Modérateur : Groupe des modérateurs
Code : Tout sélectionner
fm <- lme(Y ~ N * V * I, random ~ 1 | B / V / I, method = "ML")
Code : Tout sélectionner
fm <- lme(Y ~ N * V * I, random ~ 1 | B / V / I)
anova(fm)
Code : Tout sélectionner
anova(fm, type = "m")
Code : Tout sélectionner
library("nlme")
Y1<-rnorm(36)
bl<-factor(rep(1:3, times=12))
ir<-factor(rep(1:2, each=18))
va<-factor(rep(1:2,each=3, times=6))
az<-factor(rep(1:3, each=3, times=4))
m.aov<-aov(Y1~ir*va*az + Error(bl/ir/va))
m.lme<-lme(Y1~ir*va*az, random=~1|bl/ir/va)
summary(m.aov)
anova(m.lme, type="m")
anova(m.lme, type="s")
Code : Tout sélectionner
Bl <- paste(ir, va, az, sep = "_")
m.lme <- lme(Y1 ~ ir*va*az, random = ~ 1 | Bl)
Code : Tout sélectionner
> library("nlme")
> set.seed(123)
> Data <- data.frame(
+ Y1 = rnorm(36),
+ bl = factor(rep(1:3, times = 12)),
+ ir = factor(rep(1:2, each = 18)),
+ va = factor(rep(1:2, each = 3, times = 6)),
+ az = factor(rep(1:3, each = 3, times = 4)))
> Data$bloc <- with(Data, factor(paste(ir, va, az, sep = "")))
>
> m1 <- aov(Y1 ~ ir * va * az + Error(bl / ir / va), data = Data)
> m2 <- lme(Y1 ~ ir * va * az, random = ~ 1 | bl / ir / va, data = Data)
>
> m1
Call:
aov(formula = Y1 ~ ir * va * az + Error(bl/ir/va), data = Data)
Grand Mean: 0.05560446
Stratum 1: bl
Terms:
Residuals
Sum of Squares 2.019690
Deg. of Freedom 2
Residual standard error: 1.004910
Stratum 2: bl:ir
Terms:
ir Residuals
Sum of Squares 0.2855233 1.4188413
Deg. of Freedom 1 2
Residual standard error: 0.8422711
5 out of 6 effects not estimable
Estimated effects are balanced
Stratum 3: bl:ir:va
Terms:
va ir:va Residuals
Sum of Squares 1.0369324 0.1304834 1.0626924
Deg. of Freedom 1 1 4
Residual standard error: 0.5154349
4 out of 6 effects not estimable
Estimated effects may be unbalanced
Stratum 4: Within
Terms:
az ir:az va:az ir:va:az Residuals
Sum of Squares 0.114185 1.249827 1.915827 2.315214 19.306785
Deg. of Freedom 2 2 2 2 16
Residual standard error: 1.098487
Estimated effects may be unbalanced
Code : Tout sélectionner
> summary(m2)
Linear mixed-effects model fit by REML
Data: Data
AIC BIC logLik
113.0993 131.9481 -40.54964
Random effects:
Formula: ~1 | bl
(Intercept)
StdDev: 0.0402903
Formula: ~1 | ir %in% bl
(Intercept)
StdDev: 0.0001522510
Formula: ~1 | va %in% ir %in% bl
(Intercept) Residual
StdDev: 7.529249e-05 0.9951769
Fixed effects: Y1 ~ ir * va * az
Value Std.Error DF t-value p-value
(Intercept) 0.2560184 0.5750363 16 0.4452212 0.6621
ir2 -0.5357715 0.8125585 2 -0.6593635 0.5774
va2 0.1233928 0.8125585 4 0.1518572 0.8867
az2 -0.2708140 0.8125585 16 -0.3332856 0.7432
az3 -0.7530177 0.8125585 16 -0.9267242 0.3678
ir2:va2 0.2460439 1.1491313 4 0.2141130 0.8409
ir2:az2 0.8927399 1.1491313 16 0.7768824 0.4486
ir2:az3 0.5414556 1.1491313 16 0.4711869 0.6439
va2:az2 0.5296899 1.1491313 16 0.4609481 0.6510
va2:az3 0.4796553 1.1491313 16 0.4174069 0.6819
ir2:va2:az2 -1.8989230 1.6251170 16 -1.1684838 0.2597
ir2:va2:az3 0.4383414 1.6251170 16 0.2697291 0.7908
[...]
Standardized Within-Group Residuals:
Min Q1 Med Q3 Max
-2.08535315 -0.55331868 -0.04829478 0.46672070 1.68519531
Number of Observations: 36
Number of Groups:
bl ir %in% bl va %in% ir %in% bl
3 6 12
> VarCorr(m2)
Variance StdDev
bl = pdLogChol(1)
(Intercept) 1.623309e-03 4.029030e-02
ir = pdLogChol(1)
(Intercept) 2.318036e-08 1.522510e-04
va = pdLogChol(1)
(Intercept) 5.668959e-09 7.529249e-05
Residual 9.903770e-01 9.951769e-01
Code : Tout sélectionner
> intervals(m2)
Erreur dans intervals.lme(m2) : Cannot get confidence intervals on var-cov components: Non-positive definite approximate variance-covariance
Code : Tout sélectionner
> m3 <- aov(Y1 ~ ir * va * az + Error(bloc), data = Data)
> m4 <- lme(Y1 ~ ir * va * az, random = ~ 1 | bloc, data = Data)
> m3
Call:
aov(formula = Y1 ~ ir * va * az + Error(bloc), data = Data)
Grand Mean: 0.05560446
Stratum 1: bloc
Terms:
ir va az ir:va ir:az va:az
Sum of Squares 0.2855233 1.0369324 0.1141853 0.1304834 1.2498272 1.9158266
Deg. of Freedom 1 1 2 1 2 2
ir:va:az
Sum of Squares 2.3152141
Deg. of Freedom 2
Estimated effects may be unbalanced
Stratum 2: Within
Terms:
Residuals
Sum of Squares 23.80801
Deg. of Freedom 24
Residual standard error: 0.9959921
> summary(m3)
Error: bloc
Df Sum Sq Mean Sq
ir 1 0.28552 0.28552
va 1 1.03693 1.03693
az 2 0.11419 0.05709
ir:va 1 0.13048 0.13048
ir:az 2 1.24983 0.62491
va:az 2 1.91583 0.95791
ir:va:az 2 2.31521 1.15761
Error: Within
Df Sum Sq Mean Sq F value Pr(>F)
Residuals 24 23.808 0.992
> summary(m4)
Linear mixed-effects model fit by REML
Data: Data
AIC BIC logLik
109.0996 125.5924 -40.54982
Random effects:
Formula: ~1 | bloc
(Intercept) Residual
StdDev: 1.533430 0.9959921
Fixed effects: Y1 ~ ir * va * az
Value Std.Error DF t-value p-value
(Intercept) 0.2560184 1.637704 24 0.1563276 0.8771
ir2 -0.5357715 2.316063 0 -0.2313285 NaN
va2 0.1233928 2.316063 0 0.0532770 NaN
az2 -0.2708140 2.316063 0 -0.1169286 NaN
az3 -0.7530177 2.316063 0 -0.3251283 NaN
ir2:va2 0.2460439 3.275408 0 0.0751186 NaN
ir2:az2 0.8927399 3.275408 0 0.2725584 NaN
ir2:az3 0.5414556 3.275408 0 0.1653093 NaN
va2:az2 0.5296899 3.275408 0 0.1617172 NaN
va2:az3 0.4796553 3.275408 0 0.1464414 NaN
ir2:va2:az2 -1.8989230 4.632127 0 -0.4099462 NaN
ir2:va2:az3 0.4383414 4.632127 0 0.0946307 NaN
[...]
Standardized Within-Group Residuals:
Min Q1 Med Q3 Max
-2.08100636 -0.54993282 -0.04561536 0.46985780 1.68762809
Number of Observations: 36
Number of Groups: 12
Warning message:
production de NaN in: pt(q, df, lower.tail, log.p)
> VarCorr(m4)
bloc = pdLogChol(1)
Variance StdDev
(Intercept) 2.3514082 1.5334302
Residual 0.9920003 0.9959921
> intervals(m4)
Erreur dans intervals.lme(m4) : Cannot get confidence intervals on var-cov components: Non-positive definite approximate variance-covariance
De plus : Warning message:
production de NaN in: qt(p, df, lower.tail, log.p)
Code : Tout sélectionner
> m4 <- aov(Y1 ~ 1 + Error(bl / ir / va), data = Data)
> m5 <- lme(Y1 ~ 1, random = ~ 1 | bl / ir / va, data = Data, method = "ML")
>
> m4
Call:
aov(formula = Y1 ~ 1 + Error(bl/ir/va), data = Data)
Grand Mean: 0.05560446
Stratum 1: bl
Terms:
Residuals
Sum of Squares 2.019690
Deg. of Freedom 2
Residual standard error: 1.004910
Stratum 2: bl:ir
Terms:
Residuals
Sum of Squares 1.704365
Deg. of Freedom 3
Residual standard error: 0.7537384
Stratum 3: bl:ir:va
Terms:
Residuals
Sum of Squares 2.230108
Deg. of Freedom 6
Residual standard error: 0.6096595
Stratum 4: Within
Terms:
Residuals
Sum of Squares 24.90184
Deg. of Freedom 24
Residual standard error: 1.018615
> summary(m4)
Error: bl
Df Sum Sq Mean Sq F value Pr(>F)
Residuals 2 2.0197 1.0098
Error: bl:ir
Df Sum Sq Mean Sq F value Pr(>F)
Residuals 3 1.70436 0.56812
Error: bl:ir:va
Df Sum Sq Mean Sq F value Pr(>F)
Residuals 6 2.23011 0.37168
Error: Within
Df Sum Sq Mean Sq F value Pr(>F)
Residuals 24 24.9018 1.0376
> VarCorr(m5)
Variance StdDev
bl = pdLogChol(1)
(Intercept) 1.651793e-09 4.064226e-05
ir = pdLogChol(1)
(Intercept) 4.983046e-11 7.059070e-06
va = pdLogChol(1)
(Intercept) 6.647528e-12 2.578280e-06
Residual 8.571111e-01 9.258030e-01
> intervals(m5)
Erreur dans intervals.lme(m5) : Cannot get confidence intervals on var-cov components: Non-positive definite approximate variance-covariance
Code : Tout sélectionner
maov.06<- aov(Y~I*V*N + Error(B/I/V/N), data=spli06)
mlme.06<-lme(Y~I*V*N , random=~1| B/I/V/N, data=spli06)
Code : Tout sélectionner
summary(maov.06)
Error: B
Df Sum Sq Mean Sq F value Pr(>F)
Residuals 2 98.676 49.338
Error: B:I
Df Sum Sq Mean Sq F value Pr(>F)
I 1 1542.26 1542.26 3.9973 0.1836
Residuals 2 771.65 385.82
Error: B:I:V
Df Sum Sq Mean Sq F value Pr(>F)
V 1 1724.05 1724.05 10.5974 0.03122 *
I:V 1 1.75 1.75 0.0107 0.92246
Residuals 4 650.74 162.69
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Error: B:I:V:N
Df Sum Sq Mean Sq F value Pr(>F)
N 2 20720.5 10360.2 138.8727 7.748e-11 ***
I:N 2 442.5 221.3 2.9659 0.080238 .
V:N 2 1047.4 523.7 7.0199 0.006477 **
I:V:N 2 73.7 36.9 0.4943 0.619031
Residuals 16 1193.6 74.6
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
anova(mlme.06)
numDF denDF F-value p-value
(Intercept) 1 16 1285.1790 <.0001
I 1 2 7.0883 0.1169
V 1 4 10.5975 0.0312
N 2 16 138.8718 <.0001
I:V 1 4 0.0107 0.9225
I:N 2 16 2.9658 0.0802
V:N 2 16 7.0199 0.0065
I:V:N 2 16 0.4943 0.6190
Etonnant que l'on obtienne des résultats si différents entre les modèles ajustés sur les données simulées (aucune ifluence des facteurs sur réponse et/ou mauvais choix de paramètres pour générér Y)
Retourner vers « Questions en cours »
Utilisateurs parcourant ce forum : Google [Bot] et 1 invité