How to Run Mediation Bootstrap in SPSS with PROCESS Macro
When to use mediation bootstrap in SPSS
Use mediation bootstrap when you want to test whether variable X affects variable Y through variable M. In this model, X is the independent variable, M is the mediator, and Y is the dependent variable. For example, you may want to test whether service quality increases purchase intention through satisfaction.
The central question is whether the indirect effect of X on Y through M differs from 0. PROCESS Macro estimates two main relationships, X to M and M to Y after X has been included in the model, then uses bootstrap resampling to create a confidence interval for the indirect effect.
You should distinguish a mediator from a moderator. A mediator explains the mechanism through which an effect occurs, while a moderator changes the strength or direction of the relationship between two variables. If you are unsure how to assign the roles, see what is a mediator, what is an independent variable, and what is a control variable.
In quantitative research, bootstrap is often preferred to an inference procedure based only on the Sobel test because the indirect effect often has a skewed distribution. Bootstrap creates many resamples from the original data, calculates the indirect effect in each sample, and then uses percentiles to create a confidence interval. Under bootstrap reporting, the indirect effect is considered significant when the confidence interval does not contain 0 (Preacher and Hayes, 2008).
Prepare the data before running the analysis
Identify the roles of X, M, and Y correctly
Before opening SPSS, write the model on one line: X → M → Y. For example, Service quality → Satisfaction → Purchase intention. If each construct has multiple items, you will usually need to calculate a representative score for each scale before running PROCESS, such as the mean of the items after checking scale reliability and validity.
A mediation model is meaningful only when there is a theoretical basis for the paths from X to M and from M to Y. PROCESS does not prove a causal relationship simply because the output contains a small p-value. Compare the model with the hypotheses developed in Chapters 2 and 3.
Check data types and missing values
In Data View, X, M, and Y must be numeric variables measured as Scale, or variables that can be treated as continuous in the model. If the data have been entered as strings, PROCESS may not recognize the variables or may produce an incorrect result. Open Variable View to check Type, Decimals, Measure, and the variable labels.
You also need to check missing values. Open Analyze > Descriptive Statistics > Frequencies or Analyze > Descriptive Statistics > Descriptives to review the number of valid observations. Do not automatically recode a missing value as 0 when 0 is not a real response. Record the number of rows removed and the reason for removal so that you can explain it when writing the thesis.
Check reverse coding and scale scores
If the questionnaire contains items written in the opposite direction, reverse code them before calculating the scale score. An item that has not been reverse coded can distort the mean score of M or X and affect the indirect effect.
You should also review descriptive statistics, outliers, and correlations between the variables. A high correlation does not automatically prove mediation, and a low correlation does not always make the model meaningless. This step is for detecting data errors, not for replacing the bootstrap test.
Steps to run mediation bootstrap in SPSS
Install and open PROCESS Macro
PROCESS Macro is an extension for SPSS used to run mediation, moderation, and conditional process analyses. You can read what is PROCESS Macro before installing it to understand how it differs from standard regression commands.
After installing a version compatible with SPSS, PROCESS usually appears at Analyze > Regression > PROCESS v4.x. The version name may differ depending on the installation file, but the main menu remains under Regression. If PROCESS does not appear, check the installation and restart SPSS.
Select the variables and mediation model
Open Analyze > Regression > PROCESS v4.x. In the main dialog box, place the dependent variable in Y Variable, the independent variable in X Variable, and the mediator in M Variable(s). For the model X → M → Y, put Y in Y, X in X, and M in M.
In Model Number, enter 4. Model 4 is a simple mediation model for one mediator between X and Y. If you have two serial or parallel mediators, do not automatically use Model 4 because the hypothesized structure is different.
Select bootstrap and statistical options
In the options section, set Bootstrap samples to 5000. This number of repetitions is recommended in the PLS-SEM procedure, and in PROCESS you should also state the number of bootstrap samples used in the output and thesis. Reporting 5,000 bootstrap subsamples can be supported with (Hair et al., 2022) when it fits the research design.
Select the display options for Total effect model, Direct and indirect effects, and the bootstrap confidence interval. You can select Standardized coefficients if you want to interpret standardized coefficients. Select Effect size as well if your PROCESS version provides this option and your study requires effect-size reporting.
If the model includes control variables, move them into Covariates. Control variables should come from the research model. Do not add them simply to make the p-value look better. Check the order of X, M, and Y once more before clicking OK.
Save and check the output
SPSS will generate output containing a model description, regression tables, and the Indirect effect(s) of X on Y table. Save the .spv file immediately after running the analysis. If you run several versions, name the files by date or analysis round so that you do not confuse the results.
Do not save only a screenshot of the BootLLCI and BootULCI row. Keep the Model Summary table, the coefficient table for the M equation, and the coefficient table for the Y equation as well. The reader needs to see how the model was assessed.
Reading the output
PROCESS usually presents two regression equations. The first predicts M from X and the control variables, if any. The second predicts Y from X, M, and the control variables. In the output, pay attention to R-sq, Coeff, SE, t, and p in each equation.
The most important table is Indirect effect(s) of X on Y. In this table, Effect is the indirect effect, BootSE is the bootstrap standard error, and BootLLCI and BootULCI are the lower and upper bounds of the bootstrap confidence interval. If BootLLCI and BootULCI are both positive or both negative, the interval does not contain 0. You then have statistical evidence that the indirect effect differs from 0.
Model Summary output
R-sq shows the proportion of variance in the predicted variable explained in each equation. For the M equation, it is the proportion of M explained by X and the control variables. For the Y equation, it is the proportion of Y explained by X, M, and the control variables.
R-sq is not the only conclusion in a mediation analysis. A model with a modest R-sq can still have a significant indirect effect, while a high R-sq does not automatically prove a mediation mechanism.
Model output
In the coefficient table, the X row in the equation predicting M helps you examine the X → M relationship. In the equation predicting Y, the M row shows the M → Y relationship when X is included in the same model. The X row in the Y equation represents the direct effect of X on Y after M has been included.
Distinguish Total effect of X on Y from Direct effect of X on Y. The total effect is the overall effect before separating the path through M. The direct effect is the remaining effect of X on Y after controlling for M.
Indirect effect(s) of X on Y output
The example below is illustrative output, not the result of a specific study:
| Effect | BootSE | BootLLCI | BootULCI | Illustrative conclusion |
|---|---|---|---|---|
| 0.184 | 0.052 | 0.091 | 0.298 | The confidence interval does not contain 0, so the indirect effect is significant |
In this illustrative table, the indirect effect is 0.184. The 95% bootstrap confidence interval runs from 0.091 to 0.298, and both bounds are greater than 0. You can therefore conclude that the indirect effect is statistically significant at the corresponding confidence level. This only shows how to read the output and is not a finding from your data.
If BootLLCI is -0.04 and BootULCI is 0.08, the confidence interval contains 0. You then do not have enough evidence to conclude that the indirect effect differs from 0. Do not call M a significant mediator simply because X and Y are correlated or because one separate regression path has a small p-value.
Evaluation criteria
The table below focuses on the criteria commonly used when reading mediation bootstrap results in SPSS. Each row gives the source so that you do not copy a threshold without its reference.
| Item to check | How to read it | Criterion or rule | Source |
|---|---|---|---|
| Indirect effect | Check BootLLCI and BootULCI | The bootstrap confidence interval does not contain 0 | (Preacher and Hayes, 2008) |
| Number of bootstrap samples | Check the selected option and the method description | 5,000 bootstrap subsamples may be used when reported under an appropriate procedure | (Hair et al., 2022) |
| Direct effect | Check the X row in the equation predicting Y | Interpret it together with the indirect effect, not by itself as proof of mediation | (Baron and Kenny, 1986) |
| Indirect effect and direct effect | Compare the sign and statistical significance | Full or partial mediation may be described according to the model and results, but not inferred only from whether the direct effect is significant | (Baron and Kenny, 1986) |
| Practical effect | Check the coefficient magnitude or effect size if the output provides it | Interpret it in the context of the scale, model, and research setting, without assigning a small or large label without support | (Cohen, 1988) |
Use the term “full mediation” carefully. The classic logic often relies on the direct effect becoming smaller or no longer significant after M is added (Baron and Kenny, 1986), but the key decision in bootstrap analysis remains the confidence interval for the indirect effect. Report the direct effect, total effect, and indirect effect instead of using only one label.
If you use control variables, state clearly that they were included in each relevant equation according to the PROCESS settings. Do not compare models with and without control variables unless you explain the theoretical reason and how the model was changed.
How to write this in your thesis
You can use the following template and replace every bracketed part with the actual values from the output:
How to write this in your thesis: “The PROCESS Macro results with Model 4 and [number of bootstrap samples] bootstrap samples show that the indirect effect of [X] on [Y] through [M] is [Effect], with a 95% bootstrap confidence interval from [BootLLCI] to [BootULCI]. Because the confidence interval does not contain 0, the indirect effect is statistically significant. Therefore, hypothesis [H] is [accepted/supported] at the [significance level] significance level.”
If the confidence interval contains 0, write: “The 95% bootstrap confidence interval for the indirect effect ranges from [BootLLCI] to [BootULCI] and contains 0. Therefore, there is not enough statistical evidence to conclude that [M] transmits the effect of [X] on [Y].”
After this paragraph, add a summary table containing the coefficient, BootSE, BootLLCI, BootULCI, and the conclusion for each hypothesis. Do not write “the effect increases by [X]%” unless you have an interpretation that fits the measurement units. A coefficient of 0.184 in the output does not automatically mean an increase of 18.4%.
If your supervisor requires a Sobel test, you can read Sobel test analysis in SPSS. State clearly which method was used for the main result, and do not combine the bootstrap confidence interval with the Sobel conclusion in the same row.
Common errors when running mediation bootstrap in SPSS
Choosing the wrong Model Number
Model 4 fits a simple mediator. If you have a moderator, serial mediators, or several parallel mediators, select the corresponding model and redraw the path diagram before running the analysis. Entering Model 4 for every study may produce output, but the output may not answer the hypothesis correctly.
Reversing X and Y
If you place the dependent variable in X and the independent variable in Y, PROCESS can still generate result tables. The software does not know your theoretical model. Before clicking OK, read the hypothesis again as a sentence: “X affects Y through M.”
Looking only at the p-value for each path
A small p-value for X → M and a small p-value for M → Y do not replace the indirect effect table. The mediation conclusion should be based on the bootstrap confidence interval for the indirect effect.
Changing the data to reach a threshold
Do not delete a row simply because it makes BootLLCI contain 0. Remove data only under a predefined cleaning criterion, such as an invalid response or missing data handled according to the rule stated in the method. Save each file version and record the reason for removal.
Using a SmartPLS or CB-SEM threshold by mistake
PROCESS in SPSS is a regression and conditional process bootstrap tool. Measures such as AVE, HTMT, and outer loading belong to measurement-model assessment in PLS-SEM, while CFI, TLI, and RMSEA belong to CB-SEM. Do not place these thresholds in a PROCESS interpretation table when your model was not run in SmartPLS or AMOS.
Frequently asked questions
What is mediation bootstrap in SPSS?
It is a way to use bootstrap resampling to estimate and test the indirect effect of X on Y through M in SPSS, usually with PROCESS Macro. The conclusion focuses on the confidence interval for the indirect effect, especially the two bounds BootLLCI and BootULCI.
How many bootstrap samples are sufficient for mediation in SPSS?
You can select 5,000 bootstrap samples and state this number in the method section. More repetitions do not make the data better and do not guarantee that the hypothesis will be significant. Check the model setup, data quality, and confidence interval for the indirect effect.
Does a negative BootLLCI and BootULCI indicate mediation?
The indirect effect can be considered different from 0 when both bounds are negative and the confidence interval does not contain 0. Interpret the negative sign according to the coding direction of X, M, and Y. Check the reverse-coded items before writing that the effect is negative.
What should I do when the bootstrap confidence interval contains 0?
You do not have enough statistical evidence to conclude that the indirect effect differs from 0 in the current sample. Do not increase the number of bootstrap samples, change the significance level, or delete observations simply to remove 0 from the confidence interval. Check the coding, missing values, variable roles, and theoretical basis, then report the result as it appears in the output.
Do I need to run the Sobel test together with bootstrap?
You do not always need to run both. Bootstrap is often used as the main result because it directly creates a confidence interval for the indirect effect. If your supervisor requires the Sobel test, you can report it as an additional analysis, but keep the two methods separate so readers do not mistake them for the same test.
Open the .sav file, check X, M, and Y again, and run PROCESS with Model 4 before copying any number into Chapter 4. If you want to run the analysis on your own SPSS or CSV data and retain every output file, M4 data analysis is the appropriate module for this step.