Michel Tenenhaus
Variable
expliquer
Y
X1, X2, , Xk
Quantitatives
Qualitatives
Mlange
Quantitatif
Rgression multiple
Analyse de la variance
Qualitatif
- Rgression
Logistique
- Segmentation
- Analyse factorielle
discriminante
- Analyse discriminante
bayesienne
- Rgression
Logistique
- Segmentation
- Analyse factorielle
discriminante
- Rgression
Logistique
- Segmentation
- Analyse factorielle
discriminante
Rseaux de
neurones :
Optimiser la prvision
pour les modle nonlinaires (!!!!)
Plan du cours
Rgression logistique binaire simple (chd)
Rgression logistique binaire multiple
Utilisation
- Donnes individuelles (faillite, bb)
de SPSS
et de la
- Donnes agrges (job satisfaction)
Proc Logistic
Rgression logistique ordinale (bordeaux)
- pentes gales
- partiellement pentes gales (Proc Genmod)
Rgression logistique multinomiale (bordeaux, alligator)
- utilisation de SPSS et de la Proc Catmod
3
Rfrences
Agresti, A. (1990):
Categorical Data Analysis, New York: John Wiley & Sons, Inc.
Hosmer, D.W. & Lemeshow, S. (1989):
Applied Logistic Regression, New York: John Wiley &
Sons, Inc.
P. McCullagh & J.A. Nelder (1989):
Generalized Linear Models, Chapman & Hall, London.
Collet D. (1999):
Modelling binary data, Chapman & Hall/CRC, Londres
P. Allison (1999):
Logistic Regression Using the SAS System: Theory and Applications
Cary, NC: SAS Institute Inc.
Tenenhaus M. (2007):
Statistique, Dunod
AGRP
1
1
1
1
1
AGE
20
23
24
25
25
CHD
0
0
0
0
1
97
98
99
100
8
8
8
8
64
64
65
69
0
1
1
1
1.0
.8
.6
.4
.2
CHD
0.0
-.2
10
20
30
40
50
60
70
AGE
CHD
present
1
2
3
5
6
5
13
8
43
Mean
(Proportion)
0.10
0.13
0.25
0.33
0.46
0.63
0.76
0.80
0.43
1.0
.8
.6
.4
Proportion (CHD)
Age Group
20 29
30 34
35 39
40 44
45 49
50 54
55 - 59
60 - 69
Total
CHD
absent
9
13
9
10
7
3
4
2
57
.2
0.0
1
AGEGRP
Le modle logistique
0 1x
e
( x )
0 1x
1 e
.8
ou
.6
( x )
Log(
) 0 1x
1 ( x )
Prob(Y=1 / X)
.4
.2
0.0
10
20
30
40
50
60
AGE
70
Fonctions de lien
Fonction logit
g(p) = log(p / (1 - p))
Fonction normit ou probit
g(p) = -1(p)
o est la fonction de rpartition de la loi normale rduite
xi
xn
Y
y1
yi
yn
Le modle
( x i ) P ( Y 1 / X x i )
e0 1x i
0 1x i
1 e
yi = 1 si caractre prsent,
0 sinon
12
Prob( Y yi / X x i )
i 1
n
( x i ) yi (1 ( x i ))1 yi
i 1
0 1 xi
e
1
yi
1 yi
(
)
(
)
0 1 xi
0 1 xi
1
e
1
e
i 1
l (0 , 1 )
13
Log-Vraisemblance
n
1 yi
yi
i 1
( xi )
yi Log (
) Log (1 ( xi ))
1 ( xi )
i 1
n
yi (0 1 xi ) Log (1 exp(0 1 xi ))
i 1
14
)
, )
V
(
Cov
(
0
0 1
V( )
V(1 )
Cov(0 , 1 )
L( )
'
15
Rsultats
Model Summary
Step
1
-2 Log
likelihood
107.353
Nagelkerke
R Square
.341
B
.111
-5.309
AGE
Constant
S.E.
.024
1.134
Wald
21.254
21.935
df
1
1
Sig.
.000
.000
Exp(B)
1.117
.005
Model
Chi-square
29.310
df
1
Sig.
.000
16
Rsultats
Estimated Covariance Matrix
Variable
Intercept
age
Intercept
age
1.285173
-0.02668
-0.02668
0.000579
Test de Wald
Le modle
Test
e0 1x
( x ) P ( Y 1 / X x )
1 e0 1x
H0 : 1 = 0
H1 : 1 0
Statistique utilise
12
Wald 2
s1
ou
NS = P(2(1)Wald)
18
Test LRT
Le modle
Test
e0 1x
( x ) P ( Y 1 / X x )
1 e0 1x
H0 : 1 = 0
H1 : 1 0
Statistique utilise
-2 L(Cste) - -2 L(Cste, X )
ou NS = P( 2(1) )
19
1 e
e0 1x
(x)
1 e0 1x
0 1x 1.96 Var( 0 1x )
1 e
0 1x 1.96 Var ( 0 1x )
20
Case Summariesa
Intervalle de
confiance de (x)
au niveau 95%
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Total
AGE
20.0
23.0
24.0
25.0
25.0
26.0
26.0
28.0
28.0
29.0
30.0
30.0
30.0
30.0
30.0
30.0
32.0
32.0
33.0
33.0
34.0
34.0
34.0
34.0
34.0
35.0
35.0
36.0
36.0
36.0
30
PROBABILITE
CALCULE
.04
.06
.07
.07
.07
.08
.08
.10
.10
.11
.12
.12
.12
.12
.12
.12
.15
.15
.16
.16
.18
.18
.18
.18
.18
.19
.19
.21
.21
.21
30
INF95
.01
.02
.02
.03
.03
.03
.03
.04
.04
.05
.05
.05
.05
.05
.05
.05
.07
.07
.08
.08
.09
.09
.09
.09
.09
.11
.11
.12
.12
.12
30
SUP95
.14
.17
.18
.19
.19
.20
.20
.23
.23
.24
.25
.25
.25
.25
.25
.25
.28
.28
.29
.29
.31
.31
.31
.31
.31
.33
.33
.34
.34
.34
30
21
.8
.6
.4
.2
SUP95
INF95
0.0
20
PROBABILITE
30
40
50
60
70
22
Proportion observe :
yi / nClasse
iClasse
AGEGRP
1
2
3
Proportion thorique :
4
5
i / nClasse
iClasse
6
7
puisque E(yi) = i
estim par
Total
Mean
N
Mean
N
Mean
N
Mean
N
Mean
N
Mean
N
Mean
N
Mean
N
Mean
N
Maladie
cardiaque
.1000
10
.1333
15
.2500
12
.3333
15
.4615
13
.6250
8
.7647
17
.8000
10
.4300
100
Predicted
probability
.0787086
10
.1484562
15
.2299070
12
.3519639
15
.4824845
13
.6087623
8
.7302152
17
.8391673
10
.4300000
100
23
.8
.6
Proportion
.4
.2
Prop. observe
0.0
Prop. thorique
1
Classe d'age
24
10
iClasse
Effectif
2
Thorique
Observ
0
1
Groupe
10
Nb de degrs de libert
= Nb de groupes - 2
25
Step
1
1
2
3
4
5
6
7
8
9
10
Maladie cardiaque =
chd=no
Observed
Expected
9
9.213
9
8.657
8
8.095
8
8.037
7
6.947
5
5.322
5
4.200
3
3.736
2
2.134
1
.661
Maladie cardiaque =
chd=yes
Observed
Expected
1
.787
1
1.343
2
1.905
3
2.963
4
4.053
5
4.678
5
5.800
10
9.264
8
7.866
4
4.339
Total
10
10
10
11
11
10
10
13
10
5
Chi-square
.890
df
8
Sig.
.999
26
Pseudo R2 (McFadden)
l (cte) n2
R 1[
]
l (cte, X )
2
Max R 1 [l (cte)]
2
Pseudo R 2 1 [
2 L(cte, X )
]
2 L(cte)
2
n
R2 ajust de Nagelkerke
2
R
2
R adj
2
R max
27
Tableau de classification
Une observation i est affecte la classe [Y=1] si
i c.
Tableau de classification (c = 0.5)
TABLE OF CHD BY PREDICTS
CHD
PREDICTS
Frequency
0
1
0
45
12
1
14
29
Total
59
41
Total
57
43
Sensibilit = 29/43
Spcificit = 45/57
taux de faux positifs = 12/41
taux de faux ngatifs = 14/59
100
28
Objectifs
Sensibilit = capacit diagnostiquer les malades
parmi les malades
Spcificit = capacit reconnatre les non-malades
parmi les non-malades
1 - Spcificit = risque de diagnostiquer un malade
chez les non-malades.
Graphique ROC
(Receiver Operating Characteristic)
Coronary Hearth Disease ROC curve
1.0
C = 0.5
.8
.6
.4
Sensitivit
Sensibilit : capacit
prdire un vnement
Spcificit : capacit
prdire un non-vnement
Graphique ROC :
y = Sensibilit(c)
x = 1 - Spcificit (c)
.2
0.0
0.0
.2
.4
.6
.8
1.0
1 - Spcificit
i j )
nd = nombre de paires
discordantes (yi < yj et
i j )
t - nc - nd = Nb dex-aequo
i j )
(yi < yj et
ri
yi i
i (1 i )
32
D 2log l
di2
signe( y i i ) Di
33
3
16
-1
-2
-3
N=
100
100
100
Deviance value
34
Case Summariesa
Les donnes
des entreprises
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
cash flow /
total debt
-.45
-.56
.06
-.07
-.10
-.14
.04
-.07
.07
-.14
-.23
.07
.01
-.28
.15
.37
-.08
.05
.01
.12
-.28
.51
.08
.38
.19
.32
.31
.12
-.02
.22
.17
.15
-.10
.14
.14
.15
.16
.29
.54
-.33
.48
.56
.20
.47
.17
.58
a. Limited to first 100 cases.
net income /
total assets
-.41
-.31
.02
-.09
-.09
-.07
.01
-.06
-.01
-.14
-.30
.02
.00
-.23
.05
.11
-.08
.03
.00
.11
-.27
.10
.02
.11
.05
.07
.05
.05
.02
.08
.07
.05
-.01
-.03
.07
.06
.05
.06
.11
-.09
.09
.11
.08
.14
.04
.04
current
assets /
current
liabilities
1.09
1.51
1.01
1.45
1.56
.71
1.50
1.37
1.37
1.42
.33
1.31
2.15
1.19
1.88
1.99
1.51
1.68
1.26
1.14
1.27
2.49
2.01
3.27
2.25
4.24
4.45
2.52
2.05
2.35
1.80
2.17
2.50
.46
2.61
2.23
2.31
1.84
2.33
3.01
1.24
4.29
1.99
2.92
2.45
5.06
current
assets /
net sales
.45
.16
.40
.26
.67
.28
.71
.40
.34
.43
.18
.25
.70
.66
.27
.38
.42
.95
.60
.17
.51
.54
.53
.35
.33
.63
.69
.69
.35
.40
.52
.55
.58
.26
.52
.56
.20
.38
.48
.47
.18
.44
.30
.45
.14
.13
FAILLITE
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
NF
36
.2
.6
.1
0.0
40
-.1
0.0
-.2
-.2
40
-.4
-.6
-.8
N=
25
21
Non faillite
Faillite
-.3
-.4
-.5
N=
FAILLITE
25
21
Non faillite
Faillite
FAILLITE
1.2
46
1.0
27
42
26
.8
13
34
0
N=
FAILLITE
6
11
25
21
Non faillite
Faillite
34
.2
.4
.6
.4
.2
0.0
N=
FAILLITE
25
21
Non faillite
Faillite
37
.1
.3
.2
0. 0
.1
-.1
95% CI X2
95% CI X1
0. 0
-.1
-.2
N=
25
21
Non
Oui
-.2
N=
FAILLITE
25
21
Non
Oui
FAILLITE
3. 5
.6
3. 0
.5
2. 5
2. 0
1. 5
95% CI X4
95% CI X3
.4
1. 0
N=
FAILLITE
25
21
Non
Oui
.3
N=
25
21
Non
Oui
FAILLITE
38
Variable
Coefficient 1
WALD
NS
X1
X2
X3
X4
-7.526
-19.493
-3.382
.354
9.824
8.539
11.75
.040
.002
.003
.001
.841
R2 de
Nagelkerke
.466
.466
.611
.001
40
41
e0 1x1 ...4 x 4
(x) P(Y F / X x)
0 1x1 ...4 x 4
1 e
42
Prob( Y yi / X x i )
i 1
n
( x i ) yi (1 ( x i ))1 yi
i 1
n
(
i 1
1 e
jx j
j
jx j
j
) yi (1
1 e
jx j
j
jx j
)1 yi
l (0 , 1 ,..., 4 )
43
Rsultats
Model Summary
Step
1
-2 Log
likelihood
27.443
Nagelkerke
R Square
.725
X1
X2
X3
X4
Collinearity Statistics
Tolerance
VIF
0.212
4.725
0.252
3.973
0.635
1.575
0.904
1.106
X1
X2
X3
X4
Constant
B
-7.138
3.703
-3.415
2.968
5.320
S.E.
6.002
13.670
1.204
3.065
2.366
Wald
1.414
.073
8.049
.938
5.053
df
1
1
1
1
1
Sig.
.234
.786
.005
.333
.025
Exp(B)
.001
40.581
.033
19.461
204.283
44
Rsultats
Correlations
X1
X1
X2
X3
X4
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
1
.
46
.858**
.000
46
.571**
.000
46
-.053
.725
46
X2
.858**
.000
46
1
.
46
.471**
.001
46
.055
.717
46
X3
.571**
.000
46
.471**
.001
46
1
.
46
.154
.306
46
X4
-.053
.725
46
.055
.717
46
.154
.306
46
1
.
46
45
Le modle estim
Pr ob(Y F / X)
e5.3207.138X1 3.703X 2 3.415X3 2.968X 4
5.320 7.138X1 3.703X 2 3.415X 3 2.968X 4
1 e
Prvision de faillite
Classification Tablea
Predicted
FAILLITE
Observed
SITUATION
Overall Percentage
NF
NF
F
F
24
3
1
18
Percentage
Correct
96.0
85.7
91.3
46
Chi-square
5.201
df
7
Sig.
.636
Step
1
1
2
3
4
5
6
7
8
9
FAILLITE = Non
Observed
Expected
5
4.999
5
4.906
4
4.613
5
4.143
4
3.473
1
1.762
0
.667
1
.340
0
.098
FAILLITE = Oui
Observed
Expected
0
.001
0
.094
1
.387
0
.857
1
1.527
4
3.238
5
4.333
4
4.660
6
5.902
Total
5
5
5
5
5
5
5
5
6
47
Model Summary
Step
1
-2 Log
likelihood
27.516
Nagelkerke
R Square
.724
X1
X3
X4
Constant
B
-5.772
-3.289
2.979
5.038
S.E.
3.005
1.085
3.025
2.060
Wald
3.690
9.183
.970
5.983
df
1
1
1
1
Sig.
.055
.002
.325
.014
Exp(B)
.003
.037
19.675
154.193
48
-2 Log
likelihood
28.636
Nagelkerke
R Square
.709
X1
X3
Constant
B
-6.556
-3.019
5.940
S.E.
2.905
1.002
1.986
Wald
5.092
9.077
8.950
df
1
1
1
Sig.
.024
.003
.003
Exp(B)
.001
.049
379.996
49
27
26
42
4
24
40
44
35
28 45
372530
36
32
13
29
23
43
1531
18
517
7
4
10
8
9
19 12
20
3
33
2
2
21
14
22
39
38
16
41
FAILLITE
34
11
NF
-.6
-.4
-.2
-.0
.2
.4
.6
50
e
Pr ob(Y F / X)
0.5
5.940 6.556X1 3.019X 2
1 e
X3 = (5.940 - 6.556X1)/3.019
51
27
26
42
X3 = (5.940 - 6.556X1)/3.019
24
40
44
33
29
13
2
5 17
48
10
2
21
14
19
35
28 45
37 2530
36
32
23
43
1531
18
7
9
12
3
-.4
-.2
Droite
diso-probabilit 0.5
34
-.0
16
41
0
-.6
38
20
6
11
22
39
.2
.4
.6
52
53
54
Exemple II :
Low birth weight baby (Hosmer & Lemeshow)
Y = 1 si le poids du bb < 2 500 grammes,
= 0 sinon
n1 = 59, n0 = 130
Facteurs de risque :
-
Age
LWT (Last Menstrual Period Weight)
Race (White, Black, Other)
FTV (Nb of First Trimester Physician Visits)
Smoke (1 = oui, 0 = non)
55
Model Summary
Step
1
-2 Log
likelihood
214.575
Rsultats
Nagelkerke
R Square
.142
B
-.022
-.012
-.941
.289
-.008
1.053
1.269
AGE
LWT
WHITE
BLACK
FTV
SMOKE
Constant
S.E.
.035
.006
.418
.527
.164
.381
1.023
Wald
.410
3.762
5.070
.301
.002
7.637
1.539
df
1
1
1
1
1
1
1
Sig.
.522
.052
.024
.583
.962
.006
.215
Exp(B)
.978
.988
.390
1.336
.992
2.866
3.558
Model
1
AGE
weight last
menstrual period
smoking during
pregnancy
n physician visits
first trimester
WHITE
BLACK
Collinearity Statistics
Tolerance
VIF
.884
1.132
.869
1.150
.865
1.156
.939
1.065
.686
.743
1.457
1.346
Aucun problme
de multicolinarit
56
Validit du modle
Test de Hosmer et Lemeshow
Hosmer and Lemeshow Test
Step
1
Chi-square
11.825
df
8
Sig.
.159
Step
1
1
2
3
4
5
6
7
8
9
10
Total
19
19
19
19
19
19
19
19
19
18
57
Odds-Ratio
Odds Ratio(Smoke)
exp(Smoke )
Var(Smoke ) sSmoke
[e
Smoke 1.96sSmoke
, e
Smoke 1.96sSmoke
]
59
Step
a
1
AGE
LWT
WHITE
BLACK
FTV
SMOKE
Constant
B
-.022
-.012
-.941
.289
-.008
1.053
1.269
Exp(B)
.978
.988
.390
1.336
.992
2.866
3.558
60
e0 1x1 ...k x k
( x ) P ( Y 1 / X x )
1 e0 1x1 ...k x k
- Proc GENMOD (type 3)
- Rgression backwardLR
avec Removal = 1 dans
SPSS
H0 : r+1 = = k = 0
H1 : au moins un j 0
Statistiques utilises
2.
Wald ( r 1 ,..., k )
r 1
Var
k
r 1
- Proc Logistic
- Proc Genmod
(type 3 et wald)
- SPSS
61
Rgle de dcision
On rejette
H0 : r+1 = = k = 0
au risque de se tromper si
ou Wald
2
1
k r
ou si
NS = Prob( k r Wald ou )
2
62
AGE
LWT
RACE
RACE(1)
RACE(2)
SMOKE
FTV
Constant
B
-.022
-.012
S.E.
.035
.006
-.941
.289
1.053
-.008
1.269
.418
.527
.381
.164
1.023
Wald
.410
3.762
7.784
5.070
.301
7.637
.002
1.539
df
1
1
2
1
1
1
1
1
Sig.
.522
.052
.020
.024
.583
.006
.962
.215
Exp(B)
.978
.988
.390
1.336
2.866
.992
3.558
-2 Log
likelihood
222.815
Nagelkerke
R Square
.086
63
Effect
Intercept
AGE
LWT
SMOKE
FTV
RACE
-2 Log
Likelihood of
Reduced
Model
214.575a
214.990
218.746
222.573
214.577
222.815
Chi-Square
.000
.415
4.171
7.998
.002
8.239
df
Sig.
0
1
1
1
1
2
.
.520
.041
.005
.963
.016
64
e0 1x1 ...k x k
( x ) P ( Y 1 / X x )
1 e0 1x1 ...k x k
H0 : C(0, 1, k) = 0
H1 : C(0, 1, k) 0
Statistiques utilises
1.
= [-2L(H0)] - [-2L(H1)]
2.
0
CVar ( M )C '
Proc GENMOD
1
Proc Logistic
Proc Genmod
65
Rgle de dcision
On rejette
H0 : C(0, 1, , k) = 0
au risque de se tromper si
2
1 rang ( L)
ou Wald
ou si
2
66
68
Prob( Y 1 / x )
Test H0 :
j = 0
vs
0 1x1 t x t jx j
0 1x1 t x t jx j
1 e
H 1 : j 0
Statistique
L
2Score
'
H 0
L
2
2
H 0
H0
L
est calcul sur le modle t+1 variables.
69
e0 1x1t x t t 1x t 1...k x k
Prob( Y 1 / x )
1 e0 1x1t x t t 1x t 1...k x k
Test H0 : t+1 == k = 0 vs H1 : au moins un j 0
Statistique
'
L
L
L
2Score
'
k-t
H 0.
de libert
sous
degr
suit une loi du khi-deux
H0
H0
H 0
71
Job satisfaction (Y/N) by sex (M/F), race, age, and region of residence
for employees of a large U.S. corporation
Under 35
M
F
White
35-44
M
F
Over 44
M
F
288
177
60
57
224
166
35
19
337
172
70
30
38
33
19
35
32
11
22
20
21
8
15
10
90
45
19
12
96
42
12
5
124
39
17
2
18
6
13
7
7
2
0
3
9
2
1
1
Southern
Y
N
226
128
88
57
189
117
44
34
156
73
70
25
45
31
47
35
18
3
13
7
11
2
9
2
Midwest
Y
N
285
179
110
93
225
141
53
24
324
140
60
47
40
25
66
56
19
11
25
19
22
2
11
12
Northwest
Y
N
270
180
176
151
215
108
80
40
269
136
110
40
36
20
25
16
9
7
11
5
16
3
4
5
Southwest
Y
N
252
126
97
61
162
72
47
27
199
93
62
24
69
27
45
36
14
7
8
4
14
5
2
0
Pacific
Y
N
119
58
62
33
66
20
20
10
67
21
25
10
45
16
22
15
15
10
10
8
8
6
6
2
Region
Northeast
Y
N
Mid-Atlantic
Y
N
Under 35
M
F
Nonwhite
35-44
M
F
Over 44
M
F
72
73
74
DF
Wald
Chi-Square
Pr > ChiSq
1
2
1
6
1
2
0.1007
50.7100
14.0597
37.7010
7.5641
5.9577
0.7510
<.0001
0.0002
<.0001
0.0060
0.0509
75
DF
Estimate
Standard
Error
Intercept
race
age
age
sex
region
region
region
region
region
region
race*sex
age*sex
age*sex
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0.6481
-0.0099
-0.1952
-0.0227
0.1230
-0.2192
0.2228
-0.0446
-0.1291
-0.0927
0.0704
0.0856
0.0768
-0.0342
0.0346
0.0312
0.0316
0.0375
0.0328
0.0469
0.0820
0.0527
0.0462
0.0472
0.0531
0.0311
0.0315
0.0375
0
0
1
0
0
1
2
3
4
5
0 0
0 0
1 0
Chi-Square
Pr > ChiSq
350.2297
0.1007
38.2459
0.3675
14.0597
21.8470
7.3832
0.7159
7.8133
3.8616
1.7565
7.5641
5.9428
0.8352
<.0001
0.7510
<.0001
0.5444
0.0002
<.0001
0.0066
0.3975
0.0052
0.0494
0.1851
0.0060
0.0148
0.3608
76
77
Rsultats
Estimate
Age >44
Pacific
Age>44,Homme
0.2180
0.1924
-0.0425
Standard
Wald
Error Chi-Square
0.0375
0.0751
0.0375
0.1444
0.0453
-0.1159
Pr > ChiSq
<.0001
0.0104
0.2565
78
35 .20
35 44 .02
44 .22
Homme
Femme
Northeast .22
Mid Atlantic .22
Southern .04
.12
Midwest .13
.12
Northwest .09
Southwest .07
.19
Pacific
35 .08
- .08
Non Blanc .09
- .09
35 44 .03 .03
Blanc .09
.09
44 .05 .05
Homme Femme
Homme Femme
79
Rsultats
Type III Analysis of Effects
Effect
sex
region
race(sex)
age(sex)
DF
Wald
Chi-Square
Pr > ChiSq
1
6
2
4
14.0597
37.7010
7.5710
55.4078
0.0002
<.0001
0.0227
<.0001
81
Rsultats
Analysis of Maximum Likelihood Estimates
Parameter
DF
Estimate
Standard
Error
Intercept
sex
region
region
region
region
region
region
race(sex)
race(sex)
age(sex)
age(sex)
age(sex)
age(sex)
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0.6481
0.1230
-0.2192
0.2228
-0.0446
-0.1291
-0.0927
0.0704
0.0757
-0.0956
-0.1185
-0.0570
-0.2720
0.0115
0.0346
0.0328
0.0469
0.0820
0.0527
0.0462
0.0472
0.0531
0.0422
0.0459
0.0342
0.0370
0.0530
0.0652
0
0
1
2
3
4
5
0
0
0
1
0
1
0
1
0
0
1
1
Contrast
Pacific
Age>44,Homme
Age>44,Femme
Estimate
0.1924
0.1754
0.2605
Standard
Error
0.0751
0.0367
0.0654
Chi-Square
Pr > ChiSq
350.2297
14.0597
21.8470
7.3832
0.7159
7.8133
3.8616
1.7565
3.2230
4.3244
11.9881
2.3683
26.3735
0.0313
<.0001
0.0002
<.0001
0.0066
0.3975
0.0052
0.0494
0.1851
0.0726
0.0376
0.0005
0.1238
<.0001
0.8596
Wald
Chi-Square
6.5729
22.8477
15.8719
Pr > ChiSq
0.0104
<.0001
<.0001
82
0.65
Homme .12
Femme .12
Homme
Northeast .22
Midwest .13
Northwest .09
Southwest .07 ns
.19
Pacific
.08
- .08
.10
Femme .10
Non-blanc Blanc
Diffrence entre
races par sexe :
Race(Sexe)
ns
35 3544 44
Diffrence entre
les ages par sexe :
Age(Sexe)
83
yi n i i
ri
n i i (1 i )
Rsidu dviance :
yi
ni yi
d i signe( yi y i ) 2 yi log( ) 2(ni yi ) log(
)
y i
ni y i
84
sat
total
propsat
predicted
reschi
resdev
288
465
0.61935
0.58848
1.35305
1.35864
90
135
0.66667
0.68991
-0.58388
-0.58005
226
354
0.63842
0.63003
0.32704
0.32756
285
464
0.61422
0.61011
0.18152
0.18164
270
450
0.60000
0.61875
-0.81897
-0.81651
252
378
0.66667
0.65641
0.41995
0.42097
119
177
0.67232
0.68338
-0.31638
-0.31541
60
117
0.51282
0.53231
-0.42246
-0.42216
19
31
0.61290
0.63909
-0.30364
-0.30214
86
Validation du modle
Le khi-deux de Pearson :
QP ri2
i 1
La dviance :
QL d i2
i 1
[nb de
87
Remarques
Les tests de validation sont valables sil y a
au moins 10 sujets par groupe.
La dviance QL est gale
[2 L(modle tudi)]-[-2L(modle satur)]
Rsultats
Deviance and Pearson Goodness-of-Fit Statistics
Criterion
DF
Value
Value/DF
Pr > ChiSq
Deviance
Pearson
70
70
81.9676
79.0760
1.1710
1.1297
0.1552
0.2142
89
Sur-dispersion
Khi-deux de Pearson QP et dviance QL sont trop forts si :
- Modle mal spcifi
- Outliers
Htrognit de chaque groupe
La variable de rponse Yi = Nb de succs sur le groupe i ne
suit plus une loi binomiale :
- E(Yi) = nii
-
V(Yi) = ni i (1 - i)
90
Calcul de
Dans la Proc LOGISTIC :
- Option SCALE = Pearson :
-
QP
ddl
QL
ddl
91
QUALITE DU VIN :
Les donnes
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Temprature
3064
3000
3155
3085
3245
3267
3080
2974
3038
3318
3317
3182
2998
3221
3019
3022
3094
3009
3227
3308
3212
3361
3061
3478
3126
3458
3252
3052
3270
3198
2904
3247
3083
3043
Soleil
1201
1053
1133
970
1258
1386
966
1189
1103
1310
1362
1171
1102
1424
1230
1285
1329
1210
1331
1366
1289
1444
1175
1317
1248
1508
1361
1186
1399
1259
1164
1277
1195
1208
Chaleur
10
11
19
4
36
35
13
12
14
29
25
28
9
21
16
9
11
15
21
24
17
25
12
42
11
43
26
14
24
20
6
19
5
14
Pluie
361
338
393
467
294
225
417
488
677
427
326
326
349
382
275
303
339
536
414
282
302
253
261
259
315
286
346
443
306
367
311
375
441
371
Qualit
2
3
2
3
1
1
3
3
3
2
1
3
3
1
2
2
2
3
2
1
2
1
2
1
2
1
2
3
1
1
3
1
3
3
94
Correlations
Temprature
corrlations
Soleil
Chaleur
Pluie
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Temprature
1
.
34
.712**
.000
34
.865**
.000
34
-.410*
.016
34
Soleil
.712**
.000
34
1
.
34
.646**
.000
34
-.473**
.005
34
Chaleur
.865**
.000
34
.646**
.000
34
1
.
34
-.401*
.019
34
Pluie
-.410*
.016
34
-.473**
.005
34
-.401*
.019
34
1
.
34
Coefficientsa
VIF
Model
1
Temprature
Soleil
Chaleur
Pluie
Collinearity Statistics
Tolerance
VIF
.211
4.733
.451
2.216
.248
4.031
.760
1.316
95
e i 1x1k x k
Prob( Y i / x )
i 1x1 k x k
1 e
pour i = 1, , m et avec 1 2 m
Dans SPSS :
e i 1x1 k x k
Prob(Y i / x )
1 e i 1x1 k x k
96
Proprits du modle
Modle pentes gales (proportional odds ratio)
Prob(Y i/x) / Prob(Y i/x)
e i x
x ' e( x x ')
Prob(Y i/x' ) / Prob(Y i/x' ) e i
est indpendant de i.
Lorsque j > 0, la probabilit des petites valeurs de
Y augmente avec Xj.
97
e i 1ix1kix k
Prob( Y i / x )
i 1i x1 kix k
1
e
pour i = 1,,m
Test H0 :
11 = 12 = = 1m
21 = 22 = = 2m
k1 = k2 = = km
k(m-1) contraintes
98
Statistique utilise
'
H 0
'
2
H 0
H 0
Rgle de dcision
On rejette lhypothse H0 dun modle pentes
gales au risque de se tromper si
ou si
2
Score
12 k (m 1)
NS = Prob(
Conseil dAgresti :
2
2 m( k 1) Score
)
100
Rsultats SPSS
Test of Parallel Linesa
Model
Null Hypothesis
General
-2 Log
Likelihood
26.158
22.355
Chi-Square
df
3.803
Sig.
4
.433
-2 Log
Likelihood
74.647
26.158
Chi-Square
df
48.489
Sig.
4
.000
Pseudo R-Square
Cox and Snell
Nagelkerke
McFadden
Link function: Logit.
.760
.855
.650
101
Rsultats SPSS
Modle complet
Parameter Estimates
Threshold
Location
[QUALITE = 1]
[QUALITE = 2]
TEMPERAT
SOLEIL
CHALEUR
PLUIE
Estimate
-85.50748
-80.54960
-.02427
-.01379
.08876
.02589
Std. Error
34.92140
33.96555
.01277
.00850
.11929
.01235
Wald
5.99549
5.62405
3.61247
2.63346
.55364
4.39307
df
Sig.
.014
.018
.057
.105
.457
.036
1
1
1
1
1
1
Threshold
Location
[QUALITE = 1]
[QUALITE = 2]
TEMPERAT
SOLEIL
PLUIE
Estimate
-67.44675
-62.63810
-.01717
-.01499
.02224
Std. Error
22.89023
21.78872
.00759
.00832
.01046
Wald
8.68204
8.26445
5.11905
3.24843
4.52311
df
1
1
1
1
1
Sig.
.003
.004
.024
.071
.033
102
Prvision de la
qualit du vin
avec le 2e modle
Case Summariesa
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Qualit
Moyen
Mdiocre
Moyen
Mdiocre
Bon
Bon
Mdiocre
Mdiocre
Mdiocre
Moyen
Bon
Mdiocre
Mdiocre
Bon
Moyen
Moyen
Moyen
Mdiocre
Moyen
Bon
Moyen
Bon
Moyen
Bon
Moyen
Bon
Moyen
Mdiocre
Bon
Bon
Mdiocre
Bon
Mdiocre
Mdiocre
a. Limited to first 100 cases.
Estimated
Cell
Probability for
Response
Category: 1
.01
.00
.01
.00
.64
.99
.00
.00
.00
.42
.94
.08
.00
.67
.04
.05
.13
.00
.21
.97
.58
1.00
.04
1.00
.11
1.00
.75
.00
.95
.14
.00
.29
.00
.00
Estimated
Cell
Probability for
Response
Category: 2
.48
.05
.44
.00
.35
.01
.01
.01
.00
.57
.06
.83
.08
.33
.78
.81
.82
.01
.76
.03
.42
.00
.81
.00
.83
.00
.25
.09
.05
.81
.09
.69
.17
.36
Estimated
Cell
Probability for
Response
Category: 3
.51
.95
.56
1.00
.00
.00
.99
.99
1.00
.01
.00
.09
.92
.00
.18
.15
.05
.99
.03
.00
.01
.00
.15
.00
.06
.00
.00
.91
.00
.05
.90
.02
.83
.63
Predicted
Response
Category
Mdiocre
Mdiocre
Mdiocre
Mdiocre
Bon
Bon
Mdiocre
Mdiocre
Mdiocre
Moyen
Bon
Moyen
Mdiocre
Bon
Moyen
Moyen
Moyen
Mdiocre
Moyen
Bon
Bon
Bon
Moyen
Bon
Moyen
Bon
Bon
Mdiocre
Bon
Moyen
Mdiocre
Moyen
Mdiocre
Mdiocre
103
Qualit de la prvision
Qualit * Predicted Response Category Crosstabulation
Count
Qualit
Total
Bon
Moyen
Mdiocre
Total
11
11
12
34
104
Qualit
2
2
3
3
1
1
Type
1
2
1
2
1
2
Rponse
0
1
0
0
1
1
(Y=1) faux
(Y 2) vrai
105
Le modle complet
Pr ob(Rponse 1 / Type , x )
e 1T1 2T2 1T 4 P 5T1T 8T1P
Le code SAS
Proc genmod data=bordeaux2 descending;
class type annee;
model reponse = type tempera soleil chaleur pluie
type*tempera type*soleil
type*chaleur type*pluie
/dist=bin link=logit type3 noint;
repeated subject=annee / type=unstr;
run;
107
Rsultats tape 1
The GENMOD Procedure
Criteria For Assessing Goodness Of Fit
Criterion
DF
Value
Value/DF
Deviance
Scaled Deviance
Pearson Chi-Square
Scaled Pearson X2
Log Likelihood
58
58
58
58
22.5317
22.5317
20.4541
20.4541
-11.2659
0.3885
0.3885
0.3527
0.3527
Algorithm converged.
108
Rsultats tape 1
Analysis Of GEE Parameter Estimates
Empirical Standard Error Estimates
Parameter
Standard
Estimate
Error
Intercept
type
type
tempera
soleil
chaleur
pluie
tempera*type
tempera*type
soleil*type
soleil*type
chaleur*type
chaleur*type
pluie*type
pluie*type
0.0000
-68.1364
-251.965
0.0948
0.0079
-0.8727
-0.1036
-0.0755
0.0000
0.0013
0.0000
0.8799
0.0000
0.0852
0.0000
1
2
1
2
1
2
1
2
1
2
95% Confidence
Limits
0.0000
0.0000
0.0000
29.7166 -126.380 -9.8929
82.1239 -412.925 -91.0055
0.0330
0.0300
0.1596
0.0107 -0.0130
0.0288
0.3574 -1.5732 -0.1722
0.0437 -0.1893 -0.0179
0.0358 -0.1458 -0.0053
0.0000
0.0000
0.0000
0.0144 -0.0270
0.0295
0.0000
0.0000
0.0000
0.3795
0.1360
1.6238
0.0000
0.0000
0.0000
0.0460 -0.0050
0.1753
0.0000
0.0000
0.0000
Z Pr > |Z|
.
-2.29
-3.07
2.87
0.74
-2.44
-2.37
-2.11
.
0.09
.
2.32
.
1.85
.
.
0.0219
0.0022
0.0041
0.4598
0.0146
0.0178
0.0351
.
0.9290
.
0.0204
.
0.0641
.
109
Rsultats
Score Statistics For Type 3 GEE Analysis
Source
type
tempera
soleil
chaleur
pluie
tempera*type
soleil*type
chaleur*type
pluie*type
DF
ChiSquare
Pr > ChiSq
2
1
0
2
2
2
2
2
2
7.08
4.94
.
0.00
0.02
0.04
0.27
0.00
0.00
0.0290
0.0263
.
0.9995
0.9881
0.9799
0.8734
0.9999
1.0000
110
DF
Value
Value/DF
Deviance
Scaled Deviance
Pearson Chi-Square
Scaled Pearson X2
Log Likelihood
62
62
62
62
26.2408
26.2408
26.5218
26.5218
-13.1204
0.4232
0.4232
0.4278
0.4278
Algorithm converged.
112
DF
Estimate
Standard
Error
Intercept
type
type
tempera
soleil
chaleur
pluie
0
1
1
1
1
1
1
0.0000
-86.4800
-81.5119
0.0245
0.0140
-0.0922
-0.0259
0.0000
35.0585
34.0447
0.0127
0.0085
0.1180
0.0123
1
2
0.0000
-17.7666
-14.7855
0.0495
0.0306
0.1391
-0.0019
ChiSquare
Pr > ChiSq
.
6.08
5.73
3.70
2.73
0.61
4.46
.
0.0136
0.0167
0.0543
0.0986
0.4348
0.0347
113
e i i1x1 ik x k
r 1
1 e i i1x1 ik x k
, i 1,..., r 1
i 1
Prob(Y r / x )
1
r 1
1 e i i1x1 ik x k
i 1
114
115
Test
H0 : 1j = = r-1,j = 0
H1 : au moins un ij 0
Statistique utilise
1 j
1 j
r 1, j
r 1, j
116
Rgle de dcision
On rejette
H0 : 1j = = r-1,j = 0
au risque de se tromper si
1 r 1
Wald
ou si
2
NS = Prob( r 1 Wald )
117
Test
Statistiques utilises
1. = [-2L(Modle simplifi)] - [-2L(Modle complet)]
2.
1,p 1
Var
r 1,k
1,p1
r 1, k
118
Rgle de dcision
On rejette
H0 : 1,p+1 = = r-1,k = 0
au risque de se tromper si
1 p ( r 1)
ou Wald
ou si
2
p(r 1) Wald ou )
NS = Prob(
119
-2 Log
Likelihood
74.647
22.227
Chi-Square
df
52.420
Sig.
8
.000
Pseudo R-Square
Cox and Snell
Nagelkerke
McFadden
.786
.884
.702
120
Effect
Intercept
TEMPERAT
SOLEIL
CHALEUR
PLUIE
-2 Log
Likelihood of
Reduced
Model
34.575
29.546
22.870
25.894
31.242
Chi-Square
12.348
7.319
.642
3.667
9.015
df
2
2
2
2
2
Sig.
.002
.026
.725
.160
.011
Les tests LRT sont plus justes que les tests de Wald :
meilleure approximation du niveau de signification.
121
Qualit
Bon
Moyen
Intercept
TEMPERAT
SOLEIL
CHALEUR
PLUIE
Intercept
TEMPERAT
SOLEIL
CHALEUR
PLUIE
B
-313.557
.113
.015
-.874
-.122
-249.604
.095
.007
-.890
-.105
Std. Error
230.325
.096
.024
.934
.104
225.594
.095
.022
.923
.103
Wald
1.853
1.375
.370
.876
1.387
1.224
.999
.094
.930
1.040
df
1
1
1
1
1
1
1
1
1
1
Sig.
.173
.241
.543
.349
.239
.269
.318
.759
.335
.308
Exp(B)
1.120
1.015
.417
.885
1.099
1.007
.411
.901
122
.782
.880
.694
Effect
Intercept
TEMPERAT
CHALEUR
PLUIE
-2 Log
Likelihood of
Reduced
Model
42.197
43.392
30.419
41.634
Chi-Square
19.327
20.522
7.550
18.764
df
2
2
2
2
Sig.
.000
.000
.023
.000
123
Qualit
Bon
Moyen
Intercept
TEMPERAT
CHALEUR
PLUIE
Intercept
TEMPERAT
CHALEUR
PLUIE
B
-381.353
.145
-1.161
-.151
-308.897
.121
-1.145
-.133
Std. Error
190.219
.074
.738
.080
186.257
.072
.729
.078
Wald
4.019
3.893
2.478
3.577
2.750
2.785
2.471
2.906
df
1
1
1
1
1
1
1
1
Sig.
.045
.048
.115
.059
.097
.095
.116
.088
Exp(B)
1.156
.313
.860
1.129
.318
.875
124
Case Summariesa
Prvision de la
qualit du vin
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Qualit
Moyen
Mdiocre
Moyen
Mdiocre
Bon
Bon
Mdiocre
Mdiocre
Mdiocre
Moyen
Bon
Mdiocre
Mdiocre
Bon
Moyen
Moyen
Moyen
Mdiocre
Moyen
Bon
Moyen
Bon
Moyen
Bon
Moyen
Bon
Moyen
Mdiocre
Bon
Bon
Mdiocre
Bon
Mdiocre
Mdiocre
a. Limited to first 100 cases.
Estimated
Cell
Probability for
Response
Category: 1
.01
.00
.01
.00
.73
.94
.00
.00
.00
.63
.92
.20
.00
.30
.02
.02
.05
.00
.21
.95
.60
.99
.08
1.00
.14
1.00
.62
.00
.84
.25
.00
.49
.00
.00
Estimated
Cell
Probability for
Response
Category: 2
.88
.03
.19
.07
.26
.06
.00
.00
.00
.34
.08
.50
.04
.69
.77
.98
.95
.00
.72
.05
.40
.01
.92
.00
.86
.00
.38
.00
.16
.75
.00
.51
.38
.00
Estimated
Cell
Probability for
Response
Category: 3
.10
.97
.79
.93
.01
.00
1.00
1.00
1.00
.03
.00
.30
.96
.00
.21
.00
.00
1.00
.08
.00
.00
.00
.00
.00
.00
.00
.00
1.00
.00
.00
1.00
.00
.62
1.00
Predicted
Response
Category
Moyen
Mdiocre
Mdiocre
Mdiocre
Bon
Bon
Mdiocre
Mdiocre
Mdiocre
Bon
Bon
Moyen
Mdiocre
Moyen
Moyen
Moyen
Moyen
Mdiocre
Moyen
Bon
Bon
Bon
Moyen
Bon
Moyen
Bon
Bon
Mdiocre
Bon
Moyen
Mdiocre
Moyen
Mdiocre
Mdiocre
125
Bon
8
3
0
32.4%
Moyen
3
7
1
32.4%
Mdiocre
0
1
11
35.3%
Percent
Correct
72.7%
63.6%
91.7%
76.5%
126
Lake
Hancock
Gender
Male
Female
Oklawaha
Male
Female
Trafford
Male
Female
George
Male
Female
Size
2.3
>2.3
2.3
>2.3
2.3
>2.3
2.3
>2.3
2.3
>2.3
2.3
>2.3
2.3
>2.3
2.3
>2.3
Fish
7
4
16
3
2
13
3
0
3
8
2
0
13
9
3
8
Other
5
2
3
3
1
0
2
0
1
5
4
0
2
2
1
1
127
Exemple Alligators
The sample consisted of 219 alligators captured in four Florida lakes,
during September 1985.
The response variable is the primary food type, in volume, found in
an alligators stomach. This variable had five categories: Fish,
Invertebrate, Reptile, Bird, Other.
The invertebrates found in the stomachs were primarily apple snails,
aquatic insects, and crayfish.
The reptiles were primarily turtles (though one stomach contained
tags of 23 baby alligators that had been released in the lake during the
previous year!).
The Other category consisted of amphibian, mammal, plant material,
stones or other debris, or no food of dominant type.
128
Exemple Alligators
Likelihood Ratio Tests
Effect
Intercept
LAKE
GENDER
SIZE
-2 Log
Likelihood of
Reduced
Model
146.644a
196.962
148.859
164.244
Chi-Square
.000
50.318
2.215
17.600
df
Sig.
0
12
4
4
.
.000
.696
.001
129
Exemple Alligators
Likelihood Ratio Tests
Effect
Intercept
LAKE
SIZE
-2 Log
Likelihood of
Reduced
Model
95.028a
144.161
116.115
Chi-Square
.000
49.133
21.087
df
Sig.
0
12
4
.
.000
.000
130
Modle estim
CHOICE
B
Parameter Estimates
Intercept
[LAKE=G
[LAKE=H
[LAKE=O
[LAKE=T
[SIZE=<=2.3
[SIZE=>2.3
Intercept
[LAKE=G
[LAKE=H
[LAKE=O
[LAKE=T
[SIZE=<=2.3
[SIZE=>2.3
Intercept
[LAKE=G
[LAKE=H
[LAKE=O
[LAKE=T
[SIZE=<=2.3
[SIZE=>2.3
Intercept
[LAKE=G
[LAKE=H
[LAKE=O
[LAKE=T
[SIZE=<=2.3
[SIZE=>2.3
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
]
B
Std. Error
-.626
.642
1.847
1.317
1.300
.993
-1.265
1.233
0a
.
-.279
.806
0a
.
.379
.479
2.935
1.116
1.692
.780
.476
.634
a
0
.
.351
.580
0a
.
-.048
.505
1.813
1.127
-1.088
.908
.292
.641
a
0
.
1.809
.603
0a
.
-.009
.522
1.419
1.189
1.002
.830
-1.034
.840
0a
.
.683
.651
a
0
.
Wald
.952
1.967
1.712
1.052
.
.120
.
.626
6.913
4.703
.564
.
.367
.
.009
2.590
1.434
.207
.
9.008
.
.000
1.424
1.459
1.515
.
1.099
.
df
1
1
1
1
0
1
0
1
1
1
1
0
1
0
1
1
1
1
0
1
0
1
1
1
1
0
1
0
Sig.
.329
.161
.191
.305
.
.729
.
.429
.009
.030
.452
.
.545
.
.925
.108
.231
.649
.
.003
.
.987
.233
.227
.218
.
.295
.
131
Prvision
Case Summaries
LAKE
1
2
3
4
5
6
7
8
H
H
O
O
T
T
G
G
SIZE
<=2.3
>2.3
<=2.3
>2.3
<=2.3
>2.3
<=2.3
>2.3
Estimated
Cell
Probability for
Response
Category: B
.07
.14
.01
.03
.04
.11
.03
.08
Estimated
Cell
Probability for
Response
Category: F
.54
.57
.26
.46
.18
.30
.45
.66
Estimated
Cell
Probability for
Response
Category: I
.09
.02
.60
.25
.52
.19
.41
.14
Estimated
Cell
Probability for
Response
Category: O
.25
.19
.05
.07
.17
.20
.09
.10
Estimated
Cell
Probability for
Response
Category: R
.05
.07
.08
.19
.09
.20
.01
.02
132
LENGTH
________
1.30
1.32
1.32
1.40
1.42
1.42
1.47
1.47
1.50
1.52
1.63
1.65
1.65
1.65
1.65
1.68
1.70
1.73
1.78
1.78
1.80
1.85
1.93
1.93
1.98
CHOICE
________
I
F
F
F
I
F
I
F
I
I
I
O
O
I
F
F
I
O
F
O
F
F
I
F
I
SEX
________
M
M
M
M
M
M
M
M
M
M
M
M
M
M
F
F
F
F
F
F
F
F
LENGTH
________
2.03
2.03
2.31
2.36
2.46
3.25
3.28
3.33
3.56
3.58
3.66
3.68
3.71
3.89
1.24
1.30
1.45
1.45
1.55
1.60
1.60
1.65
CHOICE
________
F
F
F
F
F
O
O
F
F
F
F
O
F
F
I
I
I
O
I
I
I
F
SEX
________
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
LENGTH
________
CHOICE
________
1.78
1.78
1.80
1.88
2.16
2.26
2.31
2.36
2.39
2.41
2.44
2.56
2.67
2.72
2.79
2.84
133
I
O
I
I
F
F
F
F
F
F
F
O
F
I
F
F
Intercept
2
9.84
0.0073
sex
2
2.71
0.2574
length
2
10.28
0.0059
length*sex
2
2.57
0.2767
Likelihood Ratio
94
77.64
0.8890
134
Effect
Intercept
LENGTH
SEX
-2 Log
Likelihood of
Reduced
Model
92.270a
110.319
95.732
Chi-Square
.000
18.049
3.461
df
Sig.
0
2
2
.
.000
.177
135
Effect
Intercept
LENGTH
-2 Log
Likelihood of
Reduced
Model
104.563
106.681
Chi-Square
14.247
16.365
df
2
2
Sig.
.001
.000
Parameter Estimates
CHOICE
F
I
Intercept
LENGTH
Intercept
LENGTH
B
.998
.085
5.181
-2.388
Std. Error
1.176
.489
1.746
.921
Wald
.721
.030
8.807
6.718
df
1
1
1
1
Sig.
.396
.862
.003
.010
136
5.181 2.388Length
1
Prob(0)
.998 .085Length
5.181 2.388Length
1 e
e
137
.6
.4
Probabilit
.2
Prob(O)
Prob(I)
0.0
1.0
Prob(F)
1.5
Longueur
2.0
2.5
3.0
3.5
4.0
138