Hello Isabella Wakerley. What you have described so far sounds to me as though you have one dichotomous dependent variable, not two dependent variables. I wonder, therefore, if you should be estimating a binary logistic regression model. But perhaps I misunderstood. Perhaps you can clarify things by saying more about the questionnaire you are using. E.g., if you think ANOVA is appropriate, explain how the means & SDs for the 4 groups are obtained from the questionnaire items.

Can you describe both of your conditions in the 2x2 design? From what you have said so far, it seems that the bystanders can have either an active or a passive response, but what is the other condition?

More generally, it appears that you are doing ANOVA because you have different conditions for you independent variables, as opposed to MANOVA where you have multiple dependent variables.

Is there two different response dimensions in your experiment: positive/negative, construction/destructive? Or is it directly nominal category variable: Active Constructive, Active Destructive, Passive Constructive and Passive Destructive?

If it is the first case, you measure these two dimensions:

Two consecutive rating dimensions? (For example: each participant has a 1-5 rating for the “positive/negative” dimension and the “build/destructive” dimension?)

Two binary categorical variables? (For example: tagging positive vs negative, constructive vs destructive)

Hi Isabella,

Thanks for sharing the details, your study sounds really interesting, and you're asking all the right questions.

From what you’ve described, it seems like you have:

1. One independent variable (your four bullying vignettes),

2. Either one or two dependent variables (depending on how you’re recording the bystander responses), and

3. Two moderators, which you plan to test using PROCESS in SPSS.

Before deciding on the best statistical test, it’s important to clarify how the bystander responses are measured.

1. Are participants choosing one response per vignette (e.g., either “active constructive” or “passive destructive”)? If yes, that means you have a single categorical outcome. In this case, ANOVA would not be the right approach. A Chi-square test or multinomial logistic regression would work better.

2. Are the responses rated on separate Likert scales (e.g., 1–5 for “active constructive” and 1–5 for “passive destructive”)? If yes, then you have two continuous dependent variables. In that case, MANOVA would be the correct test to compare the responses across your four vignettes. If MANOVA shows significant differences, you can follow up with individual ANOVAs for each dependent variable.

Regarding the moderators, using PROCESS in SPSS is a solid approach, especially if the moderators are continuous (like empathy levels).

1. If your dependent variable is continuous, PROCESS (Model 1 or 2) will work well for moderation analysis.

2. If the dependent variable is categorical, you might need to consider logistic regression with interaction terms instead of PROCESS.

Let me know how your responses are structured, that will help narrow down the exact method.

All the best!

Mutiu Shittu Did this reply come from an uncredited AI?

Moderation is simply another word for interaction, and an interaction effect can be assessed in a straightforward manner using either ANOVA or dummy variable regression. There is no reason to use something as complicated as PROCESS.

Hi David,

Just to clarify, I wrote that response myself. It wasn’t generated by any AI.

What I shared is based on how I’ve come to understand and approach statistical design, especially when choosing the right analysis for a specific research setup. I wasn’t quoting a textbook or pulling standard phrases. I was responding directly to the way the study was described, which is why I broke it down based on how the bystander responses were measured and gave possible paths depending on whether the outcomes were categorical or continuous.

I mentioned PROCESS because it’s a tool many researchers use when dealing with moderation, particularly with continuous moderators. But I also pointed out that it’s not the only option, and sometimes it may not be necessary. That reflects my own judgment, not something automatically copied or suggested by a system.

I get that people are cautious now with AI-generated content, and that’s fair. But in this case, it was my own thinking. I’d be happy to explain it again in a different way if that helps clear things up.

Let me know if you’d like to discuss it more.

Best Regards!

It seems to me that there are nearly as many interpretations of what Isabella Wakerley wrote in her original post as there are respondents. Therefore, I think it very unlikely that anyone will be able to offer much help until she comes back to this thread and explains very clearly the design, the nature of the outcome variable(s), and the research questions. YMMV. ;-)

Mutiu Shittu My apologies about the AI, but PROCESS is most relevant with structural equation models, and since there don't seem to be any scale-variables here, it still strikes me too complex to be a viable option -- whether the variables involved in the interaction effect are either dichotomous or continuous.

Isabella Wakerley, If you're measuring two separate types of bystander responses ("active constructive" and "passive destructive"), use a MANOVA to compare the four vignette groups. If you combine them into one overall score, use a One-Way ANOVA.

If you're including moderators, and you're testing how they interact with vignette type, use PROCESS in R Programming or SPSS (e.g., Model 1 or 2 for moderation). If you're just controlling for moderators, use MANCOVA (for two DVs) or ANCOVA (for one DV).

So:

Two DVs → MANOVA/MANCOVA

One DV → ANOVA/ANCOVA

Testing moderation → PROCESS

Hello Anuraj Nayarisseri. You recommended MANOVA (or MANCOVA) in the case of two DVs. But in doing so, you have assumed that Isabella Wakerley has one multivariate research question. If she actually has two univariate research questions, two univariate models will be more appropriate. See the following articles, for example.

Article Multivariate Analysis Versus Multiple Univariate Analyses

Article MANOVA: A Procedure Whose Time Has Passed?

HTH.

Bruce Weaver You're absolutely right to raise the distinction between a multivariate vs multiple univariate approach.

In my original reply, I assumed that the two response types (“active constructive” and “passive destructive”) were part of a single, multivariate research question — in other words, that Isabella Wakerley was interested in the combined profile of bystander behavior across conditions. In that case, a MANOVA (or MANCOVA with covariates) would be appropriate.

But if Isabella Wakerley ’s research is actually addressing two distinct univariate questions — like:

Then you're right — two separate ANOVAs or ANCOVAs would be more suitable, and MANOVA would not be necessary.

This is especially relevant considering:

• MANOVA has stricter assumptions (e.g. multivariate normality, homogeneity of covariance matrices).
• MANOVA results can be harder to interpret directly in applied psychology research.
• Many researchers now favor multiple univariate models due to better clarity and more direct interpretability (as discussed in the articles you referenced).

So ultimately, the choice depends on how Isabella Wakerley frames her research questions:

• One multivariate question → MANOVA/MANCOVA.
• Two univariate questions → Two ANOVAs or ANCOVAs (or moderated models via PROCESS, if testing interactions).

Thanks for pointing this out — it’s a good reminder to always tie the statistical method directly to the research question structure, not just the number of DVs.