Julius Verrel
2007-Jun-12 08:57 UTC
[R] Post-hoc tests for interactions of between- and within-subject factors
Is there a standardized way in R to perform post-hoc comparisons for main adn interaction effects of between- and within-subject factors? For instance, I have a data set of performance of adults of different age groups (20-30, 60-70,70-80) performing a WM task (n-back, with n=1,2,3,4) in two different conditions (while sitting or walking). The corresponding ANOVA produces the following output> gm = aov(cog ~ agegp*nback*cond+Error(id/(cond*nback)),data = g); > summary(gm)Error: id Df Sum Sq Mean Sq agegp 1 7.0268 7.0268 Error: id:cond Df Sum Sq Mean Sq cond 1 5.1788 5.1788 Error: id:nback Df Sum Sq Mean Sq nback 2 586.88 293.44 Error: id:cond:nback Df Sum Sq Mean Sq nback:cond 2 2.8901 1.4451 Error: Within Df Sum Sq Mean Sq F value Pr(>F) agegp 2 139.29 69.64 45.4349 < 2.2e-16 *** nback 2 145.40 72.70 47.4288 < 2.2e-16 *** cond 1 0.38 0.38 0.2495 0.6176019 agegp:nback 4 30.32 7.58 4.9448 0.0006367 *** agegp:cond 2 1.44 0.72 0.4698 0.6254042 nback:cond 2 0.25 0.12 0.0815 0.9217123 agegp:nback:cond 4 2.22 0.56 0.3621 0.8356007 Residuals 546 836.92 1.53 --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 TukeyHSD does not seem to work for within-person factors. Any suggestions how to perform post-hoc comparions? Thanks in advance, Julius -- Julius Verrel (PhD student) MPI for Human Development Lentzeallee 94 D-14195 Berlin +49 30 82406-410
Richard M. Heiberger
2007-Jun-25 19:17 UTC
[R] Post-hoc tests for interactions of between- and within-subject factors
The glht function in package multcomp requires "aov" objects. As you discovered, it does not work with "aovlist" objects. See the documentation ?MMC in the HH package from CRAN. The maiz example is similar to your example. Come back to the list if you need further help. Rich