Morning, Nila,

I think the easiest way is to perform a univariate analysis. You add your IV into "fixed factors" field (random factors is a rather seldom option); add DV, as your response/dependent variable. Then you go to "Model", click on "Custom", and add 1. both your IV as "main effects", then choose one of your IVs and MV and add them as "interaction". Add the second variable similarly to the model. 

Of course, it depends on your model: you can test for several relationships, e.g.: direct for both IVs, direct for MV, moderation IV1*MV, IV2*MV. You can also test for only one moderation, without direct DV or MV effects, and even the moderation IV1*IV2*MV. Remember that the more relationship you add, the worse will your model become. So try to remain "parsimonious". Moreover, the interpretation of a moderated moderation is really painful, theoretically painful.

The alternative option is "generalized linear modeling". The dialog field in SPSS looks differently because you define your effects and variables not in the single window but stepwise. First, you select your response/dependent variable, choose the right procedure, define your DVs. Similarly to univariate analysis, you can define your model including different constellations of direct and moderation effects. The generalized linear modeling is a relatively new procedure and it should produce similar results as the usual univariate procedure. It has several advantages and disadvantages, so look in the literature whether you should use it.

Alternatively, you can try to make dummy variables from your categorical variables. It works, however, only if you have two categories. You code them, for example, as 0 for "male" and 1 for "female". In this event, you can run a simple regression procedure. 

I would like to remind you that some scholars do not accept single items with Likert scales as "ratios" and treat them as ordinal. Consequently, they would criticize my suggestions. If you are a proponent of this approach, you can still use generalized linear models, but use the one for ordinal scales. It might, however, be problematic with moderation by an ordinal variable. 

Sincerely,

Eugene

Dear! You can use Macro develop by Hayes2015 named "Modprob".Its more convenient for you.

Professor Hayes wrote a program called “Process” which is a extra tool for SPSS to deal with moderator effect or mediator effect， click this address "http://www.processmacro.org/index.html"you could find it，also，there is a brief introduction of this program which could help you get it easier.

 Morning, Guan.

I am not sure whether PROCESS plug in for spss works with categorical variables. 

Best,

Eugene 

Thank you Dr. Bogodistov, I create dummy variable (for categorical IV) and run regression to test the moderation effect. The result looks pretty reasonable. While I plan to try PROCESS as suggested by Dr. Wang and use it as benchmark. If the outputs are similar, then I have stronger confidence to write my results.

Process is used for comolex models having nediation and moderation at the same time. But ModProb is only used for moderation.

Thank you Dr. Ahmad, will look to that too

Hi，Dr. Bogodistov

PROCESS could analysis  mediation model that contains categorical variables of X，there is a paper wrote by  author，please see the attachment

Thank you, Guan, good to know!

I have 3 IV likert scale variables and 1 DV which is also likert scale data. I want to do moderation analysis. can i use regression for this data. And suggest ways to do.

Thanks

Dear Yogesh

it depends on the variables:

1. Are these 3 variables items of the same latent construct?

1.1. If so, you can use the "dimension reduction" procedure in SPSS and aggregate them to a single variable which you can use in your calculation.

1.2. If not, you have to see on the scale. Technically, you can regress anything on anything. Yet, single items measurement might produce concerns on the side of your supervisor/reviewer. Moreover, it depends on the scale size and type (5-, 7, 9-, 100-Likert?).

2. Of course, you can run a moderation analysis. Moderation is nothing but regressions of the IV on DV, of the Moderator (MV) on DV, and of a multiplication of IV*MV on DV.

3. Is your moderating variable also a Likert scale? Or do you investigate a group moderation? This would assume different procedures but both should not be an issue.

As you can see, technically this would work. You may discover some theoretical issues, nonetheless.

Good luck!

Yevgen

Thanks Yevgen

Dear Yevgen and all researchers,

I have also a question regarding how to run in the right way a moderation analysis for ordinal likert scales variables (4-point) in SPSS. I have four IV, one DV and would like to test how MV affects the relationship between IV and DV. So far, I am applying ordinal logistic regression to test relationships between IVs and DV.

Thank you in advance!

Dear Irina

Moderation means that you calculate the direct effects from your IVs, the direct effect from your MV as well as the multiplication of IVs*MV. In a usual regression, you can build a model with Main Effects and Interaction Effects. In logistic regression, it simply works in Stata - you can directly add an interaction using a "*"-sign.

In SPSS, it might be tricky. Caution: it is only my opinion and may not the correct solution:

1. Calculate the z-scores (mean centres standardized scores for your IVs and MV). You do it if you go to Analyze-Descriptive Stats-Descriptives. Check the box "Save the z-scores" below. Rename your newly created variables accordingly.

2. Calculate the interaction variable. Go to "Compute" - and calculate multiplication of z-scored-variables. For each interaction - z_IV1*z_MV, z_IVn*z_MV.

3. Run a logistic regression, whereby you add all z_IVs, z_MV, and interactions.

Take care of the interpretation as the DVs and MV are standardized. I.e. an increase in a 1 standard deviation will cause an increase of 1 unit in your binary DV.

It may also work with direct (not standardized) IVs, yet I am not sure.

Good luck! I hope it helps you in your research!

Best

Yevgen (Eugene)

Thank you very much for the fast response! It confirms the main way for conducting it. However, my DV is also ordinal not dichotomous. Wondering if the extension of SPSS i.e. with the Process macro, is it correct to apply moderator analysis with ordinal and categorial data?

Please watch this video. It may help you with your calculations: https://www.youtube.com/watch?v=r2_zw4G55X0