Moderating Variables in SmartPLS: How to Run and Interpret the Results

What is moderation in SmartPLS used for

A moderating variable, or moderator, changes the strength or direction of the effect of an independent variable on a dependent variable. The question you are testing is whether the effect of X on Y becomes stronger, weaker, or changes direction as W changes.

For example, X may be service quality, Y may be satisfaction, and W may be age. You are not only asking whether service quality affects satisfaction. You are also testing whether this effect differs across age groups or changes as age increases.

In SmartPLS, moderation is usually tested through an interaction variable. The general model is:

Y = β1X + β2W + β3(X × W) + ε

Here, X is the independent variable, W is the moderating variable, X × W is the interaction term, and β3 is the moderation coefficient you need to examine. If β3 is statistically significant, the relationship between X and Y depends on W. A path from X to Y with a small p-value is not enough to conclude that moderation exists when the interaction effect is not significant.

SmartPLS is suitable when your model uses PLS-SEM, includes latent variables measured by multiple indicators, and requires you to assess the measurement model and structural model together. You can also review the SmartPLS topic before starting if the PLS-SEM workflow is still unfamiliar.

Preparing the model and data

Distinguishing the measurement model from the structural model

The measurement model shows which observed indicators measure each latent variable. For example, CL1, CL2, and CL3 measure service quality. The structural model shows the hypothesized relationships between latent variables, such as service quality affecting satisfaction.

For moderation, W can be a latent variable such as trust or perceived risk. W can also be a single observed variable such as gender, age, or usage experience. The way you specify and interpret the model will differ, so identify what W represents before opening SmartPLS.

Naming indicators consistently

Each observed indicator must be placed in its own column in the data file. The first row contains the indicator names, and the following rows contain each respondent's answers. Keep the names short, without accents or spaces, such as CL1, CL2, NT1, and HL1.

Do not assign the same name to two indicators. Do not use special characters in column headers. If you use a Likert scale, the data are usually coded numerically in the same direction. Reverse-coded indicators must be processed before you import the file into SmartPLS. Otherwise, outer loading and reliability may be reduced.

Checking the CSV file

SmartPLS 4 can import data from .csv. Save the file with the appropriate delimiter, usually a comma or semicolon depending on your Excel settings. Check these three points before importing:

• Each row represents one respondent, and each column represents one variable.
• The data area contains no total rows, secondary headings, or merged cells.
• Missing data, text symbols, and values outside the scale have been reviewed.

Categorical variables such as gender must be coded numerically according to the coding scheme you report in the methods chapter. If the file contains a respondent ID column, you can keep it for checking records, but do not drag it into the model as an indicator.

If you are new to the platform, the articles Smart PLS and What is Partial Least Squares can help you distinguish PLS-SEM from running a direct regression in SPSS.

Steps for testing moderation in SmartPLS 4

Step 1: Create a project and import the data

Open SmartPLS 4, create a new project, and import the .csv file. Create a new model from the imported dataset. Drag the indicators into the correct constructs, then connect the indicators by giving the constructs clear names, such as CHAT_LUONG, NIEM_TIN, and HAI_LONG.

For reflective constructs, the arrows usually run from the construct to the indicators. Check the measurement direction before running PLS Algorithm. An incorrect arrow direction can lead you to misread the outer loading and reliability results.

Step 2: Draw the main relationships in the model

Draw a path from X to Y and from W to Y. Including W as a predictor of Y is necessary because a moderation model usually contains both the main effect of the independent variable and the main effect of the moderator. If you create only the interaction and omit the path W → Y, the model may be underspecified.

For example, the model may contain CHAT_LUONG, NIEM_TIN, and HAI_LONG. Draw CHAT_LUONG → HAI_LONG and NIEM_TIN → HAI_LONG. The interaction between CHAT_LUONG and NIEM_TIN will be created in the next step.

Step 3: Create the moderation effect

In SmartPLS 4, select the tool for creating a moderation effect from the model toolbar. Select the independent variable X, the moderating variable W, and the dependent variable Y. SmartPLS will create an interaction construct or interaction path depending on the model type and the interface version you are using.

Read the path name carefully after creating it. A name such as CHAT_LUONG * NIEM_TIN -> HAI_LONG identifies an interaction effect, not the main effect of CHAT_LUONG. If you have several moderators, create each interaction separately and use names that distinguish them.

SmartPLS 3 and SmartPLS 4 may place the tools differently. Do not spend time looking for the exact button position shown in an old screenshot. Look for the group of functions related to Create Moderation or Moderation Effect in the current model, then check the path structure on the canvas.

Step 4: Run PLS Algorithm

Select Calculate > PLS-SEM Algorithm. Keep the default settings if you do not have a methodological reason to change them. Run the model first to check outer loading, CR, AVE, and the path coefficients.

At this stage, do not use p-values to decide whether the hypotheses are supported. PLS Algorithm gives you estimated coefficients and measurement model information. If an indicator has a very low loading or a construct fails to meet the reliability requirement, resolve the measurement model issues before interpreting moderation.

Step 5: Run Bootstrapping

Select Calculate > Bootstrapping in SmartPLS 4. Set the number of bootstrap subsamples according to your research design. The current PLS-SEM procedure commonly uses 5,000 bootstrap subsamples to estimate standard errors and test paths (Hair et al., 2022).

Choose a two-tailed test if your hypothesis only states that an effect exists without specifying its direction. Choose a one-tailed test only when the hypothesis has a clear direction and your methodological justification supports that choice. After running the analysis, open the Path Coefficients table or the results section for the interaction effect.

Step 6: Check the simple slope when the interaction is significant

When the interaction effect is significant, you need to explain how W changes the relationship between X and Y. In SmartPLS, review the moderation plot or simple slope output if your version and analysis settings provide that function.

If the software does not produce the plot automatically, you can present the relationship at low, medium, and high levels of W using the options supported by the software. Do not write only that moderation exists and leave the direction unexplained. A positive β3 generally indicates that W strengthens the relationship between X and Y, while a negative β3 generally indicates that W weakens it. Your final interpretation still needs to follow the coding scheme and the model plot.

Reading the moderation results table

The table below is illustrative output, not the result of a real study. The column names follow the result groups commonly shown in SmartPLS. When reading your own file, focus on the interaction effect row and check it against the bootstrap settings.

The row CHAT_LUONG * NIEM_TIN → HAI_LONG determines the moderation test. In this illustrative example, P values of 0.019 indicate that the interaction effect is significant at the 5% level if the study uses this threshold. The coefficient of 0.146 is positive, which supports the interpretation that trust strengthens the relationship between service quality and satisfaction.

You also need to examine the bootstrap confidence interval. If the confidence interval for the interaction effect does not contain 0, the moderation conclusion receives additional support (Preacher and Hayes, 2008). Do not use the p-value from CHAT_LUONG → HAI_LONG as a substitute for the p-value of the interaction effect.

Then check the R Square of the dependent variable. After the interaction effect is added, R² may increase compared with a model containing only the main effects. You can also review What is SEM to understand where the structural model fits into the broader SEM workflow.

Evaluation thresholds in SmartPLS

The thresholds below apply specifically to PLS-SEM. CFI, TLI, and RMSEA belong to fit indices commonly used for CB-SEM, so they are not included in this SmartPLS evaluation table.

There is no single fixed number that tells you whether a moderating variable is good. Assess the sign of the interaction coefficient, statistical significance, the confidence interval, the f² of the interaction if available, and how the simple slope changes. A large coefficient may not be significant, while a small coefficient may still be significant in a specific research context.

You should also assess the interaction in the same model as the main effects. Removing X → Y or W → Y merely to make the interaction reach a target p-value is difficult to defend. If you remove an indicator, record the reason based on its loading, scale content, and effect on convergent validity, rather than removing it simply to make the results table look better.

How to write this in your thesis

Report the results in this order: measurement model results, structural model results, and then the moderation test. The reader needs to know that the scale meets reliability and validity requirements before accepting your interpretation of the relationships between variables.

A paragraph you can adapt for your thesis is: “The bootstrapping results show that the interaction between [independent variable] and [moderating variable] on [dependent variable] has a coefficient of β = [value], t = [value], and p = [value]. Because [the p-value is smaller than or larger than] [significance level], hypothesis H[ ] concerning the moderating role of [moderating variable] is [accepted or not supported]. The [positive or negative] sign of the interaction coefficient indicates that [describe how the relationship between X and Y changes].”

After this paragraph, add a table with the columns Hypothesis, Relationship, β, T statistics, P values, and Conclusion. If the interaction is significant, include a moderation plot or simple slope table in the appendix or Chapter 4, depending on your department's requirements.

Do not call W a “mediator” in this section. A mediator answers how X affects Y through M. A moderator answers how the strength or direction of the relationship between X and Y changes according to W. A control variable is included to hold the effects of other factors constant, so these three concepts need separate names in the model and hypotheses.

Common errors

Looking only at the path from X to Y

This is the most common error. A small p-value for X → Y only demonstrates the main effect in the sample. You must read the interaction row X × W → Y to decide whether W moderates the relationship.

Omitting the main effect of W from the model

If W is a moderator, consider including W → Y in the model together with X → Y and the interaction. When the main effect is missing, the interpretation of the interaction can raise questions about model specification.

Using CFI, TLI, or RMSEA for SmartPLS

These indices belong to CB-SEM. SmartPLS uses indices appropriate for PLS-SEM, including outer loading, CR, AVE, HTMT, VIF, R², f², Q², and the path coefficient. If your proposal requires fit indices, discuss the SEM tool and method again instead of mixing criteria from the two approaches.

Changing the data to make the interaction significant

Do not delete respondents simply because they make the p-value larger. Process missing data, outliers, or invalid responses only when you have predefined criteria and can explain the decision. If the interaction is not significant, report that it was not supported and discuss the result within the limitations of the sample.

Confusing a categorical variable with a reflective construct

Gender or age group can be coded as an observed variable, but creating an interaction is different from creating a construct measured with multiple indicators. Check the SmartPLS guidance for the version you use and keep the specification consistent with the methods model in your proposal.

Frequently asked questions

What is a moderating variable?

A moderating variable changes the strength or direction of the effect of X on Y. In a SmartPLS model, the key result is the interaction effect between X and W, not only the path from X to Y.

What is a good level for a moderating variable?

There is no single β value that is good for every topic. Check whether the interaction effect is statistically significant, whether the confidence interval excludes 0, whether the coefficient sign matches the hypothesis, and whether the simple slope has a clear interpretation.

How are a moderating variable and a mediating variable different?

A mediating variable explains the mechanism through which an effect is transmitted, in the form X affects M and then M affects Y. A moderating variable changes the relationship X → Y according to the level of W. A mediation model therefore focuses on the indirect effect, while a moderation model focuses on the interaction effect.

Where do you run moderation in SmartPLS 4?

Create the model, draw X → Y and W → Y, and then use the Moderation Effect function to select X, W, and Y. Next, run PLS-SEM Algorithm and then Bootstrapping. SmartPLS 3 and 4 may use different menu locations, so check the version installed on your computer.

What should you do if the interaction effect is not significant?

First check the data coding, indicator direction, variable type, sample size, model, and bootstrap settings. If the procedure contains no error, report that the interaction was not supported. Do not delete variables or alter the data to obtain the desired p-value. You can discuss that the moderating role was not demonstrated in the current sample.

Open your data file now, check the indicator names, redraw the X → Y and W → Y paths, and run the interaction effect separately in SmartPLS 4. If you need to run the analysis on your own .sav or .csv file and track each step, you can use the M4 data analysis module.