Université MOHAMMED V AGDAL

ECOLE MOHAMMADIA D’INGENIEURS

Département Génie des procédés industriels

TP de la Modélisation, simulation et
optimisation des procédés

Régression multilinéaire

Réalisé Par : Encadré Par :

 ZOUHEIR ABDELLAH  Pr. MAATAH SOUFIANE

 ALAOUI SAED

Année Universitaire 2022 –2023

 Table des matières

1 Introduction…………………………………………………………………………………………………………….3

2 Effet principal………………………………………………………………………..………………………………..4

2.1 least square……………………………………………………………………………………………………………….4

2.2 Forward stepwise……………………….……………………………………………………………………………..5
2.3 Backward stepwise selection…………….……………………………………………………………………….6

3 Effet d’interaction……………………………..…………………………………………………………………….8

3.1 least square………………………………………………………………………………………………………………8

3.2 Forward stepwise……………………………………………………………………………………………………..10
3.3 Backward stepwise selection…………………………………………………………………………………….11

4 Effet quadratique……………………………………………………………………………………………………13

4.1 least square……………………………………………………………………………………………………………..13

4.2 Forward stepwise……………………………………………………………………………………………………..15
4.3 Backward stepwise selection…………………………………………………………………………………….17

5 Analyse des résultats………….………………….……………………………………………………………….19

2
 Introduction
L’adsorption de l’acide tannique sur la fluorine dépend de la concentration initiale de l’acide
tannique (Ci), de pH du mélange (pH) et de la température (T). Cette étude a donné les résultats
expérimentaux récapitulés dans le tableau suivant. Ces données sont obtenues en se basant sur un
plan d’expérience PCC (plan d’expérience composite centré)

1. Rappeler les méthodes de la régression multiple. Expliquer le principe de chaque méthode.

Méthode des moindres carrés (least squares):La méthode des moindres carrés consiste à minimiser
la somme des carrés des écarts, écarts pondérés dans le cas multidimensionnel, entre chaque point
du nuage de régression et son projeté, parallèlement à l'axe des ordonnées, sur la droite de
régression .

La sélection directe (Forwardstepwiseselection) : commence sans aucune variable dans le modèle,

teste chaque variable au fur et à mesure qu’elle est ajoutée au modèle, puis conserve celles qui sont
considérées comme les plus statistiquement significatives – en répétant le processus jusqu’à ce que
les résultats soient optimaux.

L’élimination en amont (Backwardsstepwiseselection): commence par un ensemble de variables

indépendantes, en supprimant une à la fois, puis en testant pour voir si la variable supprimée est
statistiquement significative.

2. Ajuster les différents modèles statistiques en utilisant les méthodes de régression multiple via le
logiciel de Statraphics.

3. Donner les résultats de l’analyse de variance pour chaque méthode de régression utilisée.

4. Analyser statistiquement les résultats obtenus.

5. Sélectionner le meilleur modèle. Justifier.

3
 Effet principal
Least square
Avec constant
Dependent variable: q
Independent variables:
T
PH
C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT 16,594 7,07093 2,34679 0,0354
T -0,0542286 0,0219484 -2,47073 0,0281
PH -0,120781 0,131691 -0,917158 0,3758
C 0,0599837 0,00823067 7,28783 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 65,0972 3 21,6991 20,02 0,0000
Residual 14,0907 13 1,0839
Total (Corr.) 79,1879 16

R-squared = 82,206 percent

R-squared (adjusted for d.f.) = 78,0996 percent
Standard Error of Est. = 1,04111
Mean absolute error = 0,747661
Durbin-Watson statistic = 1,92865 (P=0,4868)
Lag 1 residual autocorrelation = -0,00811005

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 3 independent
variables. The equation of the fitted model is

q = 16,594 - 0, 0542286*T - 0, 120781*PH + 0, 0599837*C

Plot of q Residual Plot

8,5 4

6,5
Studentized residual

2
observed

4,5
0
2,5

-2
0,5

-1,5 -4
-1,5 0,5 2,5 4,5 6,5 8,5 0 2 4 6 8
predicted predicted q

Sans constante
Dependent variable: q
Independent variables:
T
PH
C
Number of observations: 17

4
 Standard T
Parameter Estimate Error Statistic P-Value
T -0,00338543 0,00404273 -0,837412 0,4164
PH -0,0776124 0,149929 -0,517661 0,6128
C 0,0613327 0,00944022 6,49696 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 130,046 3 43,3487 30,25 0,0000
Residual 20,0602 14 1,43287
Total 150,106 17

R-squared = 86,636 percent

R-squared (adjusted for d.f.) = 84,7268 percent
Standard Error of Est. = 1,19703
Mean absolute error = 0,836892
Durbin-Watson statistic = 1,5777
Lag 1 residual autocorrelation = 0,193091

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 3 independent
variables. The equation of the fitted model is

q = -0, 00338543*T - 0, 0776124*PH + 0, 0613327*C

Plot of q Residual Plot

9,3 4

7,3
Studentized residual

2
observed

5,3
0
3,3

-2
1,3

-0,7 -4
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

Forward stepwise selection

Avec constante
Dependent variable: q
Independent variables:
T
PH
C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT 15,6881 6,96178 2,25347 0,0408
T -0,0542286 0,0218236 -2,48486 0,0262
C 0,0599837 0,00818385 7,32953 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 64,1854 2 32,0927 29,95 0,0000
Residual 15,0025 14 1,07161
Total (Corr.) 79,1879 16

5
R-squared = 81,0546 percent
R-squared (adjusted for d.f.) = 78,3481 percent
Standard Error of Est. = 1,03518
Mean absolute error = 0,79122
Durbin-Watson statistic = 1,95229 (P=0,5100)
Lag 1 residual autocorrelation = -0,0057418

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 3 independent
variables. The equation of the fitted model is

q = 15,6881 - 0,0542286*T + 0,0599837*C

Plot of q Residual Plot

8,8 4

6,8

Studentized residual
2
observed

4,8
0
2,8

-2
0,8

-1,2 -4
-1,2 0,8 2,8 4,8 6,8 8,8 0 2 4 6 8
predicted predicted q

Sans constante
Dependent variable: q

Independent variables:
T
PH
C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
T -0,00520442 0,00194971 -2,66933 0,0175
C 0,0612845 0,00920654 6,65662 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 129,662 2 64,831 47,57 0,0000
Residual 20,4442 15 1,36295
Total 150,106 17

R-squared = 86,3802 percent

R-squared (adjusted for d.f.) = 85,4722 percent
Standard Error of Est. = 1,16745
Mean absolute error = 0,838633
Durbin-Watson statistic = 1,58714
Lag 1 residual autocorrelation = 0,18624

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 3 independent
variables. The equation of the fitted model is

q = -0,00520442*T + 0,0612845*C
.

6
 Plot of q Residual Plot

9,4 5

7,4 3

Studentized residual
observed
5,4 1

3,4 -1

1,4 -3

-0,6 -5
-0,6 1,4 3,4 5,4 7,4 9,4 0 2 4 6 8
predicted predicted q

Backward stepwise selection

Avec constant
Dependent variable: q
Independent variables:
T
PH
C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT 15,6881 6,96178 2,25347 0,0408
T -0,0542286 0,0218236 -2,48486 0,0262
C 0,0599837 0,00818385 7,32953 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 64,1854 2 32,0927 29,95 0,0000
Residual 15,0025 14 1,07161
Total (Corr.) 79,1879 16

R-squared = 81,0546 percent

R-squared (adjusted for d.f.) = 78,3481 percent
Standard Error of Est. = 1,03518
Mean absolute error = 0,79122
Durbin-Watson statistic = 1,95229 (P=0,5100)
Lag 1 residual autocorrelation = -0,0057418

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 3 independent
variables. The equation of the fitted model is

q = 15,6881 - 0,0542286*T + 0,0599837*C

Plot of q Residual Plot

8,8 4

6,8
Studentized residual

2
observed

4,8
0
2,8

-2
0,8

-1,2 -4
-1,2 0,8 2,8 4,8 6,8 8,8 0 2 4 6 8
predicted predicted q

7
Sans constant
Dependent variable: q
Independent variables:
T
PH
C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
T -0,00520442 0,00194971 -2,66933 0,0175
C 0,0612845 0,00920654 6,65662 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 129,662 2 64,831 47,57 0,0000
Residual 20,4442 15 1,36295
Total 150,106 17

R-squared = 86,3802 percent

R-squared (adjusted for d.f.) = 85,4722 percent
Standard Error of Est. = 1,16745
Mean absolute error = 0,838633
Durbin-Watson statistic = 1,58714
Lag 1 residual autocorrelation = 0,18624

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 3 independent
variables. The equation of the fitted model is

q = -0,00520442*T + 0,0612845*C

Plot of q Residual Plot

9,4 5

7,4 3
Studentized residual
observed

5,4 1

3,4 -1

1,4 -3

-0,6 -5
-0,6 1,4 3,4 5,4 7,4 9,4 0 2 4 6 8
predicted predicted q

Effet d’interaction
Least square
Avec constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
Number of observations: 17

8
 Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT -13,9258 21,5284 -0,646854 0,5323
T 0,0371079 0,0675145 0,549628 0,5946
PH 0,149647 2,48233 0,0602848 0,9531
C 0,559422 0,156298 3,57919 0,0050
T*PH -0,000232057 0,00777971 -0,0298284 0,9768
T*C -0,00149327 0,000486232 -3,0711 0,0118
PH*C -0,00327723 0,00291739 -1,12334 0,2875

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 72,3789 6 12,0632 17,72 0,0001
Residual 6,80894 10 0,680894
Total (Corr.) 79,1879 16

R-squared = 91,4015 percent

R-squared (adjusted for d.f.) = 86,2425 percent
Standard Error of Est. = 0,825163
Mean absolute error = 0,568811
Durbin-Watson statistic = 0,773113 (P=0,0091)
Lag 1 residual autocorrelation = 0,542258

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 6 independent
variables. The equation of the fitted model is

q = -13,9258 + 0,0371079*T + 0,149647*PH + 0,559422*C - 0,000232057*T*PH - 0,00149327*T*C -

0,00327723*PH*C

Plot of q Residual Plot

9,5 2,5

7,5 1,5
Studentized residual
observed

5,5 0,5

3,5 -0,5

1,5 -1,5

-0,5 -2,5
-0,5 1,5 3,5 5,5 7,5 9,5 0 2 4 6 8
predicted predicted q

Sans constante
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
T -0,0064448 0,00485414 -1,32769 0,2112
PH -1,23894 1,21308 -1,02131 0,3290
C 0,515381 0,13692 3,76409 0,0031
T*PH 0,00410514 0,00383967 1,06914 0,3079
T*C -0,00135773 0,000427011 -3,17961 0,0088
PH*C -0,00316215 0,00283394 -1,11581 0,2883

Analysis of Variance

9
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 143,012 6 23,8354 36,96 0,0000
Residual 7,09384 11 0,644895
Total 150,106 17

R-squared = 95,2741 percent

R-squared (adjusted for d.f.) = 93,126 percent
Standard Error of Est. = 0,803053
Mean absolute error = 0,56896
Durbin-Watson statistic = 0,792475
Lag 1 residual autocorrelation = 0,550737

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 6 independent
variables. The equation of the fitted model is

q = -0,0064448*T - 1,23894*PH + 0,515381*C + 0,00410514*T*PH - 0,00135773*T*C - 0,00316215*PH*C

Plot of q Residual Plot

9,4 2,5

7,4 1,5

Studentized residual
observed

5,4 0,5

3,4 -0,5

1,4 -1,5

-0,6 -2,5
-0,6 1,4 3,4 5,4 7,4 9,4 0 2 4 6 8
predicted predicted q

Forward stepwise selection

Avec constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT -1,55656 0,434649 -3,58118 0,0030
C 0,396562 0,0785613 5,0478 0,0002
T*C -0,00105842 0,000246214 -4,29878 0,0007

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 69,8692 2 34,9346 52,48 0,0000
Residual 9,31873 14 0,665624
Total (Corr.) 79,1879 16

R-squared = 88,2321 percent

R-squared (adjusted for d.f.) = 86,551 percent
Standard Error of Est. = 0,815858
Mean absolute error = 0,600797
Durbin-Watson statistic = 1,21842 (P=0,0579)
Lag 1 residual autocorrelation = 0,352375

10
.

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 6 independent
variables. The equation of the fitted model is

q = -1,55656 + 0,396562*C - 0,00105842*T*C

Plot of q Residual Plot

9,3 3

2
7,3

Studentized residual
1
observed

5,3
0
3,3
-1

1,3
-2

-0,7 -3
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

Sans constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
PH -0,18674 0,0490054 -3,81061 0,0019
C 0,394501 0,0761411 5,18118 0,0001
T*C -0,00105842 0,000238782 -4,43258 0,0006

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 141,342 3 47,1139 75,26 0,0000
Residual 8,76464 14 0,626046
Total 150,106 17

R-squared = 94,161 percent

R-squared (adjusted for d.f.) = 93,3269 percent
Standard Error of Est. = 0,79123
Mean absolute error = 0,632485
Durbin-Watson statistic = 1,16044
Lag 1 residual autocorrelation = 0,361405

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 6 independent
variables. The equation of the fitted model is

q = -0,18674*PH + 0,394501*C - 0,00105842*T*C

11
 Plot of q Residual Plot

8,9 2,5

6,9 1,5

Studentized residual
observed
4,9 0,5

2,9 -0,5

0,9 -1,5

-1,1 -2,5
-1,1 0,9 2,9 4,9 6,9 8,9 0 2 4 6 8
predicted predicted q

Backward stepwise selection

Avec constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT -1,55656 0,434649 -3,58118 0,0030
C 0,396562 0,0785613 5,0478 0,0002
T*C -0,00105842 0,000246214 -4,29878 0,0007

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 69,8692 2 34,9346 52,48 0,0000
Residual 9,31873 14 0,665624
Total (Corr.) 79,1879 16

R-squared = 88,2321 percent

R-squared (adjusted for d.f.) = 86,551 percent
Standard Error of Est. = 0,815858
Mean absolute error = 0,600797
Durbin-Watson statistic = 1,21842 (P=0,0579)
Lag 1 residual autocorrelation = 0,352375

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 6 independent
variables. The equation of the fitted model is

q = -1,55656 + 0,396562*C - 0,00105842*T*C

Plot of q Residual Plot

9,3 3

2
7,3
Studentized residual

1
observed

5,3
0
3,3
-1

1,3
-2

-0,7 -3
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

12
Sans constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
T -0,00482825 0,00138149 -3,49496 0,0036
C 0,377404 0,0792588 4,76167 0,0003
T*C -0,000999058 0,000249642 -4,00197 0,0013

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 140,571 3 46,8569 68,79 0,0000
Residual 9,53562 14 0,681116
Total 150,106 17

R-squared = 93,6474 percent

R-squared (adjusted for d.f.) = 92,7399 percent
Standard Error of Est. = 0,825297
Mean absolute error = 0,599861
Durbin-Watson statistic = 1,2628
Lag 1 residual autocorrelation = 0,330959

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 6 independent
variables. The equation of the fitted model is

q = -0,00482825*T + 0,377404*C - 0,000999058*T*C

Plot of q Residual Plot

9,2 3

2
7,2
Studentized residual

1
observed

5,2
0
3,2
-1

1,2
-2

-0,8 -3
-0,8 1,2 3,2 5,2 7,2 9,2 0 2 4 6 8
predicted predicted q

Effet quadratique
Least square
Avec constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH

13
 T*C
PH*C
T*T
PH*PH
C*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT -7,52663 115,313 -0,065271 0,9498
T 0,0455029 0,729152 0,0624052 0,9520
PH -1,816 1,41145 -1,28662 0,2391
C 0,521385 0,0821914 6,34354 0,0004
T*PH -0,000232057 0,00397637 -0,058359 0,9551
T*C -0,00149327 0,000248523 -6,00857 0,0005
PH*C -0,00327723 0,00149114 -2,1978 0,0639
T*T -0,0000131997 0,00114518 -0,0115263 0,9911
PH*PH 0,131043 0,0412265 3,17862 0,0155
C*C 0,000316975 0,000161041 1,96829 0,0897

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 77,9427 9 8,6603 48,69 0,0000
Residual 1,24516 7 0,177879
Total (Corr.) 79,1879 16

R-squared = 98,4276 percent

R-squared (adjusted for d.f.) = 96,4059 percent
Standard Error of Est. = 0,421757
Mean absolute error = 0,205582
Durbin-Watson statistic = 3,03713 (P=0,9350)
Lag 1 residual autocorrelation = -0,520757

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 9 independent
variables. The equation of the fitted model is

q = -7,52663 + 0,0455029*T - 1,816*PH + 0,521385*C - 0,000232057*T*PH - 0,00149327*T*C -

0,00327723*PH*C -0,0000131997*T*T + 0,131043*PH*PH + 0,000316975*C*C

Plot of q Residual Plot

9,3 4

7,3
Studentized residual

2
observed

5,3
0
3,3

-2
1,3

-0,7 -4
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

Sans constante
Dependent variable: q

Independent variables:
T
PH
C
T*PH
T*C
PH*C
T*T
PH*PH
C*C
Number of observations: 17

14
 Standard T
Parameter Estimate Error Statistic P-Value
T -0,00202747 0,0348579 -0,0581637 0,9550
PH -1,80979 1,31769 -1,37346 0,2069
C 0,521601 0,0768439 6,7878 0,0001
T*PH -0,000210711 0,00370808 -0,056825 0,9561
T*C -0,0014926 0,000232346 -6,42404 0,0002
PH*C -0,00327666 0,00139523 -2,34847 0,0468
T*T 0,0000611966 0,000103707 0,590092 0,5714
PH*PH 0,130174 0,0365067 3,56575 0,0073
C*C 0,000313367 0,000141528 2,21416 0,0577

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 148,86 9 16,54 106,20 0,0000
Residual 1,24591 8 0,155739
Total 150,106 17

R-squared = 99,17 percent

R-squared (adjusted for d.f.) = 98,34 percent
Standard Error of Est. = 0,394638
Mean absolute error = 0,205043
Durbin-Watson statistic = 3,05291
Lag 1 residual autocorrelation = -0,528597

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 9 independent
variables. The equation of the fitted model is

q = -0,00202747*T - 1,80979*PH + 0,521601*C - 0,000210711*T*PH - 0,0014926*T*C - 0,00327666*PH*C +

0,0000611966*T*T + 0,130174*PH*PH + 0,000313367*C*C

Plot of q Residual Plot

9,3 4

7,3
Studentized residual

2
observed

5,3
0
3,3

-2
1,3

-0,7 -4
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

Forward stepwise selection

Avec constant
Dependent variable: q

Independent variables:
T
PH
C
T*PH
T*C
PH*C
T*T
PH*PH
C*C
Number of observations: 17

15
 Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT 17,0014 6,09342 2,79013 0,0145
T -0,0542286 0,019131 -2,83459 0,0132
C*C 0,000503339 0,0000584619 8,60968 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 67,6591 2 33,8295 41,08 0,0000
Residual 11,5288 14 0,823488
Total (Corr.) 79,1879 16

R-squared = 85,4412 percent

R-squared (adjusted for d.f.) = 83,3613 percent
Standard Error of Est. = 0,907462
Mean absolute error = 0,668486
Durbin-Watson statistic = 2,7527 (P=0,9670)
Lag 1 residual autocorrelation = -0,405786

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 9 independent
variables. The equation of the fitted model is

q = 17,0014 - 0,0542286*T + 0,000503339*C*C

Plot of q Residual Plot

9,1 4

7,1
Studentized residual

2
observed

5,1
0
3,1

-2
1,1

-0,9 -4
-0,9 1,1 3,1 5,1 7,1 9,1 0 2 4 6 8
predicted predicted q

Sans constant
Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
T*T
PH*PH
C*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
C*C 0,000471587 0,0000442782 10,6506 0,0000

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 131,551 1 131,551 113,43 0,0000
Residual 18,5553 16 1,15971
Total 150,106 17

R-squared = 87,6385 percent

16
R-squared (adjusted for d.f.) = 87,6385 percent
Standard Error of Est. = 1,0769
Mean absolute error = 0,728186
Durbin-Watson statistic = 1,96978
Lag 1 residual autocorrelation = 0,00778721

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 9 independent
variables. The equation of the fitted model is

q = 0,000471587*C*C

Plot of q Residual Plot

8 5

Studentized residual
6
observed

1
4
-1

2
-3

0 -5
0 2 4 6 8 0 2 4 6 8
predicted predicted q

Backward stepwise selection

Avec constant

Dependent variable: q
Independent variables:
T
PH
C
T*PH
T*C
PH*C
T*T
PH*PH
C*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
CONSTANT -5,65028 5,18994 -1,0887 0,3046
T 0,0353675 0,0153149 2,30935 0,0463
PH -1,88737 0,521252 -3,62085 0,0056
C 0,521461 0,0722728 7,21517 0,0001
T*C -0,00149327 0,000219232 -6,81136 0,0001
PH*C -0,00327723 0,00131539 -2,49145 0,0343
PH*PH 0,130882 0,0342059 3,82629 0,0041
C*C 0,000316345 0,000133617 2,36755 0,0421

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 77,9421 7 11,1346 80,44 0,0000
Residual 1,24578 9 0,138421
Total (Corr.) 79,1879 16

R-squared = 98,4268 percent

R-squared (adjusted for d.f.) = 97,2032 percent
Standard Error of Est. = 0,372049
Mean absolute error = 0,205483

17
Durbin-Watson statistic = 3,04771 (P=0,9845)
Lag 1 residual autocorrelation = -0,526309

The StatAdvisor
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 9 independent
variables. The equation of the fitted model is

q = -5,65028 + 0,0353675*T - 1,88737*PH + 0,521461*C - 0,00149327*T*C - 0,00327723*PH*C +

0,130882*PH*PH + 0,000316345*C*C

Plot of q Residual Plot

9,3 4

7,3

Studentized residual
2
observed

5,3
0
3,3

-2
1,3

-0,7 -4
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

Sans constante
Dependent variable: q

Independent variables:
T
PH
C
T*PH
T*C
PH*C
T*T
PH*PH
C*C
Number of observations: 17

Standard T
Parameter Estimate Error Statistic P-Value
PH -1,88431 0,406548 -4,63491 0,0009
C 0,52054 0,0517819 10,0526 0,0000
T*C -0,00148931 0,000149838 -9,93945 0,0000
PH*C -0,00327108 0,00122437 -2,67164 0,0234
T*T 0,0000551033 0,0000134728 4,08996 0,0022
PH*PH 0,130644 0,0273596 4,77508 0,0008
C*C 0,000313112 0,000124722 2,51049 0,0309

Analysis of Variance
Source Sum of Squares Df Mean Square F-Ratio P-Value
Model 148,86 7 21,2657 170,61 0,0000
Residual 1,24648 10 0,124648
Total 150,106 17

R-squared = 99,1696 percent

R-squared (adjusted for d.f.) = 98,6714 percent
Standard Error of Est. = 0,353056
Mean absolute error = 0,205267
Durbin-Watson statistic = 3,06056
Lag 1 residual autocorrelation = -0,532849

The StatAdvisor

18
The output shows the results of fitting a multiple linear regression model to describe the relationship between q
and 9 independent
variables. The equation of the fitted model is

q = -1,88431*PH + 0,52054*C - 0,00148931*T*C - 0,00327108*PH*C + 0,0000551033*T*T + 0,130644*PH*PH

+0,000313112*C*C

Plot of q Residual Plot

9,3 4

7,3

Studentized residual
2
observed

5,3
0
3,3

-2
1,3

-0,7 -4
-0,7 1,3 3,3 5,3 7,3 9,3 0 2 4 6 8
predicted predicted q

Analyse des résultats

Récapitulatif des résultats obtenus:

Modèle P-value R-square Mea

1 0.000 82.206 0.747661
2 0.000 86,636 0,836892
3 0.000 81,0546 0,79122
4 0.000 86,3802 0,838633
5 0.000 81,0546 0,79122
6 0.000 86,3802 0,838633
7 0,0001 91,4015 0,568811
8 0.000 95,2741 0,56896
9 0.000 88,2321 0,600797
10 0.000 94,161 0,632485
11 0.000 88,2321 0,600797
12 0.000 93,6474 0,599861
13 0.000 98,4276 0,205582
14 0.000 99,17 0,205043
15 0.000 85,4412 0,668486
16 0.000 87,6385 0,728186
17 0.000 98,4268 0,205483
18 0.000 99,1696 0,205267

 Le critère P-value ne nous donne aucun renseignement en ce qui concerne la comparaison

entre la signification des modèles, puisque pour tous les modèles on a P<0.05.

 D’après l’analyse des « R-sqaured »et des « Meanabsoluteerror » on voit que le modèle 14
est le plus significatif du point de vue estimation des paramètres et description du procède,
c’est le modèle qui explique mieux la variabilité des paramètres par l’intermédiaire de la
régression et qui nous donne les valeurs les plus proches à celle obtenu par l’expérience.

19
 C’est le modèle moindre carrées avec le main effect, les interactions, les carrées et sans
constante

q = -0,00202747*T - 1,80979*PH + 0,521601*C - 0,000210711*T*PH - 0,0014926*T*C - 0,00327666*PH*C + 0,0000611966*T*T

+0,130174*PH*PH + 0,000313367*C*C

 On trouve aussi que les modèles 13,17 et 18 offrent aussi une très bonne régression vis-à-vis
les procédés vu leurs « R-squared » qui atteignent les 98% et leurs « MEA » qui sont aussi
faibles que celle du modèle 14.

Modèle 13: moindre carrées avec le main effect, les interactions, les carrées et constante :

q = -7,52663 + 0,0455029*T - 1,816*PH + 0,521385*C - 0,000232057*T*PH - 0,00149327*T*C - 0,00327723*PH*C -

0,0000131997*T*T + 0,131043*PH*PH + 0,000316975*C*C

Modele 17 : backwardsstepwiseselection avec le main effect, les interactions, les carrées et constante:

q = -5,65028 + 0,0353675*T - 1,88737*PH + 0,521461*C - 0,00149327*T*C - 0,00327723*PH*C + 0,130882*PH*PH +

0,000316345*C*C

Modele 18 : backwardsstepwiseselection avec le main effect, les interactions, les carrées et sans constante:

q = -1,88431*PH + 0,52054*C - 0,00148931*T*C - 0,00327108*PH*C + 0,0000551033*T*T + 0,130644*PH*PH +

0,000313112*C*C

 D’après le tableau récapitulatif on peut constater que les meilleurs modèles sont ceux qui
incluent la main effect, les interactions et les carrées
 Dans ce cas on peut appliquer le test de Fisher pour un coefficient de confiance α = 0,95
pour comparer entre ces modèles :

Comparons entre le modèle 14 et le modèle 13:

On a F (0.95, 3 , 4)=6.591382

𝛔𝟐𝟏𝟑
Le rapport de variance donne: 1.00526 < 6.591382.
𝛔𝟐𝟏𝟒

 Donc le test de Fisher est validé et on prend le modèle le plus simple avec le plus grand
résidu étant le modèle 14.

Comparons entre le modèle 17 et le modèle 14:

On a F (0.95, 4,9)=3.633088

𝛔𝟐𝟏𝟒
Le rapport de variance donne: 0.997858<3.633088.
𝛔𝟐𝟏𝟕

 Donc le test de Fisher est validé et on prend le modèle le plus simple étant le modèle 13.

Comparons entre le modèle 18 et le modèle 14:

On a F (0.95 , 4 ,9)=3.633088

20
 𝛔𝟐𝟏𝟒
Le rapport de variance donne:0.998908<3.633088.
𝛔𝟐𝟏𝟖

 Donc le test de Fisher est validé et on prend le modèle le plus simple étant le modèle 13.

D’après l’analyse des résidus, des erreurs et le test de Fisher on peut déduire que le modèle le
plus significatif au procédé est le modèle 14, c’est :

le modèle moindre carrées avec le main effect, les interactions, les carrées et sans constante.

En négligeant les variables non significatives on obtient:

q = 0,521601*C - 0,0014926*T*C - 0,00327666*PH*C+ 0,130174*PH*PH

21