What is the difference between moderation and mediation?

Moderation changes the strength or direction of a relationship depending on a third variable. Mediation explains a relationship through an intervening variable that sits on the path between predictor and outcome. Moderation is about when an effect holds; mediation is about how it works.

When should I use moderation instead of multi-group analysis?

Use a continuous interaction term (moderation) when the moderator is continuous, because splitting it into groups for MGA throws away information. Use MGA when the grouping is genuinely categorical, such as country or gender.

What is a simple-slopes plot?

A plot of the predictor-to-target relationship computed at low, mean and high values of the moderator. It turns a significant interaction coefficient into a concrete picture of how the effect changes.

PLS-SEM METHOD

Moderation in PLS-SEM: when an effect depends on a third variable

Moderation analysis tests whether the strength or direction of a relationship between two constructs changes depending on the value of a third construct, the moderator. This page explains the interaction effect, how to read it and simple slopes, the estimation approaches, and the common pitfalls.

What a moderation analysis tests

In a plain model, a predictor affects a target with one fixed coefficient. Moderation asks whether that coefficient itself depends on a moderator variable. For example, the effect of perceived usefulness on adoption might be stronger for younger users than older ones: age moderates the usefulness-to-adoption path.

Mechanically, moderation adds an interaction term, the product of the predictor and the moderator, as an extra predictor of the target. A significant interaction effect means the predictor-to-target relationship changes as the moderator changes.

A worked example

The effect of trust on loyalty is tested with satisfaction as the moderator. The interaction term is positive and significant (interaction coefficient 0.16, p = 0.01). Plotting simple slopes shows the trust-to-loyalty slope is shallow when satisfaction is low and steep when satisfaction is high.

So trust matters more for loyalty among highly satisfied customers: satisfaction strengthens the relationship.

Simple slopes: the trust-to-loyalty slope steepens as the moderator (satisfaction) rises, a positive interaction.

How to read the result

Moderation reports the interaction effect: its coefficient, standard error, t-value and p-value, usually from bootstrapping.

Interaction p < 0.05: Significant moderation: the predictor-to-target effect genuinely depends on the moderator. Interpret the interaction, not just the main effects.
Positive interaction: The predictor-to-target relationship gets stronger as the moderator increases.
Negative interaction: The relationship gets weaker, or reverses, as the moderator increases.
Interaction p ≥ 0.05: No evidence of moderation. Report the direct effect and do not interpret the interaction.

Read a significant interaction with a simple-slopes plot: it shows the predictor-to-target slope at low, mean and high values of the moderator, which makes the direction and size of the moderation concrete.

Estimation approaches

Two-stage

Estimates the latent variable scores first, then forms the interaction term from those scores. Robust and the default in OpenPLS; it works with both reflective and formative constructs.

Product indicator

Builds the interaction by multiplying the indicators of the predictor and the moderator. Requires reflective constructs and works best with equal indicator counts.

Orthogonalizing

A variant of the product-indicator approach that removes the correlation between the interaction term and its components to reduce collinearity.

Simple slopes

Not an estimation method but the standard way to interpret a significant interaction: plot the predictor-to-target relationship at different moderator levels.

Running a moderation step by step

Specify the direct path from predictor to target that you think is moderated.
Add the moderator and the interaction term (predictor times moderator) as predictors of the target.
Estimate the model and bootstrap the interaction effect to get its t-value and p-value.
If the interaction is significant, plot simple slopes to see how the effect changes across moderator levels.
Report the interaction coefficient, its significance and the simple-slopes interpretation, and the f² effect size of the interaction where possible.

Common pitfalls

Interpreting the main effect as unconditional when a significant interaction is present. With moderation, the main effect is the effect at the mean of the moderator only.
Splitting a continuous moderator into groups and running a group comparison instead. That discards information; a continuous interaction term is usually better.
Reporting only the interaction coefficient without a simple-slopes plot, which leaves the direction and magnitude unclear.
Ignoring the f² effect size of the interaction, which is often small even when the interaction is significant.

Moderation in OpenPLS

OpenPLS includes a moderation panel that uses the two-stage approach. Pick the predictor, the moderator and the target, and it adds the interaction term, reports the base and interaction-model paths, and gives the interaction effect with its bootstrap t-value and p-value, together with a simple-slopes plot. Free and runs in the browser.

Test a moderation on your own model

Build your model in the visual editor, add a moderator, and get the interaction effect and simple slopes in one run. Free, no install.

Open OpenPLS What is PLS-SEM?

Frequently asked questions

What is the difference between moderation and mediation?: Moderation changes the strength or direction of a relationship depending on a third variable. Mediation explains a relationship through an intervening variable that sits on the path between predictor and outcome. Moderation is about when an effect holds; mediation is about how it works.
When should I use moderation instead of multi-group analysis?: Use a continuous interaction term (moderation) when the moderator is continuous, because splitting it into groups for MGA throws away information. Use MGA when the grouping is genuinely categorical, such as country or gender.
What is a simple-slopes plot?: A plot of the predictor-to-target relationship computed at low, mean and high values of the moderator. It turns a significant interaction coefficient into a concrete picture of how the effect changes.