QuickPLS Using an assistant can mean several different activities
Yoon and Lee studied voice-assistant experiences among 515 US users. Their model distinguished system features, user experiences and use, and treated use as formative. The paper reports a PLS analysis with bootstrap inference. Its description of the use-item count is not fully consistent across passages, so this lesson does not reconstruct that exact block.
In QuickPLS you will create reflective Responsiveness and Connection blocks and a formative Use block, then fit their paths and inspect the indirect relationship through Connection. The original teaching data practise the paper’s measurement distinction without reconstructing its inconsistent use-item count. Different activities define Use; related feelings reflect the experience constructs.
Your study: responsive assistants and a broader range of use
Imagine 300 fictional voice-assistant users. You want to know whether a responsive assistant is associated with feeling connected to it, and whether that connection is associated with a broader pattern of use. Responsiveness and Connection are reflective constructs with three indicators each.
Use is formed from three activities: planning, finding information and routine tasks. A person can use the assistant often for one activity and rarely for another. For this exercise these are different ingredients of use, not interchangeable expressions of one attitude. Removing an activity changes what the overall score includes.
Draw Responsiveness → Connection, Connection → Use and Responsiveness → Use. The direct arrow lets you examine whether responsiveness has a remaining association with the formed Use score after accounting for Connection. The file contains complete continuous scores from 1 to 7, not actual usage logs or the paper’s respondents.
On a small screen, swipe the table sideways to see every column.
| Construct | Columns | Interpretation |
|---|---|---|
| Responsiveness | response1, response2, response3 | Reflective experience measure |
| Connection | connection1, connection2, connection3 | Reflective experience measure |
| Use | planning, information, routine | Three formative ingredients |
| Case label | case_id | One fictional user; not an indicator |
Step 1. Make a measurement argument before opening the results
Write two sentences before analysis. First: the responsiveness questions are intended to reflect one experience, so similar answers are expected. Second: the three activities define the breadth of use, so one activity cannot automatically replace another. These explanations are the reason for the two measurement roles.
Import the CSV in a writable project through File → Import Data… → Choose data file…. In Review & import choose raw data, Header row 1, comma separator and Decimal convention Dot (1.25). Keep numeric teaching columns Continuous. Choose Import dataset and confirm Dataset imported and verified with 300 cases.
Check nine measurement columns plus case_id. All values are complete and already point in the intended direction. No reversal is required. In a real usage study, define the observation period and whether each activity is a count, duration or rating. Mixing minutes and frequency counts without a stated construction changes the meaning of the composite.
Step 2. Build the reflective blocks and the formative Use block
Create the model
Choose File → New Model…, enter Voice-assistant activities, check Dataset version and Cases, then choose Create model.
Create Responsiveness and Connection
Ctrl-select response1, response2 and response3 in Indicators and choose Create construct. Select it and set Properties → Model → Name to Responsiveness. Repeat for the three connection columns. Under Parameter → Measurement model choose Reflective for both blocks.
Create the activity composite
Select planning, information and routine, then choose Create construct. Name it Use. In Properties → Parameter → Measurement model choose Formative. Check that the three activity columns belong only to this block.
What you should see: Use is labelled Formative and contains planning, information and routine; the experience blocks remain Reflective.
Draw the proposed relationships
Choose Connect and drag Responsiveness to Connection, Connection to Use and Responsiveness to Use. Return to Select and save. Keep case_id outside the model.
Calculate the point estimates
Choose Calculate… → PLS-SEM Algorithm. Use Starting weights → Standard (+1) and Missing data → Treatment → Casewise deletion. Review Readiness and start calculation. Confirm that the result belongs to this saved mixed-measurement model.
Step 3. Check the experiences with reflective criteria
In Results, select the completed calculation. Open Estimates → Outer loadings and inspect the three indicators for Responsiveness and Connection. A loading relates an indicator to its construct. Check its sign, its size and whether the item definition fits the block.
Open Reliability and validity → Construct reliability and validity for the reflective constructs. Reliability concerns consistency among their indicators; average variance extracted concerns the indicator variation accounted for. Then inspect Discriminant validity → HTMT+ for eligible reflective pairs. Similar feelings about responsiveness and connection can overlap, so content distinctions still matter.
These checks do not automatically apply to the formative Use block. Low agreement among planning, information and routine use may reflect real differences in activity rather than a defective measure. Treating their alpha as an item-deletion target would contradict the definition you wrote before analysis.
In this result, apply reflective loadings, reliability and HTMT+ to Responsiveness and Connection. Then switch to the Use activity weights, loadings and collinearity checks described next. The measurement role determines which evidence you need.
Step 4. Read both weights and loadings for Use
Find Outer weights using Search result outputs. A weight describes an activity’s contribution to forming the Use score under the selected model. Read it with its uncertainty and the collinearity information available for that block. Overlapping activity measures can make their separate contributions unstable.
Also inspect Outer loadings for those activities. A loading describes the activity’s relationship with the formed score, which differs from its unique contribution expressed by the weight. A small weight and a substantial loading can coexist. Neither result by itself decides whether the activity belongs in the construct.
A useful content question is: if routine use were removed, would the score still represent the intended range of voice-assistant use? If not, a small or uncertain weight needs discussion rather than automatic deletion. You can investigate whether the activity was measured badly, but that is a different claim from saying it is unimportant.
Do not interpret a negative formative weight as proof that performing the activity reduces overall use in the real world. First examine overlapping ingredients, coding and the score definition. Formative weights are model-dependent contributions, not causal effects of activities on behaviour.
Hypothetical example
Illustrative reading example
Suppose planning has weight 0.09 with a 95% interval of [-0.07, 0.25], while its loading is 0.68. The unique weight is uncertain, yet planning remains related to the score. Keeping it may preserve necessary content. These invented numbers illustrate a decision; they are not expected values from the download.
Check the fitted answers for these 300 cases
Use all 300 cases in voice-assistant-formative-use.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. Responsiveness and Connection are reflective blocks and use Mode A. Use is formative: planning, information and routine form the composite, estimated with Mode B.
Response → connection is 0.523. Connection → USE is 0.477, while the direct response → USE estimate is 0.064. USE is the formative composite of planning, information and routine use.
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 |
|---|---|---|
| Responsiveness → Connection | 0.523 | 0.436 to 0.605 |
| Connection → Use | 0.477 | 0.369 to 0.580 |
| Responsiveness → Use | 0.064 | -0.055 to 0.180 |
How these answers were checked
The checks compared 23 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 202611075 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
Use the block type when interpreting these values. A reflective loading relates an indicator to its construct score; a formative weight is its conditional contribution to the composite. A high loading alone does not establish validity or distinct concepts. These tables do not substitute for reliability, discriminant-validity and content checks.
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 |
|---|---|---|
| Responsiveness / response1 | 0.958 | 0.347 |
| Responsiveness / response2 | 0.956 | 0.348 |
| Responsiveness / response3 | 0.948 | 0.353 |
| Connection / connection1 | 0.954 | 0.360 |
| Connection / connection2 | 0.955 | 0.356 |
| Connection / connection3 | 0.948 | 0.334 |
| Use / planning | 0.751 | 0.664 |
| Use / information | 0.688 | 0.560 |
| Use / routine | 0.403 | 0.288 |
Explained variation in each outcome
R² describes this fitted sample. It is not a test of prediction on new observations.
| Outcome | R² |
|---|---|
| Connection | 0.273 |
| Use | 0.264 |
Read each indirect route separately
Each row follows one specific route. Multiply its component coefficients within each repeated fit before taking the interval; multiplying the endpoints of component intervals is incorrect. These are statistical associations in synthetic data, not evidence of causal mediation.
About these answers: Each indirect effect was recalculated in every saved reference resample. These intervals were not produced by the desktop app’s Bootstrapping command.
On a small screen, swipe the table sideways to see every column.
| Specific route | Point product | Reference 95% percentile interval |
|---|---|---|
| Responsiveness → Connection → Use | 0.249 | 0.179 to 0.329 |
Step 5. Read the relationships to the composite you actually defined
Open Effects → Direct effects. Read Responsiveness → Connection, Connection → Use and Responsiveness → Use by their full labels. The Use coefficients describe the defined composite, including all three activities. If you later remove an activity, the outcome itself changes; coefficients from the two versions no longer describe precisely the same measure.
Open Estimates → R-squared for Connection and Use. An R-square for Use describes fitted variation in that formed score. It does not mean the model explains the same percentage of every activity separately, nor that it predicts future logs accurately.
For uncertainty choose Calculate… → PLS-SEM Bootstrapping on the same saved model. For this exercise request 1000 Bootstrap samples, Confidence level (%) 95 and Seed 202611075. Review the original estimate, interval method and usable or failed fits. Read the specific indirect Responsiveness → Connection → Use row with its own interval. Keep its calculation separate from the formative measurement weights.
Hypothetical example
Illustrative reading example
Suppose Responsiveness → Connection is 0.50 and Connection → Use is 0.30 in one unmoderated fitted model. Their specific indirect product is 0.15. That product describes the route to the defined Use composite. It does not mean responsiveness raises every activity by 15%, and its interval cannot be obtained by multiplying two path-interval endpoints.
Match the diagnosis to the measurement role
On a small screen, swipe the table sideways to see every column.
| What you notice | Check | Reasonable next step |
|---|---|---|
| Alpha for Use is low | Whether the activities were defined as formative ingredients | Assess content and weights; do not apply an interchangeable-item deletion rule. |
| A Use weight is uncertain but its loading is substantial | Overlap, weight interval and activity meaning | Explain the two statistics separately and consider content coverage. |
| An activity has a negative weight | Coding and collinearity among activities | Investigate suppression or redundancy before making a substantive claim. |
| Responsiveness and Connection overlap strongly | Reflective item wording and HTMT | Revisit whether the measures distinguish the two experiences. |
| The indirect interval crosses zero | The product’s own uncertainty and the defined outcome | Report an uncertain indirect association; do not infer it from component significance alone. |
| Deleting an activity changes the structural path | The altered definition of Use | Present the alternative as a different composite, not a correction with identical meaning. |
Choose a comparison with a clear measurement purpose
A defensible sensitivity comparison might use an independently justified activity set. Save it as a new model and state which definition of Use it represents. Compare the activity coverage, weights and structural relationships while acknowledging that the outcome changed. Do not describe this as a test of the same score under a harmless technical adjustment.
Another useful check keeps all activities but examines whether two columns duplicate the same task category. If they do, the solution is to revisit the activity definition and data collection, not simply to choose the combination with the smallest probability. A study of distinct activities needs content breadth as well as stable calculation.
For a real study, consider validating the formed score against an independently collected overall-use measure when the design permits. This exercise supplies no such external criterion, so it cannot establish that kind of validity. The current task is to understand measurement roles and their consequences for interpretation.
Describe the composite before reporting its coefficients
Methods outline: We treated Responsiveness and Connection as reflective constructs measured by [items]. We defined Use as a formative composite of planning, information and routine activities, each measured over [period and scale]. We assessed reflective measurement separately from formative weights, collinearity and content coverage. We estimated the three stated PLS paths with [settings] and assessed uncertainty using [procedure and counts].
Results outline: The reflective measurements showed [findings]. The Use weights were [estimates and intervals]; [activity] was retained because [content reason], despite [uncertainty if applicable]. Connection’s association with the defined Use composite was [estimate and interval]. The specific indirect association from Responsiveness through Connection was [estimate and interval].
| Include | Purpose |
|---|---|
| Measurement roles | Shows which blocks are reflective and which are formative. |
| Activity definition | Explains what the Use composite includes and excludes. |
| Weights, loadings and overlap | Separates unique contributions from relationships with the formed score. |
| Structural estimates and intervals | Keeps measurement weights distinct from research paths. |
| Alternative definitions | Makes any revised outcome transparent to readers. |
Put this result into a report
We fitted the specified ordinary PLS model to 300 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 why the measurement choice matters
- Name an activity that could be high while another is low, without implying bad data.
- Explain why reflective reliability is not the main criterion for the formative block.
- Read a small weight together with its loading and content role.
- State exactly what the Use outcome includes when interpreting a path.
- Save the original definition before testing any alternative activity set.