You want a logistic regression with two (DVs)[multivariate] or if 1 DV, univariate.. See this link to get started. Model your moderation (interaction) variable as a product term of two IVs as usual in a factorial design: https://stats.stackexchange.com/questions/166922/what-is-a-multivariate-logistic-regression This link should get you started.This link should keep you going: .

https://www.google.com/search?client=firefox-b-1-d&q=multivariate+logistic+regression+r

The attached file covers univariate logistic regression.

Best wishes, David Booth

Hi David Eugene Booth ,

Thank you for the advice and the readings - they are very much appreciated!

This morning I found this on the IBM website, and wondered if this was the technique that you speak of? http://www-01.ibm.com/support/docview.wss?uid=swg21476743

If it is, I have one final question...

When I do the singular regression of X to Y1 including M on PROCESS (rather than X to Y1 and Y2), the output includes both the regression results and the conditional effects for each of the five categories of M.

These conditional effects are much easier to interpret (especially for someone very new to stats and quant research in general), and I was wondering whether there is any way (ie. an equation/software/a website) to determine the these conditional effects from the regression output from the multivariate linear regression analysis?

Many thanks

Hello Edward. I don't understand why logistic regression has been suggested given that both DVs are scale variables. There must have been a misunderstanding.

If all you are doing is testing an interaction effect, you do not need the PROCESS macro. From what you've said, a straightforward multivariate model with 2 DVs estimated via the GLM command will do the trick. Here's an example using one of the sample datasets that comes with SPSS.

* Modify path to show where the sample datasets are stored on your computer.

GET FILE = "C:/SPSSdata/survey_sample.sav".

* Rename some variables to match Edward's variable names.

RENAME VARIABLES ( race age educ paeduc = X M Y1 Y2).

* Decide in advance which of the 4 multivarate F-statistics you will use.

* Then estimate a multivarate model with Y1 and Y2 as DVs

* and with X, M and X*M as explantory variables.

GLM Y1 Y2 BY X WITH M

/METHOD=SSTYPE(3)

/DESIGN=M X M*X.

This assumes that you really do have a multivariate question rather than a pair of univariate questions. As David noted above, multivariate and univariate describe the number of dependent variables in the model. See Hubery & Morris (1989) for more good info (link below). HTH.

https://pdfs.semanticscholar.org/ee6c/77c99c8e4530d0cccaedf85ed525fb22a02d.pdf

Hi Bruce Weaver ,

Thank you for the response. I think you captured the core factor leading to my confusion in your final paragraph - I am trying to determine whether two univariate or one multivariate model is most appropriate.

Having read the reading you sent me (thank you for that!), I feel as if I may be trying to over-complicate things, and that a number of univariate analyses would be better.

Whilst my dependent variables are conceptually similar and statistically correlated, they are different behaviours which are conceptually independent. My understanding is, therefore, that Huberty and Morris suggest that multiple univariate analyses are most appropriate for what I am looking to do

Hello Edward. If you really do have two univariate questions, you could just make two copies of the GLM command above, and then remove Y2 from the 1st copy and Y1 from the second copy. HTH.