Checking MANOVA assumptions - tests on data vs residuals?

12 views Asked by At

I'm pretty new on MANOVA and I want to check assumptions to my model which is a 2 way MANOVA with CBD factor. Here 2 questions

Most of the assumption checking proposals I see on internet use the data. This for normality, multinormality, variance and covariance homogeneity and for outlier detection.

In my opinion only the outlier detection makes sense on the original data and I'd rather check normality, multinormality, homogeneity of variances and covariances on the residuals.

For example, using {MVN} for multivariate normality via MD after having already checked the univariate QQ plots on residuals with qqnorm() should be performed as

res <- mvn(mydata[,depvar],mvnTest = "mardia", univariateTest = "SW", univariatePlot = "SW",multivariatePlot = "qq" )

or as I would naturally do as

res <- mvn(resid(fit),mvnTest = "mardia", univariateTest = "SW", univariatePlot = "SW",multivariatePlot = "qq" )

The results are of course different, although not very different.

Is there any graphical approach to assess the homogeneity of covariances in a TA style ?

TIA

gfwp

0

There are 0 answers