Do fairness and belonging answer different employee questions?
Employees may connect responsible organisational behaviour with fair treatment, a sense of belonging or both. Job satisfaction and organisational commitment are also different outcomes. A useful mediation study keeps those concepts distinct instead of treating every positive workplace response as the same result.
Zhang and colleagues analysed 385 international employees using confirmatory measurement analysis and PROCESS models. Their paper examined fairness and identification pathways from CSR to employee attitudes, including moderated mediation. The underlying survey was collected in 2016 despite the paper's recent publication. This lesson creates two original observed-score exercises and does not reproduce the paper's scales or estimates.
Prepare two outcome questions on one fictional employee file
Use 260 fictional employees. CSR measures perceived responsible conduct, FAIR measures overall fair treatment, ALLOC measures perceived fairness of resource allocation, SAT measures job satisfaction, BELONG measures belonging and COMMIT measures organisational commitment. All are continuous 1–7 teaching assessments.
The first question is whether CSR relates to SAT through FAIR, with ALLOC changing FAIR → SAT. The second is whether CSR relates to COMMIT through BELONG. They are two separate analyses with different mediators and outcomes, not one generic employee-attitude mechanism.
FAIR and ALLOC are deliberately distinct definitions: one concerns an overall judgement of treatment, while the other concerns allocation. In real measurement, overlapping item content could make their separate roles difficult to estimate. The fictional scores do not validate that distinction.
The exercise treats rows as independent fictional employees. If actual respondents share employers, teams or repeated survey occasions, the dependence needs a suitable design and inference method. Do not assume that the ordinary case bootstrap resolves clustering.
On a small screen, swipe the table sideways to see every column.
| Column | Meaning | Used in |
|---|---|---|
| CSR | Perceived responsible conduct | Both models |
| FAIR | Overall fair treatment | Satisfaction mediator |
| ALLOC | Fairness of resource allocation | Satisfaction moderator |
| SAT | Job satisfaction | First outcome |
| BELONG | Sense of belonging | Commitment mediator |
| COMMIT | Organisational commitment | Second outcome |
| case_id | Fictional employee | Identifier only |
Step 1. Inspect concept overlap before building two models
Choose File > Import Data… and Choose data file… in a writable project. In Review & import confirm raw data, Header row 1, comma separator, Decimal convention Dot (1.25) and Continuous scores. Choose Import dataset and confirm Dataset imported and verified.
Check the 260 complete rows and score directions. No reverse coding is needed. Do not interpret a missing value as neutral fairness or low commitment. The supplied file has no missing entries, so an unexpected blank after import needs investigation.
Read the codebook's distinction between FAIR and ALLOC and between SAT and COMMIT. A shared positive direction does not make two measures interchangeable. In a real study, assess the measurement model before reducing item sets to observed scores.
After import, confirm 260 rows and all six analytical columns: CSR, FAIR, ALLOC, SAT, BELONG and COMMIT. The two questions will use different subsets of this same file.
Step 2. Build the moderated satisfaction model
The observed workflow evaluates a continuous moderator at its sample mean and mean ± one sample standard deviation. Use those printed raw-score values. Do not import the standardised −1, 0 and 1 probes from a different PLS workflow.
Open observed path analysis
Choose Calculate… and Regression, then Regression type Graph-defined Path Analysis / PROCESS.
Specify the three core paths
In Graph-defined relationships choose SAT as Outcome and CSR as Focal predictor. Under Directed paths use Add path for CSR → FAIR, FAIR → SAT and CSR → SAT.
Add allocation fairness as moderator
Use Add moderator for ALLOC, then Moderate a path for FAIR → SAT. Inspect the preview for FAIR, ALLOC and their interaction in the SAT equation. Do not also enter ALLOC as an ordinary control.
Set the bootstrap
Choose Bootstrap inference Case-resampling bootstrap, Bootstrap samples 5000, Parallel workers 1 and Seed 71017. In Configure PROCESS inference set Confidence level (%) 95 and a two-sided Test direction.
Calculate and inspect
Review the data profile and graph, retain original observed units, then Start calculation. Open the completed result. Read the actual automatic ALLOC probe values, conditional indirect effects, eligible index and usable/failed counts.
Step 3. Save the commitment question as a separate analysis
This second model does not test whether the fairness route is stronger than the belonging route. It uses a different outcome and therefore answers another question. Comparing the two coefficients as though they shared an identical outcome scale would overstate the evidence.
Keep the first result
Retain the satisfaction model and its result before preparing the second question. Give the new analysis a name that identifies commitment through belonging.
Change the graph deliberately
Open Graph-defined Path Analysis / PROCESS for the same dataset. Set Outcome to COMMIT and Focal predictor to CSR. Add CSR → BELONG, BELONG → COMMIT and CSR → COMMIT.
Remove the previous moderation from this question
Check that no ALLOC interaction remains in the relationship preview. The planned commitment model is an ordinary indirect pathway.
Calculate matching uncertainty
Use Case-resampling bootstrap with 5000 samples, 95% confidence, two-sided inference and Seed 71017. Start calculation and read the specific indirect, direct and total effects with their own intervals and fit counts.
Step 4. Check this 260-case observed-score analysis
Use the exact equations below, all 260 complete cases, raw units and intercepts. The reference uses 5,000 whole-row resamples, R seed 71017, and 95% percentile intervals with quantile type 7. It preserves the joint variation of every equation in the system.
For satisfaction, the conditional indirect estimates are 0.082, 0.123 and 0.165 at the three printed ALLOC values. The separate commitment model gives CSR → BELONG → COMMIT = 0.144, with interval 0.083 to 0.215. These are different outcomes, so the coefficients do not rank fairness against belonging.
About these answers: Reference answer calculated in R and independently repeated with NumPy.
| Check | Answer |
|---|---|
| Cases | 260 |
| Requested resamples | 5000 |
| Reference seed | 71017 |
| Independent numerical comparisons | 85054 |
What was checked
Every coefficient, indirect product, conditional effect and saved resampled fit was independently recalculated using NumPy least squares. 85,054 comparisons passed tolerance 0.00000001. This verifies the external answer, not the app’s native PROCESS or bootstrap command. The same seed in another random generator does not guarantee the same sampled rows.
Step 5. Read the satisfaction model
Fit these equations with intercepts and the original observed units: FAIR = CSR; SAT = CSR + FAIR * ALLOC. Here FAIR * ALLOC includes FAIR, ALLOC and their product. The satisfaction equation must contain both main effects as well as the interaction.
The fixed original-sample moderator probes are 2.899, 4.000, 5.102 (mean minus one sample standard deviation, mean, mean plus one sample standard deviation). They are raw scores, not standardised −1, 0 and 1.
About these answers: Reference answer calculated in R and independently repeated with NumPy.
On a small screen, swipe the table sideways to see every column.
| Equation / quantity | Estimate in raw units | 95% reference percentile interval |
|---|---|---|
| FAIR/intercept | 2.523 | 2.066 to 3.004 |
| FAIR/CSR | 0.369 | 0.256 to 0.478 |
| SAT/intercept | 3.426 | 1.582 to 5.383 |
| SAT/CSR | 0.075 | -0.039 to 0.190 |
| SAT/FAIR | -0.073 | -0.519 to 0.360 |
| SAT/ALLOC | -0.273 | -0.726 to 0.154 |
| SAT/FAIR:ALLOC | 0.102 | 0.002 to 0.206 |
| Indirect effect at low | 0.082 | 0.015 to 0.154 |
| Indirect effect at mean | 0.123 | 0.069 to 0.184 |
| Indirect effect at high | 0.165 | 0.096 to 0.243 |
| Change in indirect effect per moderator unit | 0.038 | 0.001 to 0.080 |
Equation fit and refit counts
FAIR R² = 0.132; SAT R² = 0.202
5,000 resamples requested; 5,000 usable. Every equation is refitted on the same resampled rows. No failed refit has been silently removed.
Step 6. Read the commitment model
Fit these equations with intercepts and the original observed units: BELONG = CSR; COMMIT = CSR + BELONG.
About these answers: Reference answer calculated in R and independently repeated with NumPy.
On a small screen, swipe the table sideways to see every column.
| Equation / quantity | Estimate in raw units | 95% reference percentile interval |
|---|---|---|
| BELONG/intercept | 2.612 | 2.136 to 3.138 |
| BELONG/CSR | 0.347 | 0.216 to 0.466 |
| COMMIT/intercept | 1.855 | 1.355 to 2.385 |
| COMMIT/CSR | 0.120 | 0.008 to 0.228 |
| COMMIT/BELONG | 0.416 | 0.308 to 0.529 |
| indirect/CSR → BELONG → COMMIT | 0.144 | 0.083 to 0.215 |
Equation fit and refit counts
BELONG R² = 0.119; COMMIT R² = 0.220
5,000 resamples requested; 5,000 usable. Every equation is refitted on the same resampled rows. No failed refit has been silently removed.
Step 7. Keep conditional effects and separate outcomes apart
In the satisfaction result, first read CSR → FAIR. Then read how the FAIR → SAT slope changes with ALLOC. The conditional indirect effect follows both stages at each stated allocation-fairness value.
Read the index or a directly estimated contrast to assess change in the indirect association. A conditional interval excluding zero at one probe but not another does not show that the two effects differ. The comparison itself needs uncertainty.
In the commitment result, read CSR → BELONG → COMMIT as an ordinary specific indirect effect. Its product and interval do not depend on ALLOC in this planned model. Check that you are reading the correct saved result before writing the paragraph.
Neither model establishes that responsible conduct caused the employee response. Perceptions can share unmeasured influences, and the ordering of CSR, fairness and attitudes needs support from the research design.
On a small screen, swipe the table sideways to see every column.
| Result | Question answered | Do not infer |
|---|---|---|
| Conditional fairness route | Does the CSR-to-SAT indirect association vary with ALLOC? | A difference from separate significance labels |
| Index or contrast | How does that indirect association change? | A generic improvement in all employee outcomes |
| Belonging route | Is CSR indirectly associated with COMMIT through BELONG? | That belonging is stronger than fairness for the same outcome |
| Direct effects | What relationship remains under each specified model? | A causal effect free of omitted influences |
| Fit counts and intervals | How uncertain are the estimates? | That recent publication means recent data collection |
Two positive products need not support one common conclusion
The two analyses can support different substantive discussions. Fairness concerns the experience of treatment and satisfaction; belonging concerns identification with the organisation and commitment. Do not erase that distinction because both products are positive.
Hypothetical example
Illustrative reading example
Suppose illustrative CSR → FAIR is 0.50 and FAIR → SAT is 0.20 + 0.04w in raw ALLOC units. The indirect function is 0.10 + 0.02w. At hypothetical probes 3, 4 and 5, it gives 0.16, 0.18 and 0.20.
Suppose the index interval is [−0.01, 0.05]. The estimated association increases with allocation fairness, but the change is uncertain. Report that uncertainty even if an individual conditional effect has an interval above zero.
For the separate commitment model, suppose CSR → BELONG is 0.40 and BELONG → COMMIT is 0.45. The indirect product is 0.18. Its numerical equality to the satisfaction indirect value at ALLOC 4 is coincidental. The outcomes and equations differ.
The conditional products 0.16, 0.18 and 0.20 and the separate belonging product 0.18 are invented. They are not expected CSV results, copied paper coefficients or a statistical test comparing the two pathways.
Investigate overlapping fairness measures honestly
Inspect the relationship between FAIR and ALLOC and their conceptual definitions. If a real questionnaire makes them nearly redundant, a large sample or more bootstrap draws will not resolve the underlying measurement ambiguity. Review the items before interpreting an unstable interaction.
A simpler fairness model without moderation can be a clearly labelled exploratory comparison. It does not become the planned hypothesis merely because its indirect interval is narrower. Keep the moderated model and explain why the comparison was added.
Preserve the two outcome questions when reporting. If you later want to compare mediators for the same outcome, specify that new model and the relevant effect contrast directly. The two separate exercises do not supply that comparison automatically.
Check which model and which concept the result belongs to
On a small screen, swipe the table sideways to see every column.
| Symptom | Check | Action |
|---|---|---|
| SAT and COMMIT are described as the same outcome | Their definitions and saved graph | Write separate interpretations. |
| ALLOC remains in the commitment model accidentally | Relationship preview and interaction list | Remove the unintended term before calculating the planned model. |
| The moderation coefficient is unstable | FAIR/ALLOC overlap and observed range | Review measurement and report uncertainty rather than deleting cases. |
| The two indirect estimates are ranked directly | Outcome scale and model differences | Avoid claiming a tested difference from separate analyses. |
| The conditional probes are mislabelled | Printed raw mean and standard deviation | Use the actual values and metric. |
| Publication year is used as survey year | Source methods and data-collection date | Report the actual observation period. |
Step 8. Write one paragraph for each research question
For satisfaction, insert the actual probes, conditional indirect effects and index or contrast interval. For commitment, insert the specific indirect, direct and total estimates. State that the analyses address distinct outcomes and do not test a ranking of mediators.
| Model | Report |
|---|---|
| Satisfaction | Graph, ALLOC probes, interaction and conditional indirect evidence |
| Commitment | Graph and ordinary indirect evidence |
| Both | Observed-score definitions, case unit and uncertainty settings |
| Comparison | Conceptual differences, without unsupported coefficient ranking |
Methods writing outline
We fitted two observed-score teaching models. The satisfaction model specified CSR → FAIR → SAT with ALLOC moderating the second stage and required main effects retained. The commitment model specified CSR → BELONG → COMMIT without moderation. Both retained a direct CSR-to-outcome path and used [bootstrap settings, interval method and counts].
Step 9. Explain the finding in this file
Write the outcome and the precise path being discussed. Include the estimate, its matching interval, the metric and the actual probe value when relevant. In a real paper, report the procedure you ran and its usable refits; do not copy reference intervals as if they came from a different calculation.
Keep the employee questions distinct
- Explain FAIR versus ALLOC.
- Explain SAT versus COMMIT.
- Verify the moderated arrow in the first model.
- Verify the absence of that interaction in the second.
- Read the conditional-change interval directly.
- Keep publication date separate from data-collection date.