Regression Model Selection - Lea Strength

Dependent variable: Lea Strength

Independent variables:
A=b_
B=ELG%
C=INV
D=M_R
E=MIC
F=MOIS
G=Nep
H=RD
I=STR
J=TR
K=UHML
L=UI

Number of complete cases: 22

Number of models fit: 1586
Models with Largest Adjusted R-Squared
Adjusted Included
MSE R-Squared R-Squared Cp Variables
0.863355 74.2513 66.2048 1.76943 ADEIK
0.93293 72.1763 63.4814 2.71789 AEFIK
0.9554 69.7253 62.6019 1.83821 AEIK
0.978531 70.8163 61.6964 3.33953 AEFGK
0.983077 70.6807 61.5185 3.4015 AEFJK
0.988111 70.5306 61.3214 3.47012 AEHIK
1.00557 68.1355 60.638 2.56487 AEFK
1.10294 62.9941 56.8264 2.91498 AEK
1.13025 64.1846 55.7575 4.3708 AEKL
1.13114 64.1565 55.7227 4.38367 AEJK
1.14472 63.7263 55.1913 4.5803 AEFJ
1.24825 58.1188 51.1386 5.1434 AIK
1.2562 57.8519 50.8272 5.26542 AFK
1.28087 54.6368 49.8617 4.73499 AK
1.29844 56.4347 49.1738 5.91321 ACK
1.32349 55.5944 48.1935 6.29727 AJK
1.53459 45.6509 39.9299 8.84235 AJ
1.60839 43.0373 37.0412 10.037 CK
1.66834 40.9142 34.6946 11.0074 AE
1.67672 40.6174 34.3665 11.1431 AF
1.75037 34.7464 31.4837 11.8267 A
2.0511 23.5349 19.7117 16.9513 C
2.05535 23.3766 19.5454 17.0237 K
2.19753 18.0761 13.9799 19.4465 J
2.46123 8.24539 3.65766 23.94 E
2.55467 0.0 0.0 25.7088

The StatAdvisor
This table shows the models which give the largest adjusted R-Squared values. The adjusted R-Squared statistic measures the
proportion of the variability in Lea Strength which is explained by the model. Larger values of adjusted R-Squared correspond to
smaller values of the mean squared error (MSE). Up to 5 models in each subset of between 0 and 5 variables are shown. The
best model contains 5 variables, b_, M_R, MIC, STR, and UHML.
 Adjusted R-Squared Plot for Lea Strength

adjusted R-Squared 80

60

40

20

0
0 2 4 6 8
Number of Coefficients

Mallows' Cp Plot for Lea Strength

30

25

20
Cp

15

10

0
0 2 4 6 8
Number of Coefficients
 MSE for Lea Strength

2.6

2.3

2
MSE

1.7

1.4

1.1

0.8
0 2 4 6 8
Number of Coefficients

R-Squared Plot for Lea Strength

80

60
R-Squared

40

20

0
0 2 4 6 8
Number of Coefficients

Regression Model Selection - Lea Strength

Dependent variable: Lea Strength
Independent variables:
A=b_
B=ELG%
 C=INV
D=M_R
E=MIC
F=MOIS
G=Nep
H=RD
I=STR
J=TR
K=UHML
L=UI

Number of complete cases: 28

Number of models fit: 1586
Models with Largest Adjusted R-Squared
Adjusted Included
MSE R-Squared R-Squared Cp Variables
1.35078 74.0596 68.1641 1.03692 ACEKL
1.40807 72.9594 66.8138 1.75949 ACEFK
1.41863 71.5182 66.5649 0.706016 ACDK
1.43306 72.4796 66.2249 2.07466 ACEGK
1.43556 72.4316 66.166 2.10617 ACDKL
1.43605 72.4221 66.1544 2.11238 ACDFK
1.44711 70.9465 65.8937 1.08154 ACEK
1.46128 70.6621 65.5599 1.26831 AEGK
1.47341 70.4185 65.2739 1.42827 ADFK
1.47629 69.0721 65.2061 0.312598 ADK
1.48745 70.1367 64.9431 1.61338 ACGK
1.5031 68.5103 64.5741 0.681546 ACK
1.51712 68.2167 64.2437 0.874404 AGK
1.5708 67.0921 62.9786 1.61301 AEK
1.62303 65.9978 61.7475 2.33172 AFK
1.63011 64.4266 61.5807 1.36363 AK
1.67896 63.3605 60.4294 2.06378 AC
1.86639 59.2702 56.0119 4.75016 AG
1.91285 58.2565 54.917 5.41596 AE
1.91456 58.2191 54.8767 5.4405 AD
1.96267 55.456 53.7428 5.25522 A
2.07673 52.8673 51.0545 6.95544 C
3.44632 21.7837 18.7753 27.3703 G
3.71577 15.6683 12.4248 31.3867 K
3.75704 14.7317 11.4522 32.0018 D
4.24295 0.0 0.0 39.6772

The StatAdvisor
This table shows the models which give the largest adjusted R-Squared values. The adjusted R-Squared statistic measures the
proportion of the variability in Lea Strength which is explained by the model. Larger values of adjusted R-Squared correspond to
smaller values of the mean squared error (MSE). Up to 5 models in each subset of between 0 and 5 variables are shown. The
best model contains 5 variables, b_, INV, MIC, UHML, and UI.
 Adjusted R-Squared Plot for Lea Strength

100

80
adjusted R-Squared

60

40

20

0
0 2 4 6 8
Number of Coefficients

Mallows' Cp Plot for Lea Strength

150

120

90
Cp

60

30

0
0 2 4 6 8
Number of Coefficients
 MSE for Lea Strength

3
MSE

0
0 2 4 6 8
Number of Coefficients

R-Squared Plot for Lea Strength

100

80
R-Squared

60

40

20

0
0 2 4 6 8
Number of Coefficients

Multiple Regression - Lea Strength

Dependent variable: Lea Strength
Independent variables:
b_
MIC
 INV
UHML
UI

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT 12.9941 9.95281 1.30557 0.2027
b_ -1.30411 0.280171 -4.6547 0.0001
MIC 4.11833 0.391247 10.5262 0.0000
INV 0.655937 0.297472 2.20504 0.0362
UHML 0.88318 0.209151 4.22269 0.0002
UI 0.214146 0.0689024 3.10797 0.0044

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 88.3134 5 17.6627 42.24 0.0000
Residual 11.2898 27 0.418141
Total (Corr.) 99.6032 32

R-squared = 88.6652 percent

R-squared (adjusted for d.f.) = 86.5662 percent
Standard Error of Est. = 0.646639
Mean absolute error = 0.481544
Durbin-Watson statistic = 1.88877 (P=0.2430)
Lag 1 residual autocorrelation = 0.052469

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between Lea Strength and 5
independent variables. The equation of the fitted model is

Lea Strength = 12.9941 - 1.30411*b_ + 4.11833*MIC + 0.655937*INV + 0.88318*UHML + 0.214146*UI

Since the P-value in the ANOVA table is less than 0.05, there is a statistically significant relationship between the variables at the
95.0% confidence level.

The R-Squared statistic indicates that the model as fitted explains 88.6652% of the variability in Lea Strength. The adjusted R-
squared statistic, which is more suitable for comparing models with different numbers of independent variables, is 86.5662%.
The standard error of the estimate shows the standard deviation of the residuals to be 0.646639. This value can be used to
construct prediction limits for new observations by selecting the Reports option from the text menu. The mean absolute error
(MAE) of 0.481544 is the average value of the residuals. The Durbin-Watson (DW) statistic tests the residuals to determine if
there is any significant correlation based on the order in which they occur in your data file. Since the P-value is greater than 0.05,
there is no indication of serial autocorrelation in the residuals at the 95.0% confidence level.

In determining whether the model can be simplified, notice that the highest P-value on the independent variables is 0.0362,
belonging to INV. Since the P-value is less than 0.05, that term is statistically significant at the 95.0% confidence level.
Consequently, you probably don't want to remove any variables from the model.

Correlation matrix for coefficient estimates

CONSTANT b_ MIC INV UHML UI
CONSTANT 1.0000 -0.5106 -0.4845 -0.5681 -0.5882 -0.6529
b_ -0.5106 1.0000 0.2052 0.6812 0.1194 0.1077
MIC -0.4845 0.2052 1.0000 0.3179 0.0484 0.3315
INV -0.5681 0.6812 0.3179 1.0000 -0.0109 0.4124
UHML -0.5882 0.1194 0.0484 -0.0109 1.0000 -0.0735
UI -0.6529 0.1077 0.3315 0.4124 -0.0735 1.0000
The StatAdvisor
This table shows estimated correlations between the coefficients in the fitted model. These correlations can be used to detect the
presence of serious multicollinearity, i.e., correlation amongst the predictor variables. In this case, there is 1 correlation with
absolute value greater than 0.5 (not including the constant term).
Unusual Residuals
Predicted Studentized
Row Y Y Residual Residual
4 63.2 64.4526 -1.25256 -2.14
10 62.87 64.1418 -1.27183 -2.25

The StatAdvisor
The table of unusual residuals lists all observations which have Studentized residuals greater than 2 in absolute value.
Studentized residuals measure how many standard deviations each observed value of Lea Strength deviates from a model fitted
using all of the data except that observation. In this case, there are 2 Studentized residuals greater than 2, but none greater than 3.

Component+Residual Plot for Lea Strength

3.1

2.1
component effect

1.1

0.1

-0.9

-1.9
8.3 8.8 9.3 9.8 10.3 10.8 11.3
b_
 Plot of Lea Strength

67

65
observed

63

61

59
59 61 63 65 67
predicted

Residual Plot

2.7

1.7
Studentized residual

0.7

-0.3

-1.3

-2.3
8.3 8.8 9.3 9.8 10.3 10.8 11.3
b_
 Residual Plot

2.7

1.7
Studentized residual

0.7

-0.3

-1.3

-2.3
0 10 20 30 40
row number

Plot of Lea Strength with Predicted Values

67

65
Lea Strength

63

61

59
8.3 8.8 9.3 9.8 10.3 10.8 11.3
b_

Response Lea Strength

Sorted Parameter Estimates
Term Estimate Std Error t Ratio t Ratio
MIC 4.1183328 0.391247 10.53

b+ -1.304113 0.280171 -4.65

UHML 0.8831796 0.209151 4.22
UI 0.2141463 0.068902 3.11
INV 0.6559372 0.297472 2.21
Correlation of Estimates
Corr Intercept b+ INV MIC UHML UI
Intercept 1.0000 -0.5106 -0.5681 -0.4845 -0.5882 -0.6529
b+ -0.5106 1.0000 0.6812 0.2052 0.1194 0.1077
INV -0.5681 0.6812 1.0000 0.3179 -0.0109 0.4124
MIC -0.4845 0.2052 0.3179 1.0000 0.0484 0.3315
UHML -0.5882 0.1194 -0.0109 0.0484 1.0000 -0.0735
UI -0.6529 0.1077 0.4124 0.3315 -0.0735 1.0000

Bayes Plot

Term
b+
INV
MIC
UHML
UI

Posterior Prob that sample is uncontaminated:

0.0000
Pareto Plot of Estimates

Term
MIC
b+
UHML
UI
INV
 Lea
Strength
63.40624
±0.230965

59
61
63
65
67
Prediction Profiler

b+
9.8985 10

11
1.5

2.5
INV
3.1097

3.5

4.5
4

4.5
MIC
4.6052

5.5
27
27.5
28
UHML
28.0245

28.5
29
75

80
UI
82.022

85

90