QuickPLS Perceptions of risk are not measurements of technical safety
Adigun studied 412 sign-language interpreters from 12 African countries, examining perceived benefits, risks, trust, cultural influence and future views of AI-enabled interpretation. The analysis included factor assessment and PLS with bootstrap inference.
In QuickPLS you will fit Benefit, Risk and Trust as predictors of Future acceptance, check their measurement and overlap, then bootstrap the Risk relationship. The original teaching model helps you investigate a positive risk coefficient without reversing its meaning. The article’s item-removal accounts differ, so this exercise does not reconstruct its final scale; perceptions also cannot establish translation accuracy or outcomes for Deaf learners.
Your study: benefits, risk, trust and future acceptance
Imagine 320 fictional interpreters rating perceived benefit, perceived risk, trust and future acceptance of an educational interpretation tool. Each concept is represented by three reflective teaching indicators. The complete continuous scores range from 1 to 7.
Draw Benefit → Future acceptance, Risk → Future acceptance and Trust → Future acceptance. Higher Risk means more perceived risk. The question is how each perception is associated with future acceptance after the other two are included. There is no cultural mediator or objective performance outcome in this smaller model.
The example is deliberately suitable for discussing a potentially positive risk association, but no particular sign, probability or interval is promised. Read the result you obtain. The data-generation design is not a result about interpreters, sign languages or AI safety.
On a small screen, swipe the table sideways to see every column.
| Construct | Columns | Higher values mean |
|---|---|---|
| Benefit | benefit1, benefit2, benefit3 | Greater perceived benefit |
| Risk | risk1, risk2, risk3 | Greater perceived risk |
| Trust | trust1, trust2, trust3 | Greater perceived trust |
| Future acceptance | future1, future2, future3 | More favourable future acceptance |
| Identifier | case_id | One fictional interpreter |
Step 1. Keep the participant perspective visible
In a real study, specify whose perspective the survey measures. Professional interpreters, Deaf students, teachers and software evaluators may answer different questions. An interpreter’s optimism is not interchangeable with a learner’s comprehension or an independent assessment of translation errors.
Create a writable project and import through File → Import Data… → Choose data file…. In Review & import confirm raw data, Header row 1, comma separation and Decimal convention Dot (1.25). Keep the twelve measurement columns Continuous, choose Import dataset and confirm 320 cases.
The practice columns are complete and already point in the declared direction. Do not reverse Risk because you expect risk to reduce acceptance. Reverse coding follows the item’s meaning and response scale; it is not a device for correcting a surprising structural result. Keep case_id outside the model.
Step 2. Build the three-predictor acceptance model
Create a model
Choose File → New Model…, enter Interpreter perceptions, check Dataset version and Cases, and choose Create model.
Create the reflective blocks
Ctrl-select benefit1, benefit2 and benefit3 in Indicators and choose Create construct. Name the block Benefit in Properties → Model. Repeat for Risk, Trust and Future acceptance using their matching columns.
Check measurement roles
For every block set Properties → Parameter → Measurement model to Reflective. Check each assignment and leave case_id unassigned. No block is a measured translation-accuracy construct.
Draw the three paths
Use Connect to draw Benefit, Risk and Trust into Future acceptance. Return to Select and save. Do not add reverse paths after seeing an unexpected sign.
What you should see: Benefit, Risk and Trust each predict Future acceptance; no reverse path or translation-accuracy construct has been added.
Calculate
Choose Calculate… → PLS-SEM Algorithm with Starting weights → Standard (+1) and Missing data → Treatment → Casewise deletion. Review Readiness and choose Start calculation. Confirm the intended model in the Result dropdown.
Step 3. Check whether the risk indicators mean what you think they mean
Open Estimates → Outer loadings. Read the Risk rows first, checking their sign and indicator assignments. If all three are positively oriented, the construct’s score remains a greater-risk measure. A positive path from that score does not redefine it as lower risk.
Inspect Reliability and validity → Construct reliability and validity and Discriminant validity → HTMT+. Risk and Trust are not automatically opposite ends of one scale. A person may recognise risks while trusting a tool for limited uses. Check whether the actual indicators distinguish these ideas rather than assuming they must correlate negatively.
Use Collinearity and effects → Inner VIF values for the Future acceptance equation. Benefit, Risk and Trust may share information. Their adjusted paths describe each predictor with the others included; that is different from a simple one-variable association.
Select the Interpreter perceptions result and locate the Risk indicators in Outer loadings. Confirm their assignments and direction before moving to Risk → Future acceptance in Direct effects. Keep Benefit and Trust in the same fitted equation.
Check the fitted answers for these 320 cases
Use all 320 cases in sign-language-risk-perceptions.csv and the exact blocks and arrows in this lesson. The check uses ordinary PLS, path weighting, standardised indicators, starting weights +1, a 3,000-iteration limit and stopping criterion 0.0000001. Perceived benefit, Perceived risk, Perceived trust and Future acceptance are reflective blocks and use Mode A.
Risk → future acceptance is positive at 0.115, alongside benefit (0.239) and trust (0.240). The risk score measures perceived concern; it does not measure the technical safety of a sign-language system.
About these answers: Calculated from QuickPLS calculation code and independently checked with the cSEM package in R.
On a small screen, swipe the table sideways to see every column.
| Specified path | Point estimate | Reference 95% percentile interval |
|---|---|---|
| Perceived benefit → Future acceptance | 0.239 | 0.143 to 0.334 |
| Perceived risk → Future acceptance | 0.115 | 0.019 to 0.217 |
| Perceived trust → Future acceptance | 0.240 | 0.144 to 0.342 |
How these answers were checked
The checks compared 28 estimates from the original data and 3,000 direct-path estimates from 1,000 repeated fits. Every difference was within the absolute tolerance of 0.00000001. QuickPLS v1.03.11 calculation code and cSEM 0.7.1 used the same model, file and sampled rows. Both calculations returned usable results for all 1,000 fits.
The reference used R seed 202611086 to sample rows and calculated percentile intervals with R quantile type 7. It repeatedly fitted the model through calculation code; it did not use the desktop app’s Bootstrapping command. The same seed can produce different sampled rows in different programs. The download includes the actual row numbers, unrounded answers and settings so this calculation can be repeated.
Check the measurement values and explained variation
Check the measurement values and explained variation
All blocks in this exercise are reflective. Loading relates an indicator to its construct score; Weight shows how that indicator contributes to forming the score. The presence of weights does not make these blocks formative. A high loading alone does not establish validity or distinct concepts, so also assess reliability, discriminant validity and item content.
About these answers: Estimates from the original data, calculated with QuickPLS code and cSEM using the same model and file.
On a small screen, swipe the table sideways to see every column.
| Block / indicator | Loading | Weight |
|---|---|---|
| Perceived benefit / benefit1 | 0.941 | 0.326 |
| Perceived benefit / benefit2 | 0.950 | 0.383 |
| Perceived benefit / benefit3 | 0.957 | 0.344 |
| Perceived risk / risk1 | 0.942 | 0.325 |
| Perceived risk / risk2 | 0.947 | 0.360 |
| Perceived risk / risk3 | 0.955 | 0.370 |
| Perceived trust / trust1 | 0.946 | 0.364 |
| Perceived trust / trust2 | 0.939 | 0.356 |
| Perceived trust / trust3 | 0.940 | 0.342 |
| Future acceptance / future1 | 0.947 | 0.329 |
| Future acceptance / future2 | 0.954 | 0.382 |
| Future acceptance / future3 | 0.945 | 0.343 |
Explained variation in each outcome
R² describes this fitted sample. It is not a test of prediction on new observations.
| Outcome | R² |
|---|---|
| Future acceptance | 0.169 |
Step 4. Read an unexpected sign before trying to explain it
Open Effects → Direct effects and locate Risk → Future acceptance. If it is positive, a careful first sentence is: higher perceived risk was associated with higher future acceptance after accounting for the included Benefit and Trust scores. That describes the model without inventing a mechanism.
Next check whether the interval supports a stable direction and whether the contribution is practically substantial. A positive point estimate can be small and uncertain. Even a statistically clear association would not prove that increasing risk improves acceptance or that users should tolerate unsafe interpretation.
Possible explanations in a real study include exposure, experience, shared response tendencies, omitted influences or measurement problems. For example, experienced users may recognise both benefits and risks more readily. Those are hypotheses for further investigation, not conclusions supplied by the coefficient. Do not choose one merely because it makes a counterintuitive result sound appealing.
Step 5. Separate a positive coefficient from a large contribution
Read Estimates → R-squared for Future acceptance, then Collinearity and effects → f-square effect sizes for the Risk contribution. R-square summarises fitted outcome variation accounted for by the full model. The f-square measure concerns how much that explanation changes when a predictor is removed from the relevant equation.
Use the actual effect-size output for your run. A statistically supported path can still make a small additional contribution when the other perceptions explain much of the same outcome variation. Explain magnitude and uncertainty together instead of equating a small p-value with practical importance.
Neither R-square nor f-square is an accuracy percentage for sign-language translation. They describe the fitted perception model. No case in the teaching file contains a translated utterance, comprehension score or documented safety incident.
Hypothetical example
Illustrative reading example
Suppose Future acceptance R-square is 0.24 with Risk included and 0.23 without it. The f-square arithmetic is (0.24 - 0.23) / (1 - 0.24) = 0.0132 approximately. The full model accounts for 24% of fitted score variation, while Risk adds relatively little under this comparison. These are hypothetical model values, not technical accuracy measurements or fitted answers.
Step 6. Assess the sign with a matching bootstrap interval
Choose Calculate… → PLS-SEM Bootstrapping for the same model. For this teaching run enter 1000 Bootstrap samples, Confidence level (%) 95 and Seed 202611086. Start the calculation, then record the interval method and attempted, usable and failed counts.
Keep the original Risk coefficient with its own interval. If the range crosses zero, describe uncertainty about direction. If it lies entirely above zero, describe a positive adjusted association under the model’s assumptions. In either case, keep the conclusion about perceptions rather than objective system performance.
If bootstrap fits fail, inspect the messages for constant indicators, excessive overlap or convergence problems. A completed table with some intervals does not mean every requested refit succeeded. More draws may reduce random simulation variation but do not resolve a poorly defined measure.
Hypothetical example
Illustrative reading example
A Risk estimate of 0.12 with a 95% interval of [-0.02, 0.26] leaves small negative and positive associations plausible. A suitable statement is that the estimated adjusted risk relationship was positive but imprecise. These invented bounds are not a result from the download.
Check coding without forcing a preferred direction
On a small screen, swipe the table sideways to see every column.
| Observation | Check | Reasonable action |
|---|---|---|
| The Risk path is positive | Score meaning, indicator assignment and included predictors | Describe the adjusted association accurately before proposing explanations. |
| A reviewer assumes high Risk means low risk | The item direction key and construct label | Clarify coding in the table and text; do not silently rename the score. |
| Risk and Trust seem strongly related | Definitions, HTMT and predictor overlap | Investigate whether respondents can distinguish the concepts. |
| A small p-value accompanies tiny contribution | f-square, interval and substantive units | Discuss practical magnitude separately from statistical evidence. |
| The interval crosses zero | Its width and plausible signs | Report imprecision rather than recoding or deleting cases. |
| The report claims safe interpretation for Deaf learners | Actual participants and measured outcomes | Limit the claim to perceptions and specify what performance evidence would be needed. |
Use a simpler equation as an explanatory comparison
A useful sensitivity exercise can compare the planned three-predictor model with a separately saved simpler model. State exactly which adjustment was removed and why. If the Risk coefficient changes, explain that conditional and less-adjusted associations answer different questions; do not choose the preferred sign as the final truth.
Keep the original model and full measurement map. If an actual coding mistake is found, correct it from documented item meaning and rerun all affected results. A theoretically surprising but correctly calculated relationship needs interpretation and further study, not data editing.
For a real application, perceived acceptance should sit alongside appropriate technical and user-centred evaluation. This statistical exercise cannot stand in for that evidence. A case-based understanding of language, context and affected users is essential before turning acceptance findings into deployment advice.
Keep the conclusion inside the measured question
Methods outline: We modelled Future acceptance from Benefit, Risk and Trust, each represented by three reflective indicators. Higher Risk scores indicated greater perceived risk. We assessed measurement distinction and predictor overlap before interpreting paths and f-square contributions. We used [bootstrap procedure and settings], with [usable and failed counts].
Results outline: Risk’s adjusted association with Future acceptance was [estimate and interval], with f-square [value]. [Explain whether the direction was clear or uncertain.] Benefit and Trust results were [estimates and intervals]. The outcome concerns interpreters’ reported future acceptance, not translation accuracy, learner comprehension or safety.
| Report | Include |
|---|---|
| Perspective | Who answered and whose outcomes were not measured. |
| Coding | Higher Risk means more perceived risk. |
| Adjusted result | Included predictors, path estimate and interval. |
| Magnitude | Outcome-labelled R-square and predictor contribution. |
| Sensitivity | Any simpler equation, preserved alongside the planned model. |
| Scope | Perception evidence separated from technical and learner-outcome evidence. |
Put this result into a report
We fitted the specified ordinary PLS model to 320 original synthetic teaching cases. The measurement roles and complete structural model were fixed before interpreting the results. Point estimates were checked against an independent cSEM calculation.
Report the relevant path estimate and its matching interval from the table. If you use these supplied intervals, identify the separate repeated-fit reference calculation. For results from your own app analysis, report that run’s settings and intervals. A real study also needs the actual sampling design, measurement evidence and a reason for each model choice.
Explain the surprise without changing the data
- State what a positive Risk coefficient means with the other predictors included.
- Explain why Risk should not be reversed to obtain a negative path.
- Recalculate the hypothetical f-square of approximately 0.0132.
- Distinguish an uncertain positive point estimate from a supported positive direction.
- Write a result sentence that makes no claim about translation accuracy or Deaf learners’ outcomes.