If your dependent variable (DV) naturally comes in 3 categories, multinomial logistic (for unordered categories) or ordinal regression (for ordered categories) may be appropriate choices. But if your DV in its original form has more than 3 values, one would typically not recommend trichotomizing the scores because you may loose information and power.

Converting quantitative variables into ordinal ones might seem appealing for its simplicity, but it comes with several drawbacks that can compromise the integrity of your analysis. Here are several key arguments against such conversion:

In summary, keeping quantitative variables in their continuous form preserves critical information and statistical power, avoids arbitrary categorizations, and allows for more flexible and accurate modeling of relationships, including interactions and moderated effects. This approach not only supports a richer analysis but may also lead to more robust and interpretable findings.

If you use a continuous moderator—in your case, self-esteem—it’s most straightforward and informative to adopt a moderated multiple regression analysis. This method lets you maintain the full spectrum of your data without sacrificing nuance, and it’s well-suited for testing interaction effects. Let me break down why and how you’d do this:

Why Moderated Multiple Regression?

• Preserves Precision: When you treat self-esteem (and other variables like death anxiety) as continuous, you capture every subtle variation in the data. This not only increases the sensitivity of your model but also improves transparency in the relationships you’re exploring.
• Enhanced Statistical Power: Continuous variables utilize all available data points, allowing you to detect smaller effects that might be lost with categorical splits. This is critical when exploring an interaction (or moderation) effect because such effects are typically weaker compared to main effects.
• Nuanced Interaction Interpretation: Moderated regression lets you include an interaction term (i.e., the product of death anxiety and self-esteem) that tests whether the relationship between death anxiety and symptomatology changes at different levels of self-esteem. You can then perform a simple slopes analysis or use the Johnson-Neyman technique to visualize and interpret these changes across the full range of self-esteem.

How to Implement Moderated Multiple Regression

Symptomatology = b0 + b1*(Death Anxiety) + b2*(Self-Esteem) + b3*(Death Anxiety × Self-Esteem) + ε

• Here, b3 is the coefficient for the interaction term. A significant b3 indicates that self-esteem moderates the relationship between death anxiety and symptomatology.

4. Test and Interpret the Interaction: If the interaction is significant, use simple slope analysis to understand how death anxiety predicts symptoms at different levels of self-esteem. Many analytical tools (such as Hayes’ PROCESS macro in SPSS or relevant packages in R) provide these tools along with visualizations for clearer interpretation.

Since you are investigating relationships between continuous variables like death anxiety, self-esteem, and symptomatology, and also hypothesizing a moderation effect, a moderation analysis (such as using the PROCESS macro for SPSS or hierarchical regression) would generally be more appropriate.

Categorizing continuous variables into low, medium, and high and then applying multinomial logistic regression may lead to a loss of valuable information and reduced statistical power. It is usually recommended to keep variables continuous when exploring correlational and moderation relationships.

However, if your supervisor prefers categorization, multinomial logistic regression can still be applied, though it is less ideal for your main research objectives.