Learn with teaching data
RSearch Tutorials
Learn statistical methods through guided research scenarios and computed teaching examples.
Doctoral research analysis cases
Work through consequential analytical decisions in RSearch with supplied fictional datasets. These are research exercises, not publication guarantees.
When does capability translate into innovation?
Fictional teaching data. Method-specific primary references are provided below; numerical findings must come from your actual run.
Research question and study design
Is digital capability associated with subsequent innovation, and does that association vary with learning climate after accounting for baseline performance and organizational characteristics?
A fictional three-wave organizational study: 360 independent firms, capability and covariates at T1, climate at T2, and innovation at T3. Temporal ordering helps articulate the argument but does not remove selection, omitted-variable or measurement bias. The CSV is simulated teaching material, not evidence about firms.
Supplied dataset and variable roles
One row per firm. Continuous prepared composite scores, a categorical industry field, some blank covariates, and one deliberately influential observation. No results are supplied.
Inspect the actual teaching data
Fictional observations for this research exercise. The table reads the same bundled CSV offered for download.
Loading the bundled teaching data…
Download innovation-boundary-conditions.csv
- Outcome: innovation_t3
- Predictors: digital_capability_t1 and learning_climate_t2
- Controls: baseline_innovation_t1, log_employees_t1, firm_age_t1 and industry
- Identifier only: firm_id, excluded from all fitted models
Decisions before execution
- Specify the interaction from the theoretical boundary-condition argument before inspecting its p value. The interaction changes a conditional slope; it is not a separate treatment effect.
- Use linear regression for the continuous innovation score. A binary adoption outcome would call for a different model; repeated observations per firm would require a dependence-aware design, not this independent-row exercise.
- Keep a justified control set. Baseline innovation addresses prior performance; size, age and industry represent pre-outcome organizational differences. Do not add a post-outcome variable merely because it improves significance.
- Compare covariate-only, additive and interaction models on the same complete rows. If selected-variable missingness changes the analytic sample, prepare one common comparison sample first and record the exclusions.
Carry out the analysis in RSearch
Upload Dataset and Data Readiness
- Download the supplied CSV, open Upload Dataset, and select that file. In Data Readiness, inspect variable types, blank values and the firm identifier.
- Keep industry categorical. The remaining modeled fields are numeric continuous scores or covariates. Do not turn firm_id into a predictor.
- Review and record rows excluded for missing modeled values. Do not silently replace blanks with zero. The tutorial does not claim a missing-data correction from complete-case analysis.
Inspect and record: Save the starting row count, the analytic row count for every model, missingness by variable, measurement definitions and preparation decisions. A change in n can change coefficients independently of the additional predictor.
Multiple Linear Regression configuration
- Select Multiple Linear Regression. In Dependent Variable(s), choose only innovation_t3.
- In Predictor Variables, select digital_capability_t1 and learning_climate_t2. In Control Variables (Optional), select baseline_innovation_t1, log_employees_t1, firm_age_t1 and industry.
- Under Reference Levels, choose Manufacturing as the industry reference and retain that choice for comparisons.
- Keep Include standardized coefficient output selected. For the interaction model, select Center numeric predictors before creating interactions, then add digital_capability_t1 with learning_climate_t2 in Interaction terms (optional). Both main effects stay in the model.
- Run an additive model without the interaction first, then the theory-specified interaction model with the same variables and sample. If estimating a covariate-only model separately, keep its data exclusions aligned with the full model.
Inspect and record: Confirm variable roles, reference category and centering before running. In the interaction fit, the digital-capability main coefficient is conditional on centered climate equal to zero. Centering changes the reference point, not proof of moderation.
Computed results and diagnostics
- Inspect the coefficient table and confidence intervals for the main and interaction terms, not only the omnibus model test. Read the industry coefficients relative to Manufacturing.
- Inspect the available heteroskedasticity, influence and collinearity output, plus robust inference where returned. The current MLR execution contract enables robust inference, interaction probing, influence and heteroskedasticity diagnostics; do not look for an invented on-screen toggle.
- Use the returned conditional-effect or interaction-probing output when available. Restrict interpretation to the observed moderator range. Do not infer a meaningful change from a significant coefficient alone.
- Investigate the large-influence record against source data. Refit a clearly documented sensitivity model only after a reasoned inclusion decision; never delete records just to obtain significance.
Inspect and record: Compare coefficient direction, magnitude, uncertainty, analytic n, explained variance and sensitivity across specifications. Report robust and conventional uncertainty distinctly when both are provided. Do not describe an unavailable result as zero or nonsignificant.
Analysis and reporting artifacts
- Create a specification table containing each model, variable roles, industry reference, centering, n, missing-data exclusions and software/version settings.
- Export the actual results using the results-screen export control available in the current build. Retain the supplied CSV, its filename and SHA-256 from the visible preview, preparation log and exported analysis evidence together.
- Draft a results paragraph reporting unstandardized coefficients and intervals, the interaction and conditional-effect evidence, practical magnitude and influential-case sensitivity. Mark the paragraph as a simulation exercise and cite the output actually obtained.
Limits on the conclusions
- The generator is deliberately heteroskedastic and includes an influential case; no numerical conclusion is prescribed.
- The example uses prepared continuous composites. Scale validation must precede substantive use of real questionnaire scores.
- No random assignment, causal identification, panel estimator, automated causal control selection or journal acceptance claim is provided.
Primary methods references
A mechanism claim with competing antecedents
Fictional teaching data. Method-specific primary references are provided below; numerical findings must come from your actual run.
Research question and study design
Does knowledge integration provide an indirect pathway between digital capability and innovation while learning climate predicts both the mediator and outcome?
A fictional organizational survey with 400 independent organizations, four constructs and four continuous indicators per construct. A five-path partial-mediation model retains both competing antecedents and the direct capability-to-innovation path. This is a researcher-specified mechanism model, not identification of a causal mediation effect.
Supplied dataset and variable roles
Sixteen continuous simulated indicators, a firm identifier and a descriptive sector field. Includes some missing indicators and one intentionally weaker integration indicator. Items are already keyed in the intended direction; these are not ordinal Likert responses.
Inspect the actual teaching data
Fictional observations for this research exercise. The table reads the same bundled CSV offered for download.
Loading the bundled teaching data…
Download capability-mediation-competing-antecedents.csv
- Digital Capability: DC1, DC2, DC3, DC4
- Knowledge Integration: KI1, KI2, KI3, KI4
- Innovation Performance: IP1, IP2, IP3, IP4
- Learning Climate: LC1, LC2, LC3, LC4
- firm_id and sector are not measurement indicators
Decisions before execution
- Choose the estimand first. For a theory in which latent common factors explain reflective indicators, examine Consistent PLS (PLSc). Ordinary PLS estimates composite relationships. Do not choose an estimator solely because a sample is small or its output looks more favorable.
- Retain the direct capability-to-innovation path to avoid forcing full mediation. Learning climate predicts integration and innovation, so the proposed capability pathway competes with an alternative antecedent.
- This exercise is first-order and linear. The interface supports additional features, but higher-order models, interactions, prediction and group tests have different qualification boundaries; do not combine them casually.
- Complete-case exclusion and Stop if any selected value is missing are the available missing-indicator choices in this PLS workflow. Neither is an automatic remedy for informative missingness.
Carry out the analysis in RSearch
Upload Dataset and Data Readiness
- Download and open the supplied CSV. Verify the sixteen indicator fields are numeric continuous and identify the blank indicator values.
- Create the four named constructs from the stated indicator sets using the existing construct setup. Review item meaning, coding and measurement direction. Do not assign firm_id or sector as indicators.
- Record starting n and missing-indicator counts before opening SEM Workspace with this dataset. The generated indicators are continuous; an ordinal-data model would require a separately justified method.
Inspect and record: Confirm that all sixteen indicators are assigned exactly once and the dataset remains available locally. No automatic item reversal or deletion is needed for these supplied fields.
SEM Workspace model board
- Place the four saved constructs on the board with reflective measurement directions.
- Draw Digital Capability → Knowledge Integration, Knowledge Integration → Innovation Performance, Digital Capability → Innovation Performance, Learning Climate → Knowledge Integration, and Learning Climate → Innovation Performance.
- Inspect the complete measurement and structural specification before running. There are four constructs, sixteen indicators and five structural paths, without a feedback cycle.
Inspect and record: Match every arrow to the stated theoretical equation. The model must contain both antecedent paths and the direct effect; an attractive diagram alone does not justify them.
Fit your PLS model
- In PLS method, select Consistent PLS (PLSc) for the common-factor version of this case, or explicitly justify ordinary PLS if the constructs are defined as composites.
- Set Bootstrap samples to 5000 and Confidence level to 95%. Open Reproducibility and advanced settings, set Random seed to 271828 and inspect Missing indicators.
- First choose Stop if any selected value is missing to make the data issue explicit. After documenting the missingness decision, choose Exclude rows with any selected indicator missing for the complete-case exercise.
- Confirm the checkbox stating that measurement directions, independent observations, numeric coding and keyed indicators were reviewed. Click Fit model and bootstrap. Keep the successful/requested resample count with the evidence.
Inspect and record: A failed or inadmissible resample is not silently a successful one. Record the actual analytic n, estimator, seed, confidence level, requested and retained resamples, warnings and package provenance.
Review My Model, Measurement and Structural
- Open Measurement and inspect loadings, reliability, AVE and HTMT evidence. Read KI4 as a content/measurement review issue, not an automatic deletion instruction.
- Open Structural. Inspect the five direct paths, their intervals, endogenous R², structural diagnostics and the Specific indirect effects table. Compare the capability and learning-climate pathways through integration.
- Use the computed indirect-effect bootstrap intervals. Do not multiply interval endpoints or require the direct path to become nonsignificant to claim an indirect association.
- Use Review My Model to organize the returned evidence, while treating its thresholds as review cues rather than validity certificates.
Inspect and record: Keep construct definition, uncertainty, discriminant evidence, path magnitude and alternative explanations together. A path interval excluding zero is not evidence that the research design identifies causation.
Sensitivity, prediction and reproducible artifacts
- Preserve the primary fitted run. If reviewing KI4 suggests a content-validity problem, record the reason before changing indicators, retain at least the supported measurement specification, and refit as a labeled sensitivity analysis.
- If the research claim concerns prediction, run a separate ordinary-PLS fit with Run an optional out-of-sample Prediction check, inspect preferred folds and repeated splits, then open Prediction to compare actual held-out evidence with its benchmark.
- Use Download printable report, Download diagram, and Export research bundle to retain the computed report, visual model, settings and evidence. Open R code and export the executed code. Raw data is excluded from the research bundle, so retain the supplied CSV separately with its filename and SHA-256 from the visible preview.
Inspect and record: Distinguish the common-factor inference question from the composite prediction exercise. Explain changed indicators or estimator choices rather than selecting whichever run yields a preferred conclusion.
Analysis and reporting artifacts
- Prepare a measurement table with coding, loadings, reliability and validity evidence, a structural table with direct and specific indirect effects plus confidence intervals, and a specification/sensitivity appendix.
- Report actual analytic n, missingness exclusions, estimator rationale, bootstrap settings and retained resamples. Include a model diagram and the exported research bundle/code.
- Write a bounded mechanism interpretation: the model is consistent or inconsistent with the proposed indirect association in simulated data. Identify nonexperimental design, omitted confounding, construct ambiguity and measurement decisions as limitations.
Limits on the conclusions
- This is synthetic teaching material and provides no organizational findings or publication-quality guarantee.
- Bootstrap uncertainty does not repair omitted-variable bias, common-method bias or weak causal identification.
- Do not claim common-factor measurement invariance from MICOM for composites, and do not claim higher-order or interaction prediction from this first-order example.
Primary methods references
Introductory prerequisites
These smaller exercises introduce individual methods. They do not replace the substantive research cases above.
Do two groups differ?
Compare satisfaction scores, understand why Welch's t test fits, and read a computed result.
Explore the guided example →Structural equation modelingHow does capability become innovation?
Explore four constructs, twelve indicators, and direct and mediated relationships. Fit the model on this device.
Explore the SEM model →Want to see the application in action? Check Out Demos.