7

I have two groups of persons, GRP0 and GRP1, on which I measured three continuous variables: VAR1, VAR2 and VAR3.

I would like to use Mancova in R with: - VAR1, VAR2 and VAR3 as outcome variables - GRP={0,1} as predictor variable - age and gender as covariates

What would be the correct way to formulate this model in R?

Also the measured were carried out on 100 instances (which are serially correlated) so any help on how to apply permutation-based multiple-comparison correction on top would be ideal.

Thanks a lot

michael
  • 293
  • 1
  • 3
  • 11

1 Answers1

11

See: http://www.statmethods.net/stats/anova.html

Y <- cbind(VAR1, VAR2, VAR3)
fit <- manova(Y ~ GRP+age+gender)
summary(fit, test="Pillai")

Other test options are "Wilks", "Hotelling-Lawley", and "Roy".

summary.aov(fit) # for univariate statistics.

Following link may be helpful for permutation based multiple comparisons: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2611984/pdf/1751-0473-3-15.pdf

R code on this page is also helpful: http://biostatistics1014.blogspot.in/2013/04/one-way-anova-permutation-test-and.html

rnso
  • 8,893
  • 14
  • 50
  • 94
  • 1
    Thanks a lot! And any idea on how to apply permutation-based multiple-comparison correction on N=100 of these MANCOVA? – michael Nov 20 '14 at 00:53
  • The article at the link does not provide R code on how to do this unfortunately. And especially how all the required variables would be extracted from the MANOVA commands in R. – michael Nov 20 '14 at 01:08
  • 1
    Please see a link in the answer above. – rnso Nov 20 '14 at 01:15
  • Thanks, it helps, but it does not say how to do it in the case of covariates. Should only the group variable be randomly sampled while keeping the original age and gender of each subject? Or should they all be sampled? Then in each way? – michael Nov 20 '14 at 21:19