Post-hoc MANCOVA Tests in SPSS

#1
Hi,

SPSS won't allow me to run post-hoc tests on a MANOVA with covariates - how can I now interpret results following a significant Wilk's Lambda score? My IV is composed of 4 conditions, so I can't simply compare means in the Descriptive Statistics table.

I think I essentially need to run a Tukey test, but SPSS won't allow this for MANCOVAs.

Any suggestions?

Thanks,
garou
 

spunky

King of all Drama
#2
SPSS has no built-in syntax to perform any post-hoc on MANCOVAs... post-hoc tests for MANOVA are themselves controversial enough so i guess they just decided to skip those... best option would be to obtain the adjusted vectors of means and probably do your own hotelling T^2s on those (while adjusting for type 1 error) or, what almost everybody does, which is breaking up the MANCOVA into several individual ANCOVAs and do the analysis 1 by 1... it's a widely criticised method but it is, by far, the most common...
 

trinker

ggplot2orBust
#3
Mahalanobis D may be an option which is related to the Hotellings T. The Hotellings is more of an effect size where as the Mahalanobis D is the Post Hoc.
 

spunky

King of all Drama
#4
The Hotellings is more of an effect size
i would like to kindly disagree with that :) Hotelling's T^2 is the two-sample instance of the more general MANOVA and was developed specifically for purposes of significance testing. It would be as if you were to say that a t-test is used as an effect size. Referring to Hotelling's original paper (Hotelling, H. (1931). "The Generalization of Student's T-Ratio" in the Annals of Mathematical Statistics, Vol. 2, No.3) on the first paragraph of page 378 (the last page) you will see how Hotelling himself comments on how one can work out significance testing on it, using the ever-ubiquitous .05 level, through a confidence ellipse. Hotelling's T^2 even has a sampling distribution... i'm not sure if people use it for effect-size purposes or not (perhaps they do because it uses a standardized distance) but that's not what it is intended to do..
 

trinker

ggplot2orBust
#5
i would like to kindly disagree with that Hotelling's T^2 is the two-sample instance of the more general MANOVA and was developed specifically for purposes of significance testing.
Spunky I would angerly agree 100%:mad: . Man that's the first time I got to use the angry face. It was fun. Anyway...

I had Hotellings and Mahalanobis mixed up. I was going from memory but if I had thought about it what you said is what makes sense (something called T is like a t-test and Mahalanobis is a distance formula). Thanks for the correction.