Hello,
I'm researching the effect of a self-compassion intervention on well-being. There are 2 groups (intervention and control) and 3 time points (pre, post and follow-up).
Since I have multiple dependent variables (life satisfaction, positive affect, psychological well-being, optimism, negative affect, depression, stress) I wanted to run a Mixed MANOVA instead of a Mixed ANOVA but can't seem to find how to include multiple dependent variables in Mixed Models in SPSS.
Is this the correct study design and is it possible to run the Mixed MANOVA in SPSS or do I have to run multiple Mixed ANOVAS?
Hello Céline Nagel. Having multiple DVs is not necessarily a good reason to use MANOVA. You should only use MANOVA if you have a multivariate question, and will be satisfied with an answer to that multivariate question. If you really have a bunch of univariate questions, then you ought to estimate a bunch of univariate models, and figure out how you will deal with the multiplicity problem that comes along with that. I recommend that you take a look at the following classic article by Huberty & Morris (1989).
Article Multivariate Analysis Versus Multiple Univariate Analyses
See also:
Article MANOVA: A Procedure Whose Time Has Passed?
Regarding multiplicity, I think the two articles by Schulz & Grimes (2005) are very good.
Article Schulz K, Grimes DMultiplicity in randomised trials I: endpo...
Article Multiplicity in randomised trials II: Subgroup and interim analyses
HTH.
PS- Even if you are still convinced that you have a multivariate question and want to use MANOVA, consider the excellent point Thom Baguley makes in the "best" answer in this discussion:
https://www.researchgate.net/post/Lower_Type_I_error_risk_in_MANOVA_compared_with_ANOVA-why_is_it_the_case
You should use a repeated measures model, and mixed model univariate approach. You essentially have a 2 (Treatment) X 3 (Time) factorial design in which Time is a repeated measures variable and Treatment is a between subjects variable. You are most interested in the Treatment X Time interaction because you expect the Control group not to change and expect the Intervention group to change over time. You should include the different dependent measures as another repeated measures factor in the design, which would enable you to test whether the Treatment X Time interaction varied as a function of dependent variables. This would enable you to test the relative strength of the Intervention across the different dependent variables. I am familiar with the measures you mention and I know that they would not form a strong single variate in a MANOVA; there is not enough common variance among them. You would lose a lot of unique variance if you adopt a MANOVA approach.
I think looking at Bruce Weaver and Francis C. Dane answers is a good idea.
If you are really set on a multivariate regression approach (not MANOVA) you can set this up as a multivariate mixed model, but I'm not really seeing a strong case for that over a repeated measures ANOVA. You can do this in MLwiN (see chapter 14 http://www.bristol.ac.uk/cmm/media/software/mlwin/downloads/manuals/3-05/manual-web.pdf ) and I know you can do it in brms in R: https://cran.r-project.org/web/packages/brms/vignettes/brms_multivariate.html I'm not sure how that is to do in SPSS as you'd have to lose the level variance parameter.
I totally agree with the critical points presented so far. If you are still interested in doing a MANOVA as described, you can have a look at the Tabachnick & Fidell book "Using Multivariate Statistics". They call it a doubly multivariate ANOVA, since you have repeated measures, which are multivariate but not commensurable, and severeral DVs, which in turn are.
Under SPSS, you can simply open the repeates measures ANOVA dialogue (general linear model -> repeated measures). On the first page you define how many repeated measures you have, BUT now, additionally you have to define all DVS at the bottom under "Measure Name". You have to add all DVs and in the next dialogue page you are able to incorporate all measures. Another possibility is to use the MANOVA command via syntax, but for this you should check the book.
Bruce Weaver thank you for bringing up Thom Baguley post in the other thread!! These are really crucial points and why I do not really like the MANOVA approach. I really cannot come up with an example from my area where I am really interested in the compound variable in the first place and not the univariate analyses.
One has to use Manova when he "measures" several Characteristics on the same item.
Decisions with Anova and Manova are different due to the correlation of the "measures" of the several Characteristics
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA