500
Le logiciel R - Effectuer des analyses statistiques
> par(mfrow=c(1,2))
> hist(residuals(modele1), main="Histogramme")
> qqnorm(resid(modele1),datax=TRUE) # Attention: quantiles
# normalisés en ordonnée.
Histogramme
residuals(modele1)
Frequency
−2000 0 2000
0
10
20
30
40
●
●
● ● ●
● ● ●
● ●
●
●
●
● ● ● ●
● ● ● ●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
● ● ●
●
●
● ●
●
●
● ●
● ●
●
●
●
● ●
●
● ● ●
●
●
● ● ● ●
● ●
●
●
●
●
●
●
●
● ●
●
●
● ● ●
●
●
●
●
● ●
●
●
●
●
●
● ●
●
●
● ●
● ● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ● ● ●
●
● ●
●
●
●
●
●
● ●
●
●
●
●
●
●
● ●
● ●
●
●
●
●
● ● ● ●
●
●
−2000
0
2000
−3
−2
−1
0
1
2
3
Normal Q−Q Plot
Sample Quantiles
Theoretical Quantiles
Figure 12.4 – Inspection graphique de la normalit´ e des r´ esidus.
> require("tseries")
> jarque.bera.test(residuals(modele1)) # Test de non-normalité de
# Jareque et Bera.
Jarque Bera Test
data: residuals(modele1)
X-squared = 1.7877, df = 2, p-value = 0.4091
Le test ne permet pas de conclure ` a la non-normalit´ e des erreurs. Par
ailleurs, le QQ-plot sugg` ere des erreurs normales puisque les quantiles observ´ es
et les quantiles th´ eoriques (obtenus si la distribution est normale) forment une
droite. L’hypoth` ese de normalit´ e ne sera donc pas remise en question.
Examinons maintenant le graphe des r´ esidus en fonction des valeurs pr´ edites. La figure ci-contre montre un nuage de r´ esidus correctement r´ epartis et
sym´ etrique autour de l’axe des abscisses, les conditions du mod` ele ne semblent
donc pas invalid´ ees.