5
2.0
1.5
1.0
17
Engineering Statistics 5th edition October 01, 2010
All the points would lie along the 45 degree line y y . That is, the regression model would estimate the values
exactly. At this point, the graph of observed vs. predicted indicates that the simple linear regression model provides a
reasonable fit to the data.
Analysis of Variance
Source DF SS MS F P
Regression 1 3.6631 3.6631 94.37 0.000
Residual Error 12 0.4658 0.0388
Total 13 4.1289
Coef 0.83075
Tx 9.7141
SE Coef 0.08552
P-valuex = 2*P(t > |9.71|): for degrees of freedom of 12 we obtain 2*(P-value < 0.0005) = P-value < 0.001
SS /(n p) 0.4658 / 12
2
R Adjusted 1 E 1 87.78%
SS T /(n 1) 4.1289 / 13
SS Total SS Re g SS E 3.6631 0.4658 4.1289
SS Error 0.4658
MS Error 0.0388
DFError 12
S MS Error 0.0388 0.197
MS Re g 3.6631
F 94.41
MS Error 0.0388
P valueregression 2 P( F1,12 ! 94.41) 2( P value 0.01) P value 0.02
b) V 2 0.0388
c) Based on the P-values from the F-test in the ANOVA table and the t-test for X in the output in part (a), 1 is
significantly different from zero. There P-values are always the same for simple linear regression.
d)
E1 t 0.025,12 se( E1 ) d E 1 d E1 t 0.025,12 se( E1 )
0.83075 2.179(0.08552) d E 1 d 0.83075 2.179(0.08552)
0.83075 0.18635 d E 1 d 0.83075 0.18635
0.6444 d E 1 d 1.0171
Because zero is not included in the 95%CI, the estimated coefficient (1) is significantly different from zero.
e) The results from part (c) and (d) are the same whenever the confidence level = 1 D.
f) y 0.6649 0.83075 x
y 0.6649 0.83075(2.18) 2.476
e y y 2.8 2.476 0.324
g) P Y 1.5 0.6649 0.83075(1.5) 1.911
1 ( x0 x ) 2 1 (1.5 1.76) 2
se( P Y 1.5 ) V 2 0.0388 0.057
n S xx 14 5.326191
95%CI
18
Engineering Statistics 5th edition October 01, 2010
1.787 d P Y x d 2.035 0
95%PI
1 ( x0 x ) 2 1 (1.5 1.76) 2
V 2 1 0.03881 0.205
n S xx 14 5.326191
1 (x x)2 1 ( x0 x ) 2
y 0 t 0.025,12 V 2 1 0 d Y0 d y 0 t 0.025,12 V 1
2
n S xx n S xx
1.911 2.179(0.205) d Y0 d 1.911 2.179(0.205)
1.911 0.447 d Y0 d 1.911 0.447
1.464 d Y0 d 2.358
The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
Analysis of Variance
Source DF SS MS F P
Regression 1 84.106 84.106 210.63 0.000
Residual Error 14 5.590 0.3993
Total 15 89.696
Coef 8.3088
Tx 14.5132
SE Coef 0.5725
P-valuex = 2*P(t > |-14.51|): for degrees of freedom of 14 we obtain P-value < 2(0.0005) = P-value < 0.001
SS /(n p) 5.590 / 14
2
R Adjusted 1 E 1 93.32%
SS T /(n 1) 89.696 / 15
SS Total SS Re g SS E 84.106 5.590 89.696
SS Error 5.590
MS Error 0.3993
DFError 14
S MS Error 0.3993 0.632
MS Re g 84.106
F 210.634
MS Error 0.3993
P value 2 P( F1,14 ! 210.63) 2( P value 0.01) P value 0.02
b) V 2 0.3993
c) Based on the P-values from the F-test in the ANOVA table and the t-test for X in the output in part (a), 1 is
significantly different from zero. There P-values are always the same for simple linear regression.
d)
19
Engineering Statistics 5th edition October 01, 2010
1 ( x0 x ) 2 1 (0.6 0.52) 2
se( P Y 0.6 ) V 2 0.3993 0.164
n S xx 16 1.218294
95%CI
P Y x t 0.025,14 se( P Y x ) d P Y x d P Y x t 0.025,14 se( P Y x )
0 0 0 0 0
4.362 d P Y x d 3.658 0
95%PI
1 ( x0 x ) 2 1 (0.6 0.52) 2
V 2 1 0.39931 0.653
n S xx 16 1.218294
1 (x x)2 1 ( x0 x ) 2
y 0 t 0.025,14 V 2 1 0 d Y0 d y 0 t 0.025,14 V 1
2
n S xx n S xx
4.01 2.145(0.653) d Y0 d 4.01 2.145(0.653)
4.01 1.401 d Y0 d 4.01 1.401
5.411 d Y0 d 2.609
The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
Section 6-3
Analysis of Variance
Source DF SS MS F P
Regression 2 12161.6 6080.8 9.35 0.051
Residual Error 3 1950.4 650.1
Total 5 14112.0
Source DF Seq SS
X1 1 10240.4
x2 1 1921.2
20
Engineering Statistics 5th edition October 01, 2010
b) -24.9866
24.3075
11.8203
-20.4595
12.8296
-3.5113
c) SSE = 1950.4 V 2 = 650.1
d) R-Sq = 86.2%, R-Sq(adj) = 77.0%; R-Sq(adj) is less than R-Sq because the model contains terms that are not
contributing significantly to the model. The adjusted R2 value will penalize the user for adding terms to the model that
are not significant.
e) See part a). Based on the P-value from the ANOVA table, the regression model is significant at the 0.10 level of
significance.
f) se( E 0 ) = 74.75, se( E1 ) = 1.169, se( E 2 ) = 0.08953
g) See part a). Based on the P-values for each coefficient, each regressor is not significantly different from zero at the
0.05 level of significance.
Residuals Versus x1
(response is y)
30
20
10
Residual
-10
-20
-30
0 10 20 30 40
x1
21
Engineering Statistics 5th edition October 01, 2010
30
20
10
Residual
-10
-20
-30
100 150 200
Fitted Value
1
Normal Score
-1
Residual
22
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 2 423.41 211.70 46.44 0.000
Residual Error 9 41.03 4.56
Total 11 464.44
Source DF Seq SS
Weight-x 1 236.70
Horsepow 1 186.71
b) 0.16464
-1.41661
2.33925
-1.31445
1.58629
-1.08273
1.12759
-3.77526
1.79269
0.63652
-2.14613
2.08818
c) SSE = 41.03 V 2 = 4.56
d) R-Sq = 91.2%, R-Sq(adj) = 89.2%. R-Sq(adj) is slightly less than R-Sq. The adjusted R2 value will penalize the user
for adding terms to the model that are not significant.
e) See part a). Based on the P-value from the ANOVA table, the regression model is significant at the 0.10 level of
significance.
E E E
f) se( 0 ) = 3.719, se( 1 ) = 0.0013, se( 2 ) = 0.0063
g) See part a). Based on the P-values for each coefficient, only x1 is not significantly different from zero at the 0.05
level of significance.
23
Engineering Statistics 5th edition October 01, 2010
Residual
0
-1
-2
-3
-4
100 200 300 400 500 600
Horsepow
1
Residual
-1
-2
-3
-4
2500 3500 4500
Weight-x
1
Residual
-1
-2
-3
-4
2 4 6 8 10 12
Observation Order
24
Engineering Statistics 5th edition October 01, 2010
Residual
0
-1
-2
-3
-4
10 20 30
Fitted Value
1
Normal Score
-1
-2
-4 -3 -2 -1 0 1 2 3
Residual
25
Engineering Statistics 5th edition October 01, 2010
b)
26
Engineering Statistics 5th edition October 01, 2010
27
Engineering Statistics 5th edition October 01, 2010
d) R-Sq = 71.2%, R-Sq(adj) = 65.5%. R-Sq(adj) is less than R-Sq because the regression equation contains terms that
are not contributing significantly to the model. The adjusted R2 value will penalize the user for adding terms to the
model that are not significant.
e) See part a). Based on the P-value from the ANOVA table, the regression model is significant at the 0.05 level of
significance.
f) se(E 0 ) 10.63 , se(E1 ) 0.2336 , se(E 2 ) 0.2185 , se(E 3 ) 5.55 , and se(E 4 ) 1.842
g) See part a). Based on the P-values for each coefficient, only Age is significantly different from zero at
the 0.05 level of significance.
i)
SRES COOK
28
Engineering Statistics 5th edition October 01, 2010
29
Engineering Statistics 5th edition October 01, 2010
30
Engineering Statistics 5th edition October 01, 2010
31
Engineering Statistics 5th edition October 01, 2010
j) The VIFs are all less than 10. There is no indication of a problem with multicollinearity.
32
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 2 0.23563 0.11782 1509.64 0.000
Residual Error 8 0.00062 0.00008
Total 10 0.23626
Source DF Seq SS
Cont 1 0.23562
Loss 1 0.00001
b)
-0.0089354
-0.0090847
-0.0030180
0.0025153
0.0074740
0.0087559
0.0079298
0.0098885
0.0008422
-0.0051274
-0.0112404
e) See part a). Based on the P-value from the ANOVA table, the regression model is significant at the 0.05 level of
significance.
g) See part a). Based on the P-values for each coefficient, only Cont is significantly different from zero at the 0.05
level of significance.
i)
SRES COOK
33
Engineering Statistics 5th edition October 01, 2010
-1.23832 0.255007
-1.44997 0.692448
-0.37215 0.008618
0.32815 0.011784
0.92827 0.058551
1.08437 0.077203
1.02796 0.109710
1.35664 0.287682
0.11001 0.001337
-0.67570 0.054084
-1.59530 0.485253
0.010
0.005
Residual
0.000
-0.005
-0.010
0.010
0.005
Residual
0.000
-0.005
-0.010
34
Engineering Statistics 5th edition October 01, 2010
0.010
0.005
Residual
0.000
-0.005
-0.010
0.010
0.005
Residual
0.000
-0.005
-0.010
1 2 3 4 5 6 7 8 9 10 11
Observation Order
95
90
80
70
Percent
60
50
40
30
20
10
5
1
-0.02 -0.01 0.00 0.01 0.02
Residual
35
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 4 4957.2 1239.3 5.11 0.030
Residual Error 7 1699.0 242.7
Total 11 6656.3
Source DF Seq SS
X1 1 3758.9
X2 1 1109.4
X3 1 88.9
X4 1 0.1
d) R-Sq = 74.5%, R-Sq(adj) = 59.9%; R-Sq(adj) is less than R-Sq because the regression equation contains terms that are
not contributing significantly to the model. The adjusted R2 value will penalize the user for adding terms to the model that
are not significant.
e) See part a). Based on the P-value from the ANOVA table, the regression model is significant at the 0.05 level of
significance.
f) se( E 0 ) = 207.9, se( E1 ) = 0.3689, se( E 2 ) = 5.301, se( E 3 ) = 2.392, se( E 4 ) = 0.7338
g) See part a). Based on the P-values for each coefficient, the regressors are not significant.
36
Engineering Statistics 5th edition October 01, 2010
i)
SRES COOK
-1.62309 0.430586
0.29990 0.092382
2.08556 0.820139
-0.83969 0.159823
0.67773 0.029524
0.72611 0.200284
0.36035 0.009353
-0.01285 0.000021
-1.67866 0.646149
-0.93670 0.047782
0.47341 0.010787
0.44599 0.010213
Residuals Versus X1
(response is Y)
20
10
Residual
-10
-20
20 30 40 50 60 70 80 90
X1
Residuals Versus X2
(response is Y)
20
10
Residual
-10
-20
21 22 23 24 25 26
X2
37
Engineering Statistics 5th edition October 01, 2010
Residuals Versus X3
(response is Y)
20
10
Residual
-10
-20
85 86 87 88 89 90 91 92 93 94
X3
Residuals Versus X4
(response is Y)
20
10
Residual
-10
-20
90 95 100 105 110
X4
20
10
Residual
-10
-20
230 240 250 260 270 280 290 300 310
Fitted Value
38
Engineering Statistics 5th edition October 01, 2010
20
10
Residual
-10
-20
1 2 3 4 5 6 7 8 9 10 11 12
Observation Order
95
90
80
70
Percent
60
50
40
30
20
10
5
1
-30 -20 -10 0 10 20 30
Residual
j) The VIFs are all less than 10. There is no indication of a problem with multicollinearity.
Analysis of Variance
Source DF SS MS F P
Regression 3 30532 10177 840.55 0.000
Residual Error 16 194 12
Total 19 30725
39
Engineering Statistics 5th edition October 01, 2010
Source DF Seq SS
Emitter 1 23959
Base 1 4233
EtoB 1 2340
b) -0.90039
1.83266
-0.31872
-6.78384
-2.18117
-1.51602
1.90876
2.29305
2.01911
-5.96711
-2.21540
3.41999
3.16536
-0.57066
-1.96916
2.64163
-0.93420
6.68822
1.14227
-1.75438
c) SSE = 194 V 2 = 12
e) See part a). Based on the P-value from the ANOVA table, the regression model is significant at the 0.10 level of
significance.
E
f) se( 0 ) = 49.58, se( E E ) = 3.692, se( E B ) = 0.0039, se( E EtoB ) = 0.2239
g) See part a). Based on the P-values for each coefficient, Emitter and EtoB are significantly different from zero at the
0.05 level of significance.
i) SRES COOK
-0.27777 0.002938
0.59321 0.023627
-0.10665 0.001012
-2.08750 0.159577
-0.67100 0.016418
-0.45280 0.004106
0.63705 0.035375
0.68266 0.008518
0.67761 0.041745
-1.77388 0.055072
-0.68134 0.016855
1.12620 0.099229
0.93545 0.012570
-0.17625 0.001204
-0.59731 0.010172
0.86378 0.054946
-0.40464 0.052052
40
Engineering Statistics 5th edition October 01, 2010
2.16785 0.319631
0.54024 0.124644
-0.53676 0.009606
Residual
-5
3 4 5 6 7 8 9 10 11
EtoB
5
Residual
-5
Base
41
Engineering Statistics 5th edition October 01, 2010
Residual
0
-5
14 15 16
Emitter
5
Residual
-5
2 4 6 8 10 12 14 16 18 20
Observation Order
5
Residual
-5
Fitted Value
42
Engineering Statistics 5th edition October 01, 2010
Normal Score
0
-1
-2
-5 0 5
Residual
j) All the VIFs are less than 10. There is no indication of a problem with multicollinearity.
6-23. a) 128.1
b) (49.0, 207.3)
c) (40.5, 296.8)
d) The prediction interval is wider than the confidence interval. The prediction interval is wider than the confidence
interval because it predicts a range for a future observation whereas the confidence interval predicts a range for the
mean response.
6-24. a) 27.398
b) (25.226, 29.570)
c) (22.091, 32.705)
d) The prediction interval is wider than the confidence interval. The prediction interval is wider than the confidence
interval because it predicts a range for a future observation whereas the confidence interval predicts a range for the
mean response.
6-25. a) 98.35
b) (87.99, 108.71)
c) (76.69, 120.02)
d) The prediction interval is wider than the confidence interval. The prediction interval is wider than the confidence
interval because it predicts a range for a future observation whereas the confidence interval predicts a range for the
mean response.
6-26. a) 0.97068
b) (0.95909, 0.98226)
c) (0.94724, 0.99411)
d) The prediction interval is wider than the confidence interval. The prediction interval is wider than the confidence
interval because it predicts a range for a future observation whereas the confidence interval predicts a range for the
mean response.
6-27. a) 287.56
b) (263.77, 311.35)
c) (243.69, 331.44)
d) The prediction interval is wider than the confidence interval. The prediction interval is wider than the confidence
interval because it predicts a range for a future observation whereas the confidence interval predicts a range for the
mean response.
6-28. a) 91.424.
b) (85.953, 96.895)
c) (83.249, 99.599)
43
Engineering Statistics 5th edition October 01, 2010
d) The prediction interval is wider than the confidence interval. The prediction interval is wider than the confidence
interval because it predicts a range for a future observation whereas the confidence interval predicts a range for the
mean response.
Analysis of Variance
Source DF SS MS F P
Regression 2 24518 12259 9.28 0.015
Residual Error 6 7928 1321
Total 8 32446
c) Based on the P-values from the t-test for each coefficient, the estimated coefficient of brightness (1) is not
significantly different from zero while contrast (2) is significant at the 0.05 level of significance.
Analysis of Variance
44
Engineering Statistics 5th edition October 01, 2010
Source DF SS MS F P
Regression 2 842.37 421.18 44.45 0.000
Residual Error 6 56.85 9.48
Total 8 899.22
b) V 2 9.48
se( E0 ) 28.40 , se( E1 ) 1.539 , and se( E 2 ) 1.539
c) Based on the P-values from the t-test for each coefficient, both regressors appear to be significant at the 0.05 level of
significance.
f) The prediction interval is wider than the confidence interval because it predicts a range for a future observation
whereas the confidence interval predicts a range for the mean response.
Analysis of Variance
Source DF SS MS F P
Regression 2 133.366 66.683 96.18 0.000
Residual Error 25 17.332 0.693
Total 27 150.698
Coef x1 0.7417
Tx1 1.2859
SE Coef x1 0.5768
Coef x 2 9.1142
Tx 2 13.8682
SE Coef x 2 0.6571
P-valuex1 = 2P(t > |1.29|) and for degrees of freedom of 25 we obtain
2(0.10) <P-value < 2(0.25) = 0.20 < P-value < 0.50
45
Engineering Statistics 5th edition October 01, 2010
P-valuex2 = 2P(t > |13.87|) and for degrees of freedom of 25 we obtain P-value < 2(0.0005) = P-value < 0.001
SS 17.332
R2 1 E 1 88.50%
SS T 150.698
DFError DFTotal DFRe g 27 2 25
SS Error 17.332
MS Error 0.6933
DFError 25
S MS Error 0.6933 0.832646
MS Re g 66.683
F 96.182
MS Error 0.6933
P valueregression 2 P( F2, 25 ! 96.18) 2( P value 0.01) P value 0.02
b) V 2
0.693
c) Based on the ANOVA from MINITAB output in part (a), F0 96.18 ! F0.05, 2, 25 3.39 and P-value = 0.000
< = 0.05. We reject the null hypothesis and conclude that the regression is significant.
d) 1: T0 1.29 t 0.025, 25 2.060 and P-value = 0.210 > = 0.05. There is not sufficient evidence to conclude that 1
differs from zero.
2: T0 13.87 ! t 0.025, 25 2.060 and P-value = 0.000 < = 0.05, we reject the null hypothesis and conclude that 2 is
different from zero.
e) 95% CI for 1
E1 t 0.025, 25 se E1 d E 1 d E1 t 0.025, 25 se E1
0.7417 2.060(0.5768) d E 1 d 0.7417 2.060(0.5768)
0.7417 1.188 d E 1 d 0.7417 1.188
0.4463 d E 1 d 1.9297
Because zero is included in the 95% CI, we fail to reject the null hypothesis that 1 differs from zero.
f) 95% CI for 2
E 2 t 0.025, 25 se E 2 d E 2 d E 2 t 0.025, 25 se E 2
9.1142 2.060(0.6571) d E 2 d 9.1142 2.060(0.6571)
9.1142 1.354 d E 2 d 9.1142 1.354
7.7602 d E 2 d 10.4682
Because zero is not included in the 95% CI, we reject the null hypothesis that 2 equals zero.
g) Only 2 is significantly different from zero (parts d) and f)), and the regression is significant (part c)). The next step
would be to fit a model with x2 as the sole predictor. A residual analysis should be carried out to make sure that the
chosen model is adequate.
Analysis of Variance
Source DF SS MS F P
Regression 3 363.01 121.00 145.37 0.000
Residual Error 44 36.62 0.83
Total 47 399.63
46
Engineering Statistics 5th edition October 01, 2010
Coef x1 9.6864
Tx1 19.4156
SE Coef x1 0.4989
Coef x 2 0.3796
Tx 2 1.6229
SE Coef x 2 0.2339
Coef x 3 2.9447
Tx 3 12.5093
SE Coef x 3 0.2354
P-valuex1 = 2P(t > |19.42|): for degrees of freedom of 44 we obtain P-value < 2(0.0005) = P-value < 0.001
P-valuex2 = 2P(t > |-1.62|): for degrees of freedom of 44 we obtain 2(0.05) < P-value < 2(0.1) = 0.1 < P-value < 0.2
P-valuex2 = 2P(t > |12.51|): for degrees of freedom of 44 we obtain P-value < 2(0.0005) = P-value < 0.001
SS 36.62
R2 1 E 1 90.84%
SS T 399.63
SS Error 36.62
MS Error 0.8323
DFError 44
S MS Error 0.8323 0.912305
MS Re g 121.00
F 145.38
MS Error 0.8323
P value 2 P( F3, 44 ! 145.38) 2( P value 0.01) P value 0.02
b) V 2
0.83
c) Based on the ANOVA from MINITAB output in part (a), F0 145.37 ! F0.05,3, 44 2.824 (interpolated) and
P-value = 0.000 < = 0.05. We reject null hypothesis and conclude that the regression is significant.
d) 1: T0 19.42 ! t 0.025, 44 2.017 (interpolated) and P-value = 0.000 < = 0.05. We reject the null hypothesis and
conclude that 1 is different from zero.
2: T0 1.62 t 0.025, 44 2.017 (interpolated) and P-value = 0.112 > = 0.05, we fail to reject the null hypothesis that
2 equals zero.
3: T0 12.51 ! t 0.025, 44 2.017 (interpolated) and P-value = 0.000 < = 0.05, we reject the null hypothesis that 3
equals zero.
e) 95%CI for 1
E1 t 0.025, 44 se E1 d E 1 d E1 t 0.025, 44 se E1
9.6864 2.017(0.4989) d E 1 d 9.6864 2.017(0.4989)
9.6864 1.006 d E 1 d 9.6864 1.006
8.6804 d E 1 d 10.6924
Because zero is not included in the CI, we reject the null hypothesis that 1 equals zero.
f) 95%CI for 2
E 2 t 0.025, 44 se E 2 d E 2 d E 2 t 0.025, 44 se E 2
0.3796 2.017(0.2339) d E 2 d 0.3796 2.017(0.2339)
0.3796 0.472 d E 2 d 0.3796 0.472
0.8516 d E 2 d 0.0924
Because zero is included in the 95% CI, we fail to reject the null hypothesis that 2 equals zero.
g) 95%CI for 3
E3 t 0.025, 44 se E3 d E 3 d E3 t 0.025, 44 se E3
2.9947 2.017(0.2345) d E 3 d 2.9947 2.017(0.2345)
2.9947 0.473 d E 3 d 2.9947 0.473
2.5217 d E 3 d 3.4677
Because zero is not included in the CI, we reject the null hypothesis that 3 equals zero.
47
Engineering Statistics 5th edition October 01, 2010
h) The tests for 1 and 3 are significant while 2 is not significant. Also, the regression is significant (part (c)). A
model with x2 omitted should be considered. Residual analysis should be applied to make sure that model is adequate.
Section 6-4
S = *
Analysis of Variance
Source DF SS MS F P
Regression 5 14112.00 2822.40 * *
Residual Error 0 * *
Total 5 14112.00
Source DF Seq SS
x1 1 10240.37
x2 1 1921.21
x1x2 1 827.86
x1^2 1 1056.39
x2^2 1 66.17
b) Because VIFs are much greater than 10, multicollinearity is present in the second-order model.
c) Because SSE(Full Model) is not available the test statistic can not be computed.
Analysis of Variance
Source DF SS MS F P
Regression 5 440.988 88.198 22.57 0.001
Residual Error 6 23.449 3.908
Total 11 464.438
Source DF Seq SS
W 1 236.699
HP 1 186.707
W*HP 1 9.307
W^2 1 0.001
HP^2 1 8.274
b) Because the VIFs are much greater than 10, multicollinearity is present in the second-order model.
c) The regression equation is
48
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 2 423.41 211.70 46.44 0.000
Residual Error 9 41.03 4.56
Total 11 464.44
Response is Sat
S
A S u A
Mallows g e r n
Vars R-Sq R-Sq(adj) C-p S e v g x
1 82.1 80.8 4.4 9.3577 X
1 57.0 54.0 27.3 14.487 X
2 87.9 86.1 1.1 7.9723 X X
2 83.0 80.4 5.5 9.4476 X X
3 88.0 85.0 3.0 8.2768 X X X
3 87.9 84.9 3.0 8.2942 X X X
4 88.0 83.6 5.0 8.6446 X X X X
Step 1 2
Constant 136.2 146.7
Sev -0.56
T-Value -2.51
P-Value 0.026
S 9.36 7.97
R-Sq 82.07 87.92
R-Sq(adj) 80.79 86.06
Mallows C-p 4.4 1.1
Step 1 2 3
Constant 146.2 146.2 146.7
49
Engineering Statistics 5th edition October 01, 2010
Surg 0.1
T-Value 0.03
P-Value 0.979
d) Model with only age and severity seems to be the best among all. It has a large R-Sq(adj) and small Cp and
values.
Response is Density
C L
o o
Mallows n s
Vars R-Sq R-Sq(adj) C-p S t s
1 99.7 99.7 1.1 0.0083967 X
1 99.5 99.5 6.9 0.010964 X
2 99.7 99.7 3.0 0.0088342 X X
Step 1
Constant -0.2005
Cont 0.4679
T-Value 57.81
P-Value 0.000
S 0.00840
R-Sq 99.73
R-Sq(adj) 99.70
Mallows C-p 1.1
Step 1 2
Constant -0.1105 -0.2005
Loss 2.1
T-Value 0.36
P-Value 0.727
50
Engineering Statistics 5th edition October 01, 2010
S 0.00883 0.00840
R-Sq 99.74 99.73
R-Sq(adj) 99.67 99.70
Mallows C-p 3.0 1.1
d) Model with only Dielectric constant seems to be the best among all. It has a high R-Sq(adj) and small Cp and S
values.
Response is y
Adj. x x x x
Vars R-Sq R-Sq C-p s 1 2 3 4
Step 1 2
Constant -90.1607 0.5287
x2 15.2 10.3
T-Value 4.26 2.36
P-Value 0.002 0.042
x1 0.50
T-Value 1.71
P-Value 0.122
S 15.4 14.1
R-Sq 64.46 73.14
R-Sq(adj) 60.90 67.17
C-p 1.7 1.4
Step 1 2 3 4
Constant -102.7132-101.6100 0.5287 -90.1607
x3 1.4 1.4
T-Value 0.60 0.65
P-Value 0.567 0.536
x4 0.01
T-Value 0.02
P-Value 0.986
51
Engineering Statistics 5th edition October 01, 2010
d) The model with only x1 and x2 seems to be the best among all. It has a high R-Sq(adj) and small Cp value.
Response is y
x x x
Vars R-Sq R-Sq(adj) C-p S 1 2 3
Step 1 2 3
Constant -23.62 66.13 47.17
x1 -5.4 -9.7
T-Value -1.72 -2.64
P-Value 0.103 0.018
x2 0.43
T-Value 1.91
P-Value 0.074
Step 1
Constant 47.17
x1 -9.7
T-Value -2.64
P-Value 0.018
x2 0.43
T-Value 1.91
P-Value 0.074
x3 18.2
T-Value 13.90
P-Value 0.000
S 3.48
R-Sq 99.37
52
Engineering Statistics 5th edition October 01, 2010
R-Sq(adj) 99.25
C-p 4.0
d) The model that contains all the first-order terms seems to be the best among all. It has the highest R-Sq(adj) and
smallest Cp value.
6-39. a) Note that x2 = 0 if using tool type 302 and x2 = 1 if using tool type 416.
Analysis of Variance
Source DF SS MS F P
Regression 2 1083.55 541.77 1171.07 0.000
Residual Error 17 7.86 0.46
Total 19 1091.41
Source DF Seq SS
x1 1 201.09
x2 1 882.46
Unusual Observations
Obs x1 y Fit SE Fit Residual St Resid
7 248 37.520 36.030 0.240 1.490 2.34R
Because the P-value < 0.01, the regression model is significant at 0.01.
Analysis of Variance
Source DF SS MS F P
Regression 1 104.46 104.46 775.21 0.000
Residual Error 8 1.08 0.13
Total 9 105.54
53
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 1 34.277 34.277 52.85 0.000
Residual Error 8 5.189 0.649
Total 9 39.466
Unusual Observations
Obs Tool2 y-Tool2 Fit SE Fit Residual St Resid
7 248 37.520 35.760 0.332 1.760 2.40R
Supplemental Exercises
6-40. a)
Source DF SS MS F P
0.00008 /1
f0 0.004
0.2631/13
54
Engineering Statistics 5th edition October 01, 2010
Because 0.01 < 4.67 we fail to reject H0. The regression model is not significant at D = 0.05.
P-value > 0.10 (from computer output the P-value = 0.950
0.0089
t0 0.0642
0.1387
Because 2.160 < 0.0642 < 2.160 we fail to reject H 0 There is not sufficient evidence to conclude that Dollars is a predictor of
electrical usage at D = 0.05.
SSE
6-41. Using R 2 1 Syy
,
( n 2) 1 S E
SS
yy Syy SSE Syy SSE
F0 SS E SS E
V 2
S yy n2
Also,
SSE ( yi E 0 E 1xi ) 2
( y y E ( x x )) 2
i 1 i
( yi y) E 12 ( xi x ) 2 2E 1 ( y i y)( x i x )
( yi y) 2 E 12 ( xi x ) 2
SST SSE E 12 ( xi x ) 2
E 12
Therefore, F0 t 20
V 2 / S xx
Because the square of a t random variable with n - 2 degrees of freedom is an F random variable with 1 and n-2 degrees of
freedom, the usual t-test that compares | t 0 | to t D / 2 ,n 2 is equivalent to comparing f 0 t 20 to f D ,1,n 2 t D / 2 ,n 2 .
55
Engineering Statistics 5th edition October 01, 2010
0.9(23)
6-42. a) From the previous exercise f 0 207 . Reject H 0 :E1 0.
1 0.9
23R 2
b) Because f0.05,1,23 4.28, H 0 is rejected if ! 4.28 .
1 R2
That is, H 0 is rejected if
23R 2 ! 4.28(1 R 2 )
27.28R 2 ! 4.28
R 2 ! 0.157
( x x) 2
V 2 1 ( n1 iS )
xx
a) Because ei is divided by its standard error (when V is known), ri has a unit standard deviation.
b) No, the term in brackets in the denominator is necessary for the standardized residuals to have unit standard
deviation.
d) If xi is far from x , the standard error of ei is small. Consequently, extreme points are better fit by least squares
regression than points near the middle range of x. Because the studentized residual at any point has variance of
approximately one, the studentized residuals can be used to compare the fit of points to the regression line over the
range of x.
6-44. a)
2
DCoutput
0
2 3 4 5 6 7 8 9 10
WindVel
The scatter diagram shows definite curvature. A higher-order polynomial model or a transformation of variables may
be appropriate.
56
Engineering Statistics 5th edition October 01, 2010
y 0.131 0.241x
c) Analysis of Variance
Source DF SS MS F P
Regression 1 8.9296 8.9296 160.26 0.000
Residual Error 23 1.2816 0.0557
Total 24 10.2112
The P-value from the ANOVA table is approximately zero. Therefore, reject H0 and conclude that the regression model
is significant at D = 0.01. The test can also be conducted in more detail as follows:
d)
Residual Plot for DC Output Residual Plot for DC Output
0.4 0.4
0.2 0.2
Residuals
Residuals
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
e) Examining the residual plots in part d), a transformation on the x-variable, y-variable, or both would be appropriate. A
simple linear regression of y on the transformed variable 1/x may be satisfactory.
57
Engineering Statistics 5th edition October 01, 2010
Analysis of Variance
Source DF SS MS F P
Regression 1 10.007 10.007 1128.43 0.000
Residual Error 23 0.204 0.009
Total 24 10.211
The P-value from the ANOVA table is approximately zero. Therefore, reject H0 and conclude that the regression model
is significant at D = 0.05. The test can also be conducted in more detail as follows:
0.19 0.19
0.09 0.09
Residuals
Residuals
-0.01
-0.01
-0.11
-0.11
-0.21
-0.21
0 0.4 0.8 1.2 1.6 2 2.4
0 0.1 0.2 0.3 0.4 0.5
Predicted
1/WindVel
Using 1/WindVel
From the random appearance of the residuals in the plots we conclude that he model is adequate. The transformation,
1/(Wind Velocity), appears to be satisfactory as a regressor of DC Output.
6-45. a) p = k + 1 = 2 + 1 = 3
Average size = p/n = 3/25 = 0.12
58
Engineering Statistics 5th edition October 01, 2010
b) Analysis of Variance
Source DF SS MS F P
Regression 3 9863398 3287799 1724.42 0.000
Residual Error 36 68638 1907
Total 39 9932036
The P-value from the ANOVA table is approximately zero. Therefore, reject H0 and conclude that the regression model
is significant at D = 0.01. The test can also be conducted in more detail as follows:
H 0 :E 3 E4 E5 0
H 1:E j z 0 for at least one j
The test statistic is
SS R / k
f0
SS E / ( n p)
9863398 / 3
f0 1724.42
68638.2 / 36
Because 1724.42 > 4.38 reject H0 and conclude that the regression model is significant at D = 0.01. The P-value <
0.00001
59
Engineering Statistics 5th edition October 01, 2010
99.9
99
95
cumulative percent
80
50
20
0.1
Residuals
The normality assumption appears reasonable. The residuals fall along a line.
f)
Residual Plot
100
70
40
Residuals
10
-20
-50
-80
The plot is satisfactory. There does not appear to be a nonrandom pattern in the residual vs. predicted plot.
g)
60
Engineering Statistics 5th edition October 01, 2010
100
70
40
Residuals
10
-20
-50
-80
There is a slight indication that variance increases as x3 increases. There is a fanning out appearance of the
residuals.
Analysis of Variance
Source DF SS MS F P
Regression 3 0.68611 0.22870 1323.62 0.000
Residual Error 36 0.00622 0.00017
Total 39 0.69233
The P-value from the ANOVA table is approximately zero. Therefore, reject H0 and conclude that the regression model
is significant at D = 0.01. The test can also be conducted in more detail as follows:
H 0:E
3 E4 E5 0
H 1:E j z 0 for at least one j
The test statistic is
SS R / k
f0
SS E / ( n p)
Reject H0 if f0 > fD,3,36 where f0.01,3,36 = 4.38
Using the results from the ANOVA table
0.686112 / 3
f0 1323.62
0.00622033 / 36
Because 1323.62 > 4.38 reject H0 and conclude that the regression model is significant at D = 0.01.
P-value < 0.00001
61
Engineering Statistics 5th edition October 01, 2010
E3: Fail to reject H0. There is not sufficient evidence that the coefficient of ln(x3) in the model differs from zero at D =
0.01.
E4: Reject H0. The coefficient of x4 in the models differs from zero at D = 0.01.
E5: Fail to reject H0. There is not sufficient evidence that the coefficient of x5 in the model differs from zero at D =
0.01.
c)
28
28
18
18
Residuals
Residuals
8
8
-2
-2
-12 -12
-22 -22
8 8.1 8.2 8.3 8.4 8.5 1026 1027 1028 1029 1030 1031 1032
(X 0.01)
Predicted ln(x3)
Curvature is evident in the residuals plots from this model, whereas non-constant variance was evident in the previous
model.
6-48. a)
62
Engineering Statistics 5th edition October 01, 2010
The P-value from the ANOVA table is approximately zero. Therefore, reject H0 and conclude that the regression model
is significant at D = 0.01. The test can also be conducted in more detail as follows:
26.633 / 2
f0 572
0.163 / 7
Because 572 > 4.74 reject H0 and conclude that the regression model is significant at D = 0.05.
c) H 0 :E11 0
H 1:E11 z 0
The test statistic is
E 1 E1,0
t0
se(E )
1
Reject H0 if t0 < tD/2,n-2 where t0.025,7 = 2.365 or t0 > t0.025,7 = 2.365
Using the results from the table given in part a)
0.005699 0
t0 9.498
0.0006
Because 9.498 < 2.365 reject H 0 and conclude the quadratic term contributes significantly to the model at D = 0.05.
d)
63
Engineering Statistics 5th edition October 01, 2010
There might be a slight indication of non-constant variance. There is greater variability in the residuals as the predicted
values decrease.
e)
6- 49. a)
64
Engineering Statistics 5th edition, SI
a)
The simple linear regression model seems appropriate. The data fall along a straight line.
65
Engineering Statistics 5th edition, SI
b)
= 0.28 + 20.6x
c) = 0.28 + 20.6(1) = 20.88
d)
= 0.28 + 20.6(0.5) = 10.58
ei yi y i 11.8 10.58 1.22
e) The least squares estimate minimizes ( y i x i ) 2 .
Upon setting the derivative equal to zero, we obtain
2 ( y x )( x ) 2[ y x x 2 ] 0
i i i i i i
Solving for ,
yi xi
2
xi
f)
= 0.28 + 20.6x
66
Engineering Statistics 5th edition, SI
Examining the plot, the model seems appropriate in this case. In part b) the coefficient for the intercept was not
significantly different from zero (P-value = 0.811) so that the zero intercept model could be expected to provide an
adequate fit in this exercise.
6-50. a) has a t distribution with n 1 degrees of freedom.
2
x2i
b) From the previous exercsie
21031461
. , 3.611768, and
x 2i 14.7073 .
Therefore,
21031461
.
t0 22.3314 and H 0 : 0 is rejected at usual values.
3.611768
14.7073
6-51. a)
A linear regression model seems appropriate. The data fall along a line.
67
Engineering Statistics 5th edition, SI
b)
d)
68
Engineering Statistics 5th edition, SI
If there were no error, the values would all lie along the 45 axis. YES, the plot indicates that age is a reasonable
regressor variable.
6-52. a)
2800 2800
2600 2600
2400 2400
Strength
Strength
2200 2200
2000 2000
1800 1800
1600 1600
-12 -8 -4 0 4 8 12 0 5 10 15 20 25
z Age
The slopes of both regression models are the same, but the y-intercepts differ.
y 2132.41 36.9618x
69
Engineering Statistics 5th edition, SI
0 2625.39 0 2132.41
vs.
36.9618
1 1 36.9618
Because the data is shifted by the mean of age, the intercept in the model Y 0* 1* z is now the average of
strength.
1
6-53. t0
2 / S xx
After the transformation
b
1 1 ,
a
S xx a 2 S xx ,
x ax , 0 b 0 ,
and b .
b 1 / a ( b / a ) 1 1
Therefore, t 0 t0 .
( b ) 2 / a 2Sxx ( b / a ) 2 Sxx 2 Sxx
6-54. H 0 :1 10
H 1:1 10
The test statistic is
1 1,0
t0
se( )
1
Reject H0 if t0 < t/2,n-2 where t0.005,10 = 3.17 or t0 > t0.005,10 = 3.17
Using the data from the referenced exercise
9.21 10
t0 23.37
0.0338
Because 23.37 < 3.17 reject H 0 . The coefficient of the regressor is significantly different from 10 at = 0.01.
P-value = 0.
Response is y*
x
x x 3 x x x
Vars R-Sq R-Sq(adj) C-p S 1 2 * 4 5 6
Step 1 2 3 4 5
Constant 7.275 6.837 6.698 6.728 -15.490
70
Engineering Statistics 5th edition, SI
x5 0.00041 0.00026
T-Value 4.33 2.11
P-Value 0.000 0.042
x3* 2.2
T-Value 1.81
P-Value 0.080
Step 1 2 3
Constant -16.26 -15.49 -23.51
x1 -0.00004
T-Value -0.37
P-Value 0.713
x2 0.00002 0.00002
T-Value 1.25 1.23
P-Value 0.220 0.227
d) The model with only x3*, x4, x5, and x6 seems to be the best among all. It has a high R-Sq(adj) and small Cp value.
Response is Sat
S
A S u A
Mallows g e r n
Vars R-Sq R-Sq(adj) C-p S e v g x
71
Engineering Statistics 5th edition, SI
Step 1 2
Constant 136.2 146.7
Sev -0.56
T-Value -2.51
P-Value 0.026
S 9.36 7.97
R-Sq 82.07 87.92
R-Sq(adj) 80.79 86.06
Mallows C-p 4.4 1.1
Step 1 2 3
Constant 146.2 146.2 146.7
Surg 0.1
T-Value 0.03
P-Value 0.979
d) Model with only age and severity seems to be the best among all. It has a large R-Sq(adj) and small Cp and S
values.
72
Engineering Statistics 5th edition, SI
Analysis of Variance
Source DF SS MS F P
Regression 2 11076473 5538237 99.67 0.000
Residual Error 37 2055837 55563
Total 39 13132310
Analysis of Variance
Source DF SS MS F P
Regression 4 3.84906 0.96227 17.28 0.000
Residual Error 16 0.89121 0.05570
Total 20 4.74028
73