DISCLAIMER: I do not use the PROCESS macro. However, on the FAQ page (https://processmacro.org/faq.html), I see that there is a multicategorical option that you can select when you have categorical variables with 3 or more categories.

Out of curiosity, what is your sample size? I ask because the variables you listed will eat up a lot of degrees of freedom, and I wonder if you are in danger of overfitting your model. You can find more info about overfitting here:

• https://discourse.datamethods.org/t/author-checklist/3407#arrow_rightoverfitting-and-lack-of-model-validation-26

I hope this helps.

Hi Bruce,

I am also a bit concerned about overfitting the model - my sample size is 145.

I will see what I can do about the categorical variables from the guide you provided. Thank you!

Hello,

The problem I initially see is conceptual, not statistical. I don't know if it makes sense to try to partial out depression when you are looking at self-esteem as a moderator, as the two are so closely connected. Same for religiosity and death anxiety (research suggests a link between the two).

With variables that are closely related, it is usually better to find a way to incorporate their relationship into the analysis. Statistically you could do this by factoring continuous variables, clustering ordinal/nominal ones, or use finite mixture modeling if there is a mix of different kinds of predictors.

I would suggest running some analyses on your variables first to see which ones are interconnected, and then decide what you want to do.

There is a good chance that a SEM/path analytic model could do what you are looking for, while accounting for correlations between your predictors. The bottom line is: try to simplify whenever you can.

I stopped using SPSS a long time ago and I am guessing you may not want to learn R in-depth, but the open-source Jamovi software has many options to do this type of analysis. JASP is also an option.

Good luck,

DJ

Dear Isabella Neergaard I would definitely treat prefer not answering as a missing and remove them as there is no way you could get any useful info to your models

Hi David Joubert

Thank you so much for your reply! Unfortunately, at St Andrews we are only taught SPSS and get no introduction to R, so I am entirely reliant on SPSS and Jasp (which I have taught myself as it is relatively intuitive).

My hypotheses are as follows: Hypothesis 1 (H1): “There will be a significant positive relationship between death anxiety and symptoms of generalised anxiety”.

Hypothesis 2 (H2): “There will be a significant negative relationship between levels of self-esteem and symptoms of generalised anxiety”.

Hypothesis 3 (H3): “High levels of self-esteem will significantly moderate the strength of the positive relationship between death anxiety and symptoms of generalised anxiety, leading to a weaker relationship”.

The reason why I wanted to control for the aforementioned variables is because there is conflicting evidence for the effects of them (i.e. some individuals feel less death anxiety as they age, where some feel a lot more). I didn't want to leave a stone unturned so to speak, but I see your point about clustering. I've tried to run a regression using PROCESS which suggests that the overall model is significant but that there are no significant main effects or significant interaction. However, symptoms of depression do seem to contribute significantly (as the only variable, interestingly). This could be due to the sparse sample size, but interesting nonetheless.

I will, however, have a look into a path analytic model!! Thank you!!

For a Master's thesis using Moderated Multiple Regression (MMR), you can use PROCESS macro in SPSS (Model 1) to test interaction effects between an independent variable and a moderator; ensure all variables are mean-centered to reduce multicollinearity, interpret the interaction term for significance, and use simple slope analysis or Johnson-Neyman technique (built into PROCESS) to explore the nature of moderation.