QuickPLS Assigned information and perceived source are different variables
Nagpal and Cotte examined responses to human and AI involvement in news. Their studies included experimental conditions and mediation analyses involving perceived source, human oversight and credibility. The article’s analysis-specific sample counts vary, so a single sample number should not be applied to every reported table.
In QuickPLS you will enter the observed-variable graph, fit its three regression equations and bootstrap the complete serial route. The new teaching file starts with a continuous perception of human involvement, not randomized assignment. It practises the paper’s mediation approach without reproducing its experimental treatment comparison.
Your study: perceived involvement, oversight, credibility and reliance
Imagine 300 fictional readers evaluating an article. The question is whether perceived human involvement is associated with perceived oversight, whether oversight is associated with credibility, and whether credibility is associated with willingness to rely on the article.
The four columns are directly observed continuous teaching scores from 1 to 7. There are no multi-item latent constructs in this exercise. The graph will connect ordinary regression equations, so measurement error is not separately estimated or corrected.
Specify source → oversight, source → credible, source → rely, oversight → credible, oversight → rely and credible → rely. This keeps the long serial route and the shorter indirect routes visible. All columns are complete; case_id identifies a reader and does not enter any equation.
On a small screen, swipe the table sideways to see every column.
| Column | Meaning | Role |
|---|---|---|
| source | Perceived human involvement | Continuous focal predictor, not treatment assignment |
| oversight | Perceived human oversight | First mediator |
| credible | Perceived credibility | Second mediator |
| rely | Reliance intention | Continuous outcome |
| case_id | Fictional reader identity | Identifier only |
Step 1. Check the scale and the timing you are claiming
Import the CSV through File → Import Data… → Choose data file… in a writable project. In Review & import confirm raw data, Header row 1, comma separation and Decimal convention Dot (1.25). Keep the four scores Continuous, choose Import dataset and verify 300 cases.
No binary recoding is needed. In particular, do not treat source as 0 for AI and 1 for human: it is a continuous perception in this file. A genuine assigned experimental condition would require its own documented coding and design interpretation.
The teaching records do not establish when perceptions arose. A reader might judge an article credible and then infer greater human oversight. Draw the proposed sequence for statistical analysis, but keep alternative orderings and common influences in the interpretation. The software’s graph does not resolve them.
Step 2. Enter the equations in the observed-variable route
Before calculating, check Relationship preview for all six arrows. Outcome must be rely and Focal predictor must be source; oversight and credible are the ordered mediators. Keep Regression type on Graph-defined Path Analysis / PROCESS rather than the separate ordinary least-squares form.
Open Regression
Select the imported numeric dataset and choose Calculate → Regression under Standalone analysis. Confirm the intended data version.
Choose the path-analysis route
Set Regression type to Graph-defined Path Analysis / PROCESS. Set Outcome to rely and Focal predictor to source. You will enter the relationships directly in the graph instead of selecting a numbered PROCESS template.
What you should see: The setup shows the intended outcome and focal predictor, ready for the six relationships.
Add the serial chain
Under Graph-defined relationships use Add path, with From and To, to enter source → oversight, oversight → credible and credible → rely.
Add the shorter and direct routes
Use Add path for source → credible, source → rely and oversight → rely. Keep Controls empty for this exercise. Review Relationship preview and confirm all six arrows with no reverse loop.
Keep the original units
Review the common complete-case profile. Under Configure groups, data metrics, scores and copulas retain All cases and set Final variable metric to Raw. Retain the equation intercepts and do not add equal-weight scores or copula controls for this first analysis.
Step 3. Write the three equations before reading the output
The graph defines three equations. Oversight is predicted from source. Credibility is predicted from source and oversight. Reliance is predicted from source, oversight and credibility. Each equation includes its intercept, and the complete system uses the common cases required by these variables.
On the Raw metric, a coefficient describes an outcome-unit difference associated with a one-unit predictor difference, accounting for the other predictors in that equation. The source → credible coefficient is therefore adjusted for oversight. It is not the simple correlation between source and credibility.
Read the calculation status and analysed count first. Inspect the equation coefficients, uncertainty and available residual or overlap diagnostics. A large coefficient from an unstable equation is not automatically strong evidence. These observed scores are not validated psychological scales merely because the equations can be fitted.
On a small screen, swipe the table sideways to see every column.
| Equation outcome | Predictors | Question |
|---|---|---|
| oversight | source | How is perceived involvement related to oversight? |
| credible | source, oversight | How is oversight related to credibility after source is included? |
| rely | source, oversight, credible | How is credibility related to reliance after the earlier perceptions are included? |
Check this 300-case observed-score analysis
Use the exact equations below, all 300 complete cases, raw units and intercepts. The reference uses 1,000 whole-row resamples, R seed 202611084, and 95% percentile intervals with quantile type 7. It preserves the joint variation of every equation in the system.
The long source → oversight → credible → rely route is 0.077, with interval 0.046 to 0.115. The shorter source → credible → rely route is 0.018, with interval −0.022 to 0.061. Source here is a continuous perception score, not the binary assignment in the separate message-attribution lesson.
About these answers: Reference answer calculated in R and independently repeated with NumPy.
| Check | Answer |
|---|---|
| Cases | 300 |
| Requested resamples | 1000 |
| Reference seed | 202611084 |
| Independent numerical comparisons | 12036 |
What was checked
NumPy independently recalculated every regression coefficient, indirect effect, conditional effect and saved resampled fit using least squares. All 12,036 comparisons were within 0.00000001. This checks the answers calculated in R; it does not test the desktop app’s PROCESS or Bootstrapping command. To reproduce the reference exactly, use its saved sampled rows: the same seed can select different rows in different programs.
Read the serial model
Fit these equations with intercepts and the original observed units: oversight = source; credible = source + oversight; rely = source + oversight + credible.
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 |
|---|---|---|
| oversight/intercept | 2.286 | 1.908 to 2.678 |
| oversight/source | 0.429 | 0.337 to 0.523 |
| credible/intercept | 1.920 | 1.459 to 2.410 |
| credible/source | 0.048 | -0.057 to 0.154 |
| credible/oversight | 0.473 | 0.361 to 0.585 |
| rely/intercept | 1.480 | 0.966 to 2.042 |
| rely/source | 0.092 | -0.023 to 0.198 |
| rely/oversight | 0.160 | 0.038 to 0.274 |
| rely/credible | 0.379 | 0.275 to 0.477 |
| indirect/source → oversight → rely | 0.069 | 0.017 to 0.121 |
| indirect/source → credible → rely | 0.018 | -0.022 to 0.061 |
| indirect/source → oversight → credible → rely | 0.077 | 0.046 to 0.115 |
Equation fit and refit counts
oversight R² = 0.180; credible R² = 0.252; rely R² = 0.265
1,000 resamples requested; 1,000 usable. Every equation is refitted on the same resampled rows. No failed refit has been silently removed.
Step 4. Follow the entire route when calculating an indirect association
Read the specific source → oversight → credible → rely effect. In this unmoderated linear system it multiplies the three component coefficients from the same fitted model. Also read source → oversight → rely and source → credible → rely before their sum.
The direct source → rely coefficient is the source coefficient in the final equation after both mediators are included. Total indirect adds the specific routes; total effect combines direct and indirect parts under this model. Keep these definitions with the output rather than using mediation as one label for everything.
If hand multiplication disagrees, check full-precision values, identical cases, intercepts, variable metric and the named route. Rounding can create a small discrepancy. Combining coefficients from separately filtered equations can create a different analysis altogether.
Hypothetical example
Illustrative reading example
Suppose source → oversight is 0.50, oversight → credible is 0.40 and credible → rely is 0.60. The serial product is 0.12. If the two shorter indirect routes are 0.05 and 0.06, total indirect is 0.23. With a direct coefficient of 0.07, the total is 0.30. These invented values teach the arithmetic; they are not fitted answers for the file.
Step 5. Resample readers and refit the whole system
A complete-system bootstrap keeps the joint variation of the coefficients that make the product. A bootstrap of only the final reliance equation would omit uncertainty from the earlier equations. Multiplying three separate interval endpoints would not repair that omission.
For a hypothetical serial estimate of 0.12, an interval of [0.04, 0.23] would support a positive indirect association under the model. It would not establish that changing the assigned source of a news article causes the same response. The starting variable in this exercise is a perception, and the quoted interval is only a reading illustration.
Enable complete-system resampling
Set Bootstrap inference to Case-resampling bootstrap. Enter Bootstrap samples 1000 and Seed 202611084, and record Parallel workers. The outer samples must refit all equations in the graph.
Set the intended inference
Open Configure PROCESS inference. Choose Confidence level (%) 95 and Test direction Two-sided before reading the effects. Leave nested studentized intervals off for this first exercise.
Read each effect and its calculation status
Start the calculation. In the completed result, use Search result outputs to open Equation coefficients and check the three equations. Open Bootstrap inference and find the complete source → oversight → credible → rely effect. Read the named percentile or eligible BCa interval actually provided, then check Bootstrap summary and Bootstrap failures for usable and failed fits. Retain any unavailable-interval explanation.
What you should see: The interval belongs to the entire serial route, and the output records which repeated fits were usable.
Do not let an experimental setting change the meaning of the predictor
In a real experiment, an assigned condition can support a treatment comparison when randomization and the design are sound. A perception measured after exposure can be influenced by the treatment, the participant and the article. Replacing assignment with that perception changes the estimand, meaning the effect or association being estimated.
Adding a post-exposure perception as a control can also change the question. It may remove part of the response to assignment or introduce other interpretation problems. Write the intended treatment, mediator and outcome roles before deciding which variables enter each equation.
This lesson makes no treatment-effect claim. It teaches associations among observed perceptions and reliance intention. Actual sharing behaviour, reading time and later reliance would be separate outcomes with their own measurement and modelling needs. A continuous intention score should not be reported as a binary behavioural event.
Find whether the disagreement comes from coding or from the question
On a small screen, swipe the table sideways to see every column.
| Observation | Check | Action |
|---|---|---|
| source is treated as a binary condition | Its continuous definition and imported values | Restore the perception interpretation; use a separate genuine assignment variable for an experimental question. |
| The serial product differs from hand arithmetic | Common cases, raw metric and the complete path | Use coefficients from the same equation system. |
| A direct source coefficient is small | Short and serial indirect components | Report the decomposition without requiring a significant direct effect first. |
| An interval is wide or unavailable | Usable refits, rank deficiency and equation stability | Distinguish uncertainty from a failed calculation and report both accurately. |
| A single-equation bootstrap was used | Whether earlier equations were also refitted | Use the complete PROCESS graph for indirect uncertainty. |
| The conclusion says assigned AI caused distrust | Whether assignment or perception was the predictor | Rewrite the conclusion to match the actual variable and design. |
Make any added control answer a defensible question
A useful real-study sensitivity analysis might add a measured pre-exposure characteristic that plausibly relates to both perceptions and reliance. State why it belongs in the equations before looking at probabilities. In this editor, a selected Control enters every equation, so review whether that matches the intended adjustment.
The practice file contains no such extra characteristic, so do not use case_id as a control or invent a variable. Keep the planned four-variable system. If a later study explores an alternative order or adjustment set, retain the primary result and label the alternative as a different analysis.
Changing the final metric from Raw to Standardized changes coefficient units. It can be a useful presentation choice, but do not compare its numeric coefficients with the raw arithmetic as though they had the same units. Record the metric with every result table.
Name the measured predictor and the complete system
Methods outline: We analysed four observed continuous scores using Graph-defined Path Analysis / PROCESS. Perceived human involvement was the focal predictor, oversight and credibility were ordered mediators, and reliance intention was the outcome. We retained the six stated paths and equation intercepts on the Raw metric, using common complete cases. We assessed indirect uncertainty with [complete-system bootstrap settings and counts].
Results outline: The source → oversight → credible → rely association was [estimate], with a [level and method] interval of [bounds]. The shorter indirect associations were [named values], and the direct association was [value and interval]. These results concern measured perceptions and reliance intention; they are not estimates of a randomized source-label treatment effect.
| Report | Include |
|---|---|
| Variables | Perception versus assignment, coding and timing. |
| Equations | All predictors, controls if any, intercepts and common cases. |
| Scale | Raw or transformed metric, stated consistently. |
| Effects | Direct, each specific indirect and total definitions. |
| Inference | Complete-system resampling, interval names and failures. |
| Design scope | Which causal claims are supported independently of the graph. |
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 causal and numerical questions separate
- Write all three regression equations from the graph.
- Explain why source is not an assigned 0/1 treatment in this file.
- Calculate the serial product and total association from the saved full-precision reference coefficients. Use the separate hypothetical example as an additional arithmetic check.
- Identify the interval for the complete serial effect.
- Explain why an experiment elsewhere in a paper does not make every perception-based regression a treatment-effect estimate.