Your workshop data
Use these files to practise an analysis and learn how to explain the result. Every row is invented. The files contain no participant records or copied questionnaire responses.
For the first PLS lesson, use the separate 240-case customer-experience exercise. The files below are for the longer method workshops. Match the filename in your lesson before comparing results.
Choose your file
| Download | Rows | Use it for |
|---|---|---|
| research-workshop.csv | 360 | Build the research model and practise measurement checks, prediction, group comparisons and weighted analysis. |
| research-workshop-missing.csv | 360 | Separate missing-indicator exercise. Blank cells are deliberate; compare decisions with the complete file. |
| observed-workshop.csv | 120 | Use directly recorded columns for regression, indirect-path arithmetic, a yes/no outcome and principal component analysis (PCA). |
| observed-workshop-challenge.csv | 120 | Find two known data-entry problems by comparing against the complete teaching file. Do not use the challenge file as the baseline analysis. |
| tetrad-workshop.csv | 360 | Four indicators of one practice concept for CTA; three target indicators are supplied if the prerequisite model needs a structural path. Separate generator seed 20261122. |
| observed-diagnostics.csv | 120 | Inspect separately calculated regression residuals and case influence. A large diagnostic value is a reason to investigate a case, not automatically remove it. |
| observed-ols-reference.csv | 4 | Independently calculated OLS reference with HC3 standard errors and Student-t intervals for y predicted by x, z and m, including an intercept. |
The research model
research-workshop.csv has 360 cases. Start with three reflective indicators for each of Quality, Trust, Satisfaction and Loyalty. Draw Quality → Satisfaction, Trust → Satisfaction and Satisfaction → Loyalty. Keep IDs, group and weight outside the indicator assignments.
quality_1 to quality_3 measure Quality in this teaching model. The trust, satisfaction and loyalty blocks follow the same naming rule. resources_1 to resources_3, support_1 to support_3 and capability_1 to capability_3 are optional blocks for other specifications explained in the guides. They do not establish a validated scale or a particular higher-order theory.
group divides cases evenly into 0 and 1. weight contains positive, unequal practice weights; these are not survey-design weights. age and experience are continuous teaching variables. adoption is a binary 0/1 response. The four columns ending in _mean are simple indicator averages, not estimated construct scores.
The observed-variable exercise
observed-workshop.csv has 120 cases. x and z are predictors, m is a proposed mediator and y is the continuous outcome. binary_outcome codes the event as 1 and the other response as 0. case_id identifies each row; group and weight are available only for lessons that explicitly use them.
Practise finding data problems
Use research-workshop-missing.csv to compare missing-data decisions with the complete research dataset. Blank indicator cells are intentional. Use observed-workshop-challenge.csv to investigate two documented data-entry problems: y for case o017 was multiplied by ten, and z for o042 was replaced by -888. Compare with the original file before correcting them. In real research, a surprising value is not automatically an error.
Numbers you can check
The following reference fits y from x, z and m with an intercept, using all 120 complete cases. These values were calculated separately in Python from the exact published CSV. They are reference calculations for the exercise, not a screenshot or a claimed QuickPLS run.
| Term | Estimate | Standard error | 95% lower | 95% upper |
|---|---|---|---|---|
| Intercept | -0.0605514 | 0.0756005 | -0.210288 | 0.089185 |
| x | 0.29162 | 0.11019 | 0.0733744 | 0.509865 |
| z | 0.00181673 | 0.0721099 | -0.141006 | 0.144639 |
| m | 0.688776 | 0.104405 | 0.481988 | 0.895564 |
R-squared is 0.506093. The model uses 116 residual degrees of freedom: 120 cases minus four estimated coefficients. These intervals use HC3 robust standard errors and a Student-t reference. Select HC3 for this exercise in QuickPLS. The JSON also contains conventional model-based results for a separate comparison; those intervals are not the HC3 output.
Download the full reference answers (JSON), including the mediator and total models, a binary logit model and unrotated correlation-based PCA. For PCA, a component can have the opposite sign and still represent the same component axis; match variables, scaling and rotation before comparing.
Use the files in your own teaching
The original synthetic data are released under CC0 1.0. Keep the data notes with your copy. The generation seed is 20261105. Download the file list and checksums if you need to identify the exact files used.
Download the complete workshop pack · Return to the learning library