Multivariate analysis is needed when there are 2 or more Dependent Variables (DV) are in your research model.  Base module of SPSS (i.e. without add-on module) can't handle multivariate analysis.  Don't know exactly the research model with how many Independent Variable(s) & how they connected to the 2 DV.  Very likely Structural Equation Modeling (SEM) can be used.  However, if you intend to use variance-based SEM / Partial Least Squares (PLS) then you might use SmarPLS etc. software.  If you intend to use Covariance-based SEM then you might want to use AMOS (add-on module from SPSS), LISREL etc.  Wishing you all the best.

It seems it is possible. I found this thread: http://stats.stackexchange.com/questions/4517/regression-with-multiple-dependent-variables

If you go down a bit, someone explains that for SPSS, you can do a multivariate regression using the GLM multivariate option. Hope this helps! I would be very interested by knowing more about your experiment :)

Best wishes,

Elsa

Yes, this is possible and I have heard it termed as joint regression or multivariate regression. In essence you would have 2 (or more) dependent variables, and examine the relationships between independent variables and the dependent variables, plus the relationship between the 2 dependent variables. I'm not familiar with SPSS, but there seem to be options in R: http://stackoverflow.com/questions/12161659/how-to-write-r-formula-for-multivariate-response Good luck!

Sounds like a MANOVA/MANCOVA type model, you can use the manova() function in R and specify multiple outcomes, and multiple predictors, both categorical and continuous, assuming the residuals from the model are multivariate normal. Here's a link:

http://stats.stackexchange.com/questions/124694/how-to-do-mancova-in-r

I've always seen people have trouble interpreting the canonical correlation analysis, although you are right on, that it sounds like a good fit here, assuming everything is continuous. If you're wanting ease of interpretation of the regression effects and if you're wanting a model for prediction, I would still stick with the good old linear model, my 2 cents of course

Another issue to think about is the within design. This might be something that's tripping up the analyses. Using a MANOVA might make interpretation easier.

As Paul R. Yarnold stated, canonical regression is you proper technique.

If we want to draw genuine multivariate conclusions (i.e. to explain by the predictors in the model what they have in common the different dependent variables), then the optimal technique is canonical correlation. By contrast, other approaches based on SPSS MANOVA-menus only allow us to address the different dependent variables separately.

In the first of the aforementioned approaches, the interpretative complexity is inevitable, no matter the statistical package we use. Basically because solving linear equations often support more than one solution, which must combine to reach a proper conclusion.

This leads to properly plan the study, so that the interpretation feasible: For example calibrating the number of dependent variables.

Again, as Paul R. Yarnold stated, the most important is a good manual from the point of view of interpretation, the kind that he has recommended.

Multivariate analysis is needed when there are 2 or more Dependent Variables (DV) are in your research model.  Base module of SPSS (i.e. without add-on module) can't handle multivariate analysis.  Don't know exactly the research model with how many Independent Variable(s) & how they connected to the 2 DV.  Very likely Structural Equation Modeling (SEM) can be used.  However, if you intend to use variance-based SEM / Partial Least Squares (PLS) then you might use SmarPLS etc. software.  If you intend to use Covariance-based SEM then you might want to use AMOS (add-on module from SPSS), LISREL etc.  Wishing you all the best.

If the independent variables (IV) (x1, x2) do not have strong inter-dependency then MV Analysis makes sense (y = f(x1, x2, xn). If the IV's are mutually dependent, you'd rather vary one while keeping the other constant and regress it against the dependent (y) variable. I tried MVA in excel with reasonable accuracy. You get data analysis tool pack which can be added on to Excel and it performs the MVA. It's  very simple. Google a bit to for videos and you can do it. Make sure to test for correlations between x1...xn. If the correlations are weak, do an MVA. If the correlation is strong don't include the variable in the MVA. Try 2 at a time. It is easy to visualize and interpret. Best,

Madhavi

Ps: Please read this paper for a detailed account on the procedure and interpretation of MVA.

Sharma, M. R. & Ali, S., 1986. Tropical Summer Index—a study of thermal comfort in Indian subjects. Building and Environment, 21(1), p. 11–24.

I think what you are trying to do is effectively a repeated measures regression rather than a multivariate one in a classical sense. You can run this as a multilevel model (linear mixed model in SPSS - though I'd probably run it using R or MLwiN).