|
Tests des effets inter-sujets |
||||||
|
Variable dépendante: VAR00001 |
||||||
|
Source |
Somme des carrés de type III |
ddl |
Moyenne des carrés |
D |
Sig. |
|
|
Ordonnée à l'origine |
Hypothèse |
484.489 |
1 |
484.489 |
680.484 |
.000 |
|
Erreur |
2.136 |
3 |
.712a |
|
|
|
|
F1 |
Hypothèse |
11.690 |
2 |
5.845 |
4.257 |
.015 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F2 |
Hypothèse |
2.557 |
1 |
2.557 |
1.862 |
.173 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F3 |
Hypothèse |
.762 |
1 |
.762 |
.555 |
.457 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F4 |
Hypothèse |
2.136 |
3 |
.712 |
.519 |
.670 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F1 * F2 |
Hypothèse |
2.279 |
2 |
1.139 |
.830 |
.437 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F1 * F3 |
Hypothèse |
1.429 |
2 |
.714 |
.520 |
.595 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F2 * F3 |
Hypothèse |
1.100 |
1 |
1.100 |
.801 |
.372 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
F1 * F2 * F3 |
Hypothèse |
.369 |
2 |
.184 |
.134 |
.874 |
|
Erreur |
374.806 |
273 |
1.373b |
|
|
|
|
a. MS(F4) |
||||||
|
b. MS(Erreur) |
#3 Mixed effect ANOVA
|
Tests des effets fixes de type IIIa |
||||
|
Source |
Numérateur ddl |
Dénominateur dll |
F |
Sig. |
|
Constante |
1 |
179.859 |
477.158 |
.000 |
|
F1 |
2 |
112.144 |
12.424 |
.000 |
|
F2 |
1 |
163.517 |
1.805 |
.181 |
|
F3 |
1 |
167.461 |
.867 |
.353 |
|
F4 |
3 |
98.404 |
2.147 |
.099 |
|
F1 * F2 |
2 |
105.482 |
.850 |
.430 |
|
F1 * F3 |
2 |
106.020 |
2.082 |
.130 |
|
F2 * F3 |
1 |
154.899 |
.061 |
.806 |
|
F1 * F2 * F3 |
2 |
108.667 |
.290 |
.749 |
|
a. Variable dépendante : VAR00001. |
#4 anovan(VOI.data,VOI.F.FM,'random',[1 5],'model','full')
NB X1 is subject, X2 is F1, X3 is F2 etc...
To view this discussion on the web visit https://groups.google.com/d/msgid/fmri_matlab_tools/8692f6a5-c28a-4d6f-827d-cfa7bcd81cb0%40googlegroups.com.
anova(lmer(Y~(F1*F2*F3)+F4 + (1|Subj) + (1|F1:Subj) + (1|F2:Subj) + (1|F3:Subj) + (1|F4:Subj),family="gaussian",REML=TRUE,data=dat))
Using this model the F3 effect is not significant. Still not 100% sure that this is specified correctly.
An alternative would be to use lme from the nlme package which I think would be specified as follows:
anova(lme(fixed=Y~(F1*F2*F3)+F4 ,random=~1|Subj/F1/F2/F3/F4,method="REML",data=dat))
This does not give the exact same results as the lmer command, but they are very similar, and F3 is not significant in either model.
Again though, for this model I don't think there is a directly corresponding LME model, and the GLM and LME results will diverge. The degree of divergence is likely dependent on how complicated the covariance structure is, and with 4 repeated measures the covariance structure gets pretty crazy.
The standard ANOVA model in R is specified as:
summary(aov(Y~(F1*F2*F3)+F4 + Error(Subj/(F1*F2*F3*F4)), data = dat))
The standard ANOVA will return the same result as GLM_Flex
If anyone else is more up to speed on LME models please feel free to weigh in, I'd like to hear your thoughts.
-Aaron
To view this discussion on the web visit https://groups.google.com/d/msgid/fmri_matlab_tools/8cc695f8-6706-487a-ba97-bb44475df23f%40googlegroups.com.