506
Le logiciel R - Effectuer des analyses statistiques
> newdata <- cbind(BWT,LWT,AGE,SMOKE)
> # Diagramme de dispersion.
> pairs(newdata,lower.panel=panel.smooth,upper.panel=add.cor)
BWT
40
80
0.19
0.09
0.0 0.4 0.8
1000 3000 5000
−0.19
40
80
●
●
● ● ●
● ●
●
●
●
●
●
●
● ● ● ●
● ● ● ●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
● ●
●
●
● ●
● ●
●
●
●
● ●
●
● ● ●
●
●
● ● ●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
●
●
● ● ●
●
●
●
●
● ●
●
●
● ●
● ● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
● ● ●
●
●
● ●
●
● ●
●
●
● ● ●
● ●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
● ● ●
●
●
●
● ● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
● ● ● ●
●
●
LWT
0.18
−0.04
●
●
● ●
●
● ●
●
●
●
● ●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ● ● ●
●
●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● ●
●
● ●
●
● ● ●
●
●
●
●
●
●
● ●
●
● ●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
● ●
● ●
●
●
●
●
● ● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
● ● ●
●
●
●
●
●
●
●
●
●
● ●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
AGE
15 25 35 45
−0.04
1000 3000 5000
0.0 0.4 0.8
● ●
● ● ●
● ● ●
● ●
● ● ● ●
● ●
●
●
●
●
● ● ● ● ● ●
●
●
●
● ●
● ●
● ●
● ● ● ● ● ●
● ● ●
● ●
● ● ●
●
● ●
● ●
● ●
●
● ● ● ● ● ● ●
●
● ●
●
● ●
● ● ●
●
●
● ●
● ●
● ● ● ● ● ●
●
● ● ● ● ● ●
● ● ●
● ●
● ●
● ●
●
● ● ● ● ● ●
●
● ● ●
●
●
●
● ● ● ● ● ● ● ● ● ● ● ●
●
● ●
●
●
●
● ● ● ● ● ●
● ● ●
● ●
● ●
●
●
● ● ● ●
● ●
●
● ● ●
●
● ●
● ● ●
● ●
● ●
●
●
● ● ●
● ●
● ● ● ●
● ● ●
● ● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
●
●
●
● ●
● ●
● ●
●
●
●●
●
●
●
●
●
●
● ● ● ●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
● ●
●
●● ●
●
●
●
● ●
● ●
●
●
●
● ●●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
● ●
● ● ●
●
●
●
● ●
●
●
●
●
●
●
● ● ●
●
●
●
● ●
●
●
●
●
● ●
●●
● ●
●
●
●●
●
● ●
● ●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
15 25 35 45
●
●
● ●
●
● ●
●
●
●
● ● ●
●
● ●
●
●
●
●
●
● ●
●
● ●
●
●
●
● ●
●
●
● ●
●
●
● ●●
●
● ●
●
● ●
●
● ●
●
● ●
●
●
●
●
●
●
●
● ● ● ● ●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ●
●
●●
●
● ●
●
● ●
● ●
●
●
●
●
●
● ● ●
●
●
●
●
●
●
●
●
●
●
●
●
● ● ●
● ● ● ● ●
●
●
●
●
●
●
● ●●
● ● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●●
●
●
●
●
●
●
●●
● ●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
SMOKE
Figure 12.6 – Diagramme de dispersion de toutes les paires de variables.
L’objectif de ce diagramme est double. Il permet de visualiser la relation entre
la variable `
a expliquer et chacune des variables explicatives, mais aussi de juger de la corr´ elation entre les variables explicatives. Le dernier point permet
entre autres choses de diagnostiquer un probl` eme potentiel de colin´ earit´ e (voir
la section sp´ ecifique sur ce sujet en 12.3.7).
Estimation des param` etres
Comme en r´ egression lin´ eaire simple, nous estimons le mod` ele par la fonction
lm() :
> modele2 <- lm(BWT∼AGE+LWT+as.factor(SMOKE))
> modele2
Call:
lm(formula = BWT ∼ AGE + LWT + as.factor(SMOKE))
Coefficients:
(Intercept)
AGE
LWT
2362.720
7.093
8.860
as.factor(SMOKE)1
-267.213
Tests sur les param` etres
Les tests sur les divers param` etres sont obtenus grˆ ace ` a la fonction summary().