How can I test interactions of between and within-subject variable in repeated measure anova
5 views (last 30 days)
Show older comments
Hi!
I have a 66 x 16 table (attached). Rows represents subejcts. Column 1:4 represents between-subejct variable (Group (PTSD/HC), Age, TBI (yes/no) and Wholebrain_FA. Age, TBI and WHolebrain_FA represents covariates. Column 5:16 represents within-subject measure: 5:10 represents FA measure from 6 white matter brain tract in left brain hemisphere and 11:16 represents FA measure from the same 6 white matter brain tracts but in right hemisphere.
I want to model the data using a mixed ANOVA with 2 within-, 1 between-subject and 3 covariates variables. To test the effect of Group, the Group X Tract and Group X Tract X Hemisphere I model the data with fitrm and test it with ranova:
% set within-subject variables
w1 = table(categorical({'CB','SLF_I','SLF_II','SLF_III','UF','OF_ST','CB','SLF_I','SLF_II','SLF_III','UF','OF_ST'}'), 'VariableNames',{'tract'});
w2 = table(categorical({'left','left','left','left','left','left', 'right','right','right','right','right','right'}'), 'VariableNames',{'hemisphere'});
w3 = table(w1.tract .* w2.hemisphere, 'VariableNames',{'tract_hemisphere'});
w = [w1 w2 w3];
% Fit model
rm = fitrm(t,'CB_left_FA-OF_ST_right_FA~group+age+tbi+Wholebrain_FA','WithinDesign',w, 'WithinModel', 'separatemeans');
% Test for sphericity assumption
mauchly(rm);
% Test model with ANOVA
a=ranova(rm,'WithinModel','tract*hemisphere');
I now have 2 questions:
1) Is it possible to test the Group X Hemisphere interaction for each level of Tract (CB, OF-ST, SLF-1, SLF-2, SLF-3, UF) using the coefficients derived from the omnibus model? I have looked into the function coeftest but can´t seem to figure it out. Can anyone help?
2) Using
multcompare(rm,'group','By','tract_hemisphere','ComparisonType','Bonferroni'));
I can test each pair of the three-way interaction. However, since I group the tract_hemisphere variable by Group the multiple comparison correction does no work the way I intend it; It corrects for each pair of test (HC vs. PTSD) at each level of tract_hemisphere (i.e. two tests) and not the whole batch of tests. Is it possible to set up the model so it corrects for for each Group X Tract X Hemisphere pairwise comparison?
1 Comment
Scott MacKenzie
on 28 Jul 2021
This design is way too complicated (IMHO). The ANOVA table has 20 F-statistics!
Answers (0)
See Also
Categories
Find more on ANOVA in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!