What the method does
Conditional process analysis combines mediation and moderation. It asks how an indirect pathway changes with a moderator rather than reporting a single indirect effect for everyone. The relevant effect is derived from the model’s constituent paths at stated moderator values, with uncertainty for that derived quantity.
Example first-stage moderation: indirect effect at W = w is (a₁ + a₃w)b.
a₁ is the predictor-to-mediator coefficient, a₃ its interaction with W, and b the mediator-to-outcome coefficient. Other moderation arrangements have different expressions. State which path is moderated and the moderator values used.
Data and analysis in QuickPLS
Provide an eligible model with defined predictor, mediator, outcome and moderator roles. Keep constituent main effects and a defensible scale or reference value for the moderator.
QuickPLS supports the conditional-process specification and its derived effects. Read the named effect and moderator value, then use the corresponding uncertainty rather than the significance of a single constituent path.
Follow a worked example
The following example shows the data, settings and decisions for one analysis. Use the linked practice files to follow it. For your own study, keep the procedure and choose the variables, model and settings that fit your research question.
Download research-workshop.csv to begin. It contains 360 rows and 31 columns. The example below identifies the variables to use, their roles and any rows to exclude; not every column is an analysis variable.
Ask how the whole indirect pathway changes
Suppose Quality is related to Loyalty through Satisfaction, and Trust changes the Quality-to-Satisfaction relationship. You need the indirect relationship at different Trust values. Conditional-process analysis follows the whole indirect pathway while allowing the specified stages to change with a moderator. A moderator is a variable that changes the size or direction of another relationship.
The first stage enters the mediator; the second leaves it for the outcome. Either stage, both stages or supported longer paths can be moderated. State the complete pathway, mediator order and evaluation values before calculating. Main effects, interaction terms, groups, weights and higher-order constructs must form one supported specification, not just a collection of individually available features.
Check the path and interactions on the model
Write the indirect path in order, then mark which arrow each moderator affects. Check that these interactions and their required main effects already exist in the PLS model. This calculation selects and evaluates relationships from the model; it does not create a missing interaction for you.
Use this route for the conditional paths of a PLS model, including its indicator-based constructs. The observed-variable Graph-defined Path Analysis / PROCESS route under Regression is a different calculation. If adding groups, weights, a higher-order construct or a three-way interaction, review the complete profile before fitting. A higher-order construct represents a broader concept through lower-order dimensions; a three-way interaction lets one moderation relationship itself vary with another moderator.
Select the pathway, moderator values and effects
Build and save the moderated workshop model before opening this calculation. The setup selects an existing indirect path and its interactions; it does not add the Trust moderation effect to the diagram. The first step in the worked exercise shows how to create that effect.
Open the PLS conditional-process setup
Choose Calculate, select PLS-SEM Conditional Process and continue to its setup. Choose a Profile envelope, meaning the supported combination of model features and interval method. For the first exercise, use Multiple two-way · percentile.
Choose the complete path and evaluation values
Select the whole pathway under Explicit indirect paths and its interaction under Authored interaction terms. Enter the moderator values under Frozen original-sample probes and check Probe metric. A probe is a value at which you want the effect evaluated. Under Requested estimands, select the quantities that answer your question; estimand means the effect or comparison you intend to estimate.
Calculate and read matching effect rows
Set Outer resamples and Alternative for the selected profile, review the interval summary, then calculate. Read each conditional indirect effect, eligible index or requested contrast with its own interval. Check the named pathway and probe values so that a result from one condition is not described as a result for all conditions.
Distinguish an effect at one value from a change between values
A conditional specific indirect effect follows one selected pathway at stated moderator values. A conditional total indirect effect combines the relevant pathways. These are effects at a condition. An index, derivative or contrast describes how the indirect relationship changes; it answers a different question.
A scalar index is one constant rate of change. It is appropriate only when the complete indirect-effect function is linear in one moderator. A local first derivative is the rate of change at a particular value, useful when that rate itself changes. A finite contrast subtracts the effect at one declared condition from the effect at another. Read the exact definition and matching interval rather than calling every row a moderation test.
An unavailable result means its calculation requirements were not met; it is not a numerical zero. Check the combined profile and the bootstrap, jackknife or inner-fit status before deciding how to describe the result.
On a small screen, swipe the table sideways to see every column.
| Output | How to read it | What to check |
|---|---|---|
| Conditional indirect effect | The indirect relationship at stated moderator values. | A single unconditional product can miss the research question. |
| Index / contrast | The defined change in an indirect relationship across conditions. | Interpret the actual contrast rather than a generic moderation label. |
| Bootstrap interval / eligibility | Uncertainty for the chosen specification and whether its calculation requirements were met. | Features that are available separately may not be available together. |

Use the conditional coefficient at every affected stage
Follow the pathway from start to finish at the same stated moderator values. If only the first arrow changes, combine that conditional coefficient with the unchanged later coefficient. If both arrows change, evaluate both before multiplying them.
This distinction matters because a product of changing coefficients may be curved rather than linear. A constant index can then misdescribe the pattern even though each individual equation is straightforward.
Illustrative example, not a software benchmark
If only the first stage is moderated, the indirect relationship changes through that stage while the later path remains as specified. If both stages are moderated, using one conditional slope with an unconditional coefficient would misstate the effect. Follow the complete pathway at the chosen moderator values.
Practice: does Trust change the Quality-to-Loyalty pathway?
The practice question is whether the Quality-to-Loyalty indirect relationship through Satisfaction changes with Trust. Trust moderates the first stage, Quality → Satisfaction. An interaction coefficient alone does not describe the whole indirect pathway; the later Satisfaction → Loyalty relationship also enters the calculation.
Download research-workshop.csv. Build reflective Quality, Trust, Satisfaction and Loyalty from their three matching indicators. Draw Quality → Satisfaction, Trust → Satisfaction, Satisfaction → Loyalty and Quality → Loyalty. Add a two-way interaction in which Trust moderates Quality → Satisfaction, retaining both main effects. The direct Quality → Loyalty path lets you distinguish direct and indirect associations.
These synthetic data illustrate an associational model. The hand calculations below use hypothetical coefficients and are not expected fitted results from the download. They teach how to read the function and its comparisons.
Run the first-stage exercise
Check the complete model
Confirm Quality → Satisfaction → Loyalty and the direct Quality → Loyalty path. Keep both Quality and Trust main effects in the Satisfaction equation. To create the first-stage interaction, open Model → Moderating Effect…; the command may also read Moderating Effect (Creates Model Revision)…. Set Relationship to Quality → Satisfaction, Moderator to Trust and Construction to Two-stage, then choose Add moderating effect. Save the revised model. The conditional-process setup can select only interactions already present in that model.
What you should see: The full model contains the mediator, moderator and required main effects.
How to read it: An omitted constituent term changes the effect definition.
Select the profile and pathway
Open PLS-SEM Conditional Process and continue to setup. Under Profile envelope, select Multiple two-way · percentile. Under Explicit indirect paths, select the entire two-edge Quality → Satisfaction → Loyalty path. Under Authored interaction terms, select the relevant Trust interaction. Select the planned relationships rather than every available path.
What you should see: The setup names the intended pathway and procedure.
How to read it: Confirm which stage of mediation is moderated.
Enter the Trust probes
Under Frozen original-sample probes, enter -1, 0, 1 for Trust and set Probe metric to Standardized score. These values mean one standard deviation below the original score mean, the mean, and one standard deviation above it. Check that they represent a sensible range in the sample. They are construct-score values, not raw questionnaire responses.
What you should see: Trust probes are within the meaningful score range.
How to read it: They define where conditional effects are evaluated.
Select effects and the eligible index
Request Conditional specific indirect effects and Scalar index only when affine in one moderator. Affine here means a straight-line function with an intercept, so its rate of change is constant. That fits this first-stage example. For a curved function or multiple moderators, use Local first derivatives or Finite probe contrasts to describe the actual change instead.
What you should see: Requested effects and the eligible index match the model.
How to read it: Not every moderation arrangement has the same index.
Set the resampling plan
Use 1000 Outer resamples and Alternative Two-sided for practice. Read the Interval summary and record the confidence level in the saved settings or result; this panel does not offer a separate confidence input. Percentile uses cut-offs from the bootstrap estimates. Full delete-one BCa also refits after omitting each case. Studentized inference adds inner resampling to scale estimates by their uncertainty. Keep the method name accurate if a requested calculation fails.
What you should see: The resampling plan records draws, interval choice and seed.
How to read it: Derived effects need their own uncertainty.
Read the conditional effects and the change itself
Choose Start calculation and open the completed result. Read Frozen probe values, then Conditional-process targets. Inspect the full indirect effect at every Trust value, followed by the eligible index or declared contrast with its own interval. If the low-probe interval includes zero and the high-probe interval excludes zero, that alone does not show a supported difference between the two effects. Use the direct comparison's uncertainty.
What you should see: The output distinguishes conditional effects from their change across probes.
How to read it: Interpret each named quantity on its stated scale.
Work through a changing indirect effect by hand
Suppose the Quality → Satisfaction coefficient is 0.30 + 0.10w, where w is standardized Trust, and Satisfaction → Loyalty is 0.50. Multiplying gives (0.30 + 0.10w) × 0.50 = 0.15 + 0.05w. At w = -1, 0 and 1, the indirect values are 0.10, 0.15 and 0.20. Its constant index is 0.05: the change in the indirect association for one unit of standardized Trust.
Now suppose the hypothetical 95% interval for the index is [-0.01, 0.11]. The point estimates increase with Trust, but the interval includes zero and allows considerable uncertainty about that change. Report the estimated pattern and its imprecision. Do not claim supported moderated mediation merely because one conditional-effect interval excludes zero and another does not.
If Trust moderates both stages, suppose their coefficients are 0.30 + 0.10w and 0.50 + 0.20w. The product is 0.15 + 0.11w + 0.02w². Its local rate of change is 0.11 + 0.04w, which varies with Trust. One constant Hayes index no longer describes the function. At w = -1 and 1, the indirect effects are 0.06 and 0.28; the high-minus-low contrast is 0.22. This contrast needs its own interval.
Finite probe contrasts use the effect at the left declared joint probe minus the effect at the right. Put the high condition on the left if you want high minus low. With multiple moderators, Explicit joint tuples let you name selected combinations of values. They replace the Cartesian grid, which would otherwise use every combination of the entered values. Choose combinations that make sense for the study rather than filling the report with implausible conditions.
Check the pathway, scale and requested comparison
Start with the model if an interaction or index is missing. Is the selected path complete, are its main effects present, and is its overall indirect function actually linear in one moderator? A missing constant index can be the correct result for a curved function; use an appropriate derivative or contrast.
If the effects are available but weak or uncertain, inspect measurement quality, the observed moderator range and the interval for the change you planned to test. If BCa or studentized inference is unavailable, inspect the extra jackknife or inner-bootstrap fits. Keep the distinction between an unsupported hypothesis and an unavailable calculation clear.
On a small screen, swipe the table sideways to see every column.
| Symptom | Check | Reasonable action |
|---|---|---|
| No scalar index appears | Whether the complete path function is linear in one moderator | Use local derivatives or explicit contrasts for nonlinear or multiple-moderator effects. |
| The interaction is absent from the selected analysis | Whether the specified interaction and its lower-order terms are present | Correct the model specification before calculating; do not infer moderation from separate subgroup fits. |
| A probe has no sensible interpretation | Score scale, observed range and moderator combinations | Choose justified evaluation values and state the scale. Avoid extrapolated extremes. |
| One conditional effect is significant and another is not | The direct contrast or eligible index interval | Test the change itself rather than comparing significance labels. |
| BCa or studentized inference is unavailable | Failures in leave-one-case-out or inner-bootstrap fits, and usable counts | Report the missing calculation and diagnose it; no automatic percentile replacement is implied. |
| A combined profile is blocked | Interactions, groups, weights and higher-order requirements together | Revise the analysis explicitly and explain how the revised question differs. |
Choose probes for their meaning, not their significance
Choose moderator values from the study question and observed data before looking at intervals. A useful sensitivity check asks whether plausible alternative anchors lead to the same substantive interpretation, while keeping the pathway and scale clear. Searching many values and reporting only favourable intervals hides both uncertainty and the number of comparisons.
For a mechanism claim, examine when each variable was measured, plausible omitted common causes and alternative pathways. A PLS conditional-process result describes associations unless the design and assumptions support a causal interpretation. Statistical support for the indirect association alone does not show that intervening on Quality would change Loyalty through Satisfaction.
Describe the whole conditional pathway and its uncertainty
Methods example: We estimated the Quality to Satisfaction to Loyalty indirect relationship with Trust moderating the first stage. The model retained the lower-order terms and the direct Quality to Loyalty path. We evaluated Trust at standardized values -1, 0 and 1, fixed from the original sample, using the multiple-two-way percentile profile with 1000 outer resamples and two-sided inference. Add the actual confidence setting, sample size and usable and failed counts.
Hypothetical Results example: The conditional indirect estimates were 0.10, 0.15 and 0.20 across the three Trust values. The constant index was 0.05, with an illustrative interval of [-0.01, 0.11]. The estimated association increased with Trust, but the rate of change remained imprecise. We assessed that change using its own interval rather than comparing separate conditional significance labels.
For a more complex model, replace the constant-index sentence with the actual derivative or ordered contrast you estimated. Keep the effect definition, scale and interval together so readers can reconstruct the comparison.
| Report item | Detail to include |
|---|---|
| Pathway | Ordered constructs and every moderated stage |
| Specification | Main effects, interactions, direct paths and selected profile |
| Evaluation | Probe values, score scale and original-sample anchors |
| Effect definition | Conditional effect, eligible index, derivative or ordered contrast |
| Inference | Interval method, confidence, alternative, resample counts and failures |
| Interpretation | Associational scope and research-design assumptions |
Can you explain the effect and the comparison?
- Reproduce the hypothetical first-stage indirect values 0.10, 0.15 and 0.20 from the two path coefficients.
- Explain why moderating both stages can produce a curve and make a constant index inappropriate.
- State which declared condition is subtracted from which in a finite probe contrast.
- Explain why raw observed values and standardized construct scores are different scales.
- Check the joint requirements for interactions, paths, groups, weights or higher-order constructs in a specialist profile.
- Use the interval for the index or contrast itself when assessing a change in the indirect association.
Tell readers exactly which conditional effect you estimated
- Name the complete indirect path, each moderated stage and the main effects, interactions and direct paths retained in the model.
- Give the evaluation values, their scale and any ordered contrast. Explain why those values represent meaningful conditions in the study.
- Report the profile, named interval procedure, confidence level, test direction and usable or failed resamples. Match uncertainty to the effect being discussed and state what the design supports about causal interpretation.
Keep the statistical pathway and the causal argument distinct
A conditional indirect association describes the specified model. Calling it a mechanism additionally needs a credible account of timing, measurement and common causes. An interaction or an interval excluding zero does not supply that study-design argument.
If a combination is unavailable, choose a method that addresses it or explicitly revise the question. Dropping a moderator, a weight or a stage changes the reported effect. Keep the original specification and explain any revision instead of presenting the simpler result as the original analysis.
Choose advanced profiles as complete analyses
For a longer indirect path, multiply the conditional coefficient of every arrow in its stated order. Standard multiple-two-way profiles support selected paths of two to six edges, subject to their interaction, path and moderator limits. An edge is one directed arrow. Adding or removing stages changes the pathway being estimated.
A bounded three-way profile needs the complete lower-order hierarchy: the relevant main effects and two-way interactions as well as the three-way term. It uses percentile inference. Higher-order constructs must use one supported approach and be disjoint and nonnested, so the selected higher-order structures do not overlap or contain one another. Grouped analysis resamples within the original groups. Positive case weights change each row's contribution; integer frequency weights represent repeated counts. Their resampling rules differ, and neither automatically supplies a complex-survey correction.
These options are complete profiles rather than switches that can be combined freely. Higher-order constructs plus groups, groups plus weights, and three-way terms plus higher-order constructs fall outside the stated combined profiles. The studentized profile has tighter limits and adds inner resamples. Choose the appropriate method or describe a justified redesign; do not silently remove a term and keep the original research claim.
Raw observed units are available for an eligible observed single-indicator moderator with a recorded scale transformation. Choose Raw observed units and use Prepare exact metric when that option is applicable; this is different from pretending a multi-item construct score is a raw questionnaire response. The original-sample anchors stay fixed during resampling, so each effect is evaluated at the same declared condition rather than a value that moves with each bootstrap mean.