I ran an ANOVA using Satterthwaite df on the same model using both type II and type III sums of squares. The model contains no interactions and all factors are effect coded (contr.sum), and so if run w/ vanilla lm type II and III SS give the same results (using car::Anova).
I find that using lmerTest (version 3.1-2), the results are also the same EXCEPT in the case of a multiple-df test on a factor w/ 3 levels (congruent.f).
anova(choice_lm.test,type="II")
Type II Analysis of Variance Table with Satterthwaite's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
congruent.f 15.3285 7.6643 2 52.468 51.1370 4.759e-13 ***
isH.f 0.0666 0.0666 1 39.174 0.4445 0.5088568
zaSNR 2.5808 2.5808 1 44.202 17.2194 0.0001492 ***
blz 22.1505 22.1505 1 42.820 147.7910 1.785e-15 ***
posX 0.1401 0.1401 1 34.631 0.9345 0.3404018
posY 2.0839 2.0839 1 38.255 13.9038 0.0006217 ***
anova(choice_lm.test,type="III")
Type III Analysis of Variance Table with Satterthwaite's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
congruent.f 15.3285 7.6643 2 57.663 51.1370 1.683e-13 ***
isH.f 0.0666 0.0666 1 39.174 0.4445 0.5088568
zaSNR 2.5808 2.5808 1 44.202 17.2194 0.0001492 ***
blz 22.1505 22.1505 1 42.820 147.7910 1.785e-15 ***
posX 0.1401 0.1401 1 34.631 0.9345 0.3404018
posY 2.0839 2.0839 1 38.255 13.9038 0.0006217 ***
In this case, the denominator DF differs between the two approaches, but all values are the same for all other parameters, including the single-df factor isH.f. Can you explain why this would be the case?
Hello,
I ran an ANOVA using Satterthwaite df on the same model using both type II and type III sums of squares. The model contains no interactions and all factors are effect coded (contr.sum), and so if run w/ vanilla lm type II and III SS give the same results (using car::Anova).
I find that using lmerTest (version 3.1-2), the results are also the same EXCEPT in the case of a multiple-df test on a factor w/ 3 levels (congruent.f).
In this case, the denominator DF differs between the two approaches, but all values are the same for all other parameters, including the single-df factor isH.f. Can you explain why this would be the case?
Thanks, Nathan