Frequently Asked Questions

Metis is a dedicated desktop application for Partial Least Squares Structural Equation Modelling (PLS-SEM). It brings dataset management, visual path model building, statistical analysis, interactive results exploration, and reporting into one unified workspace.

The statistical calculations run through R, primarily using the acclaimed seminr package, while Metis provides the visual graphical workflow around the entire analysis lifecycle.

Metis is designed for PhD students, academic researchers, university lecturers, data analysts, and quantitative researchers who work with PLS-SEM and prefer an intuitive, visual workflow without having to write R scripts for every analysis. It is also valuable to experienced R users who want a rapid graphical modelling environment while retaining full access to the underlying statistical workflow and reproducible R scripts.

Yes. Metis is distributed 100% free of charge and is released as open-source software under the GNU General Public Licence v3 (GPL-3). There are no subscriptions, seat licenses, time limits, student restrictions, or paywalled features required to run analyses.

No. Metis is an independent PLS-SEM environment with its own workflow and development priorities. Different statistical applications (such as SmartPLS, ADANCO, WarpPLS, AMOS, or lavaan) serve different user needs. The aim of Metis is to provide an accessible, transparent, local, and reproducible PLS-SEM workflow rather than to reproduce every feature or interface of another program.

Metis provides a completely free, open-source (GPL-3) desktop alternative for Windows and macOS. Unlike commercial PLS-SEM tools that require annual subscriptions or restrict features on student tiers, Metis offers full, unrestricted access to advanced modeling methods—including multi-core BCa bootstrapping, PLSpredict, CVPAT, IPMA, NCA, cIPMA, 3-step MICOM measurement invariance, MGA, and automated Tark APA 7 Word reporting—while processing all data 100% locally on your machine with no cloud dependencies.

Both installers provide the complete Metis application. The difference is how the R environment required for statistical calculation is provided:

• Bundle is recommended for most users. It includes the complete embedded R calculation environment Metis needs and does not require you to install R separately.
• Lite is intended for users who already have R installed on their computer. It connects to your existing R installation and checks that the required packages are available.

If you do not normally work in R, choose the Bundle installer.

No. Metis executes R in the background, but normal dataset preparation, model building, parameter estimation, and report generation are performed entirely through the graphical user interface. Researchers who want to inspect or reproduce their analysis in R can also export or copy the corresponding R script at any time.

Not if you download the Bundle edition. The Lite edition requires an existing R installation. Metis checks the required packages (such as seminr) during setup and provides guidance if dependencies need updating.

No. Once installed, Metis performs all statistical calculations locally on your computer without requiring an active internet connection. Internet access is only needed to download the initial installer or to fetch missing R packages when using the Lite edition.

The current release supports:
• Windows: Windows 10 and Windows 11, 64-bit
• macOS: macOS 11 (Big Sur) and newer, with native builds for both Apple Silicon (M1/M2/M3/M4) and Intel Macs

Metis currently supports:
• CSV (Comma-separated values)
• XLSX (Modern Microsoft Excel)
• XLS (Legacy Microsoft Excel)

The import preview allows you to verify variable types, inspect missing value indicators, and make quick corrections before constructing your model.

Yes. The built-in Data View supports practical data preparation tasks including editing cell values, renaming indicator variables, deleting rows or columns, filtering subsets, recoding values, and creating simple computed sum/average variables. For substantial data wrangling or complex multi-wave mergers, a dedicated statistical application may still be helpful.

No. Your research data, model specifications, calculations, workspaces, saved results, and Tark reports are processed and stored exclusively on your local machine. Metis never transmits your dataset to cloud servers, ensuring full compliance with GDPR, institutional IRB boards, and strict academic data confidentiality protocols.

No account or registration is required. You can install and start analyzing models immediately without login screens, telemetry accounts, or API keys.

Metis supports both reflective (Mode A) and formative (Mode B) measurement models. The drag-and-drop model canvas enables you to create latent constructs, assign manifest indicators, configure reflective or formative measurement directions, and draw structural paths (hypotheses) between constructs.

In reflective measurement models (Mode A), indicators are manifestations of the underlying latent construct (e.g., personality traits or satisfaction dimensions). Changes in the construct cause changes in all its indicators, meaning items should correlate positively and exhibit high internal consistency (outer loadings ≥ 0.708, composite reliability ρ_a / ρ_c, Cronbach's alpha ≥ 0.70, and AVE ≥ 0.50).

In formative measurement models (Mode B), indicators define or cause the construct (e.g., an economic index composed of distinct income and asset sources). Items are not expected to correlate with each other, and the construct is evaluated through regression-based outer weights, bootstrap significance, and indicator collinearity (VIF < 3.3 or < 5.0) rather than internal reliability.

Yes. Metis provides visual support for higher-order constructs (second-order constructs) and supports standard estimation approaches (such as the repeated indicators approach and two-stage estimation) where permitted by the selected analysis. When a specific higher-order configuration is not compatible with a chosen procedure, Metis provides an informative guidance message.

Yes. Moderation relationships can be established directly on the model canvas by connecting a moderator construct to an existing structural path. Metis estimates the interaction terms (two-stage approach) and provides interaction effect statistics, p-values, and simple slope data.

No. Metis estimates the model specified by the user. Decisions regarding construct conceptualisation, reflective vs formative operationalisation, causal ordering, hypothesis derivation, and theoretical justification remain the responsibility of the researcher. Statistical software assists the estimation process but does not replace theoretical reasoning.

The analytical suite in Metis includes:
• PLS-SEM Path Modeling: Standard PLS algorithm with path, factor, and centroid weighting
• Bootstrapping: Non-parametric significance testing, t-values, standard errors, and confidence intervals
• PLSpredict: Out-of-sample predictive assessment with k-fold cross-validation
• CVPAT: Cross-Validated Predictive Ability Test comparing rival models
• IPMA: Importance-Performance Map Analysis identifying priority constructs
• NCA: Necessary Condition Analysis identifying bottleneck thresholds and ceiling lines
• cIPMA: Combined Importance-Performance & Necessity Matrix
• MICOM: 3-step Measurement Invariance of Composite Models Beta
• MGA: 2-group Multigroup Analysis (parametric and non-parametric PLS-MGA)
• Higher-Order Constructs & Moderation: Second-order models and interaction effects
• Tark Reporting: Automated APA 7th edition table generation and Word (.docx) export

Depending on your model, the results include comprehensive assessment metrics:
• Structural Model: Path coefficients (β), direct, indirect, and total mediation effects, R², adjusted R², f² effect sizes, and collinearity VIF diagnostics
• Construct Reliability & Validity: Cronbach's alpha, composite reliability (ρ_a, ρ_c), and Average Variance Extracted (AVE ≥ 0.50)
• Discriminant Validity: Heterotrait-Monotrait Ratio (HTMT < 0.85/0.90), Fornell-Larcker criterion, and cross-loadings matrix
• Measurement Metrics: Outer loadings, outer weights, indicator collinearity, and overall model fit (SRMR)

Metis computes the Heterotrait-Monotrait Ratio of correlations (HTMT) following Henseler, Ringle, and Sarstedt (2015). Modern PLS-SEM methodology regards HTMT as superior to the traditional Fornell-Larcker criterion in detecting discriminant validity issues.

• HTMT < 0.85: Recommended conservative threshold when constructs are conceptually distinct.
• HTMT < 0.90: Acceptable threshold when constructs represent conceptually similar dimensions.
• Bootstrapped HTMT Inference: In the Bootstrapping results, Metis reports confidence intervals for HTMT. If the upper confidence interval bound is significantly below 1.0 (HTMT_inference), discriminant validity is firmly established.

Metis calculates Variance Inflation Factor (VIF) values for both inner structural predictors and outer formative indicators (Hair et al., 2019):

• VIF < 3.0 or 3.3: Ideal, indicating no critical multicollinearity.
• VIF between 3.3 and 5.0: Acceptable in exploratory research and complex models.
• VIF > 5.0: Problematic collinearity that may distort path coefficients and standard errors, suggesting that indicators should be combined, removed, or modeled as higher-order dimensions.

Some statistics are only applicable when the model contains the corresponding relationships or measurement types. For instance, a model without mediator constructs will not have specific indirect effects, a purely reflective model does not report formative outer weights, and moderation tables require an active interaction relationship. Metis clearly explains these contextual absences rather than populating invalid values.

Bootstrapping repeatedly estimates the model using random resamples with replacement from the original dataset. It evaluates the empirical sampling distribution, producing standard errors, t-statistics, p-values, and Percentile and Bias-Corrected and Accelerated (BCa) confidence intervals for path coefficients, indirect mediation effects, and indicator loadings. Researchers can configure the number of bootstrap samples (e.g. 5,000 or 10,000 resamples for final publications) and random seed values.

For exploratory modeling and initial testing, 500 to 1,000 resamples provide rapid calculation. For final doctoral dissertations, conference proceedings, and peer-reviewed journal submissions, methodological authorities (Hair et al., 2022; Streukens & Leroi-Werelds, 2016) recommend configuring 5,000 to 10,000 bootstrap resamples with Bias-Corrected and Accelerated (BCa) or Percentile confidence intervals to ensure stable, reliable standard errors and p-values.

A smaller default (e.g., 500 resamples) allows fast calculation while building, refining, and verifying a model interactively. For final dissertation or journal publication reporting, researchers should increase the setting to 5,000 or 10,000 bootstrap resamples in accordance with established methodological guidelines (Hair et al., 2022).

PLSpredict evaluates the out-of-sample predictive power of a PLS-SEM model using cross-validation (e.g., 10 folds, 10 repetitions). Metis generates predictive assessment statistics including Q²_predict, RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), and automatic benchmark comparisons against a naïve Linear Model (LM) to distinguish explanatory power from true out-of-sample predictive ability (Shmueli et al., 2019).

The coefficient of determination (R²) measures in-sample explanatory power on the observations used to estimate parameters. However, high R² does not guarantee that the model can accurately predict new, unseen cases.

PLSpredict uses k-fold cross-validation to assess genuine out-of-sample predictive performance on held-out test data. If PLS prediction errors (RMSE or MAE) are lower than the naïve Linear Model (LM) benchmark across all indicators and Q²_predict > 0, the model exhibits strong out-of-sample predictive validity (Shmueli et al., 2019).

The Cross-Validated Predictive Ability Test (CVPAT) provides a formal statistical test (Liengaard et al., 2021) for comparing the predictive performance between alternative models or benchmarks based on cross-validated prediction loss functions. In Metis, CVPAT can be activated alongside PLSpredict to test whether a proposed model demonstrates statistically superior predictive capability.

These post-hoc analyses provide actionable managerial and empirical insights:
• IPMA (Importance-Performance Map Analysis): Contrasts total importance (total effects) against performance (unstandardized construct scores on a 0-100 scale) to pinpoint high-priority areas for improvement.
• NCA (Necessary Condition Analysis): Identifies whether specific construct levels represent non-negotiable bottlenecks (using ceiling lines CE-FDH and CR-FDH and necessity effect size d) for reaching outcome targets.
• cIPMA (Combined IPMA & NCA): Integrates importance, performance, and necessity conditions into a comprehensive decision-making matrix.

IPMA, NCA, and cIPMA rely directly on construct scores, path coefficients, and measurement parameters calculated during the baseline PLS-SEM estimation. Metis requires a valid, converged base model to ensure that downstream advanced analyses are computed on verified parameters.

Yes. Metis includes the full 3-step Measurement Invariance of Composite Models (MICOM) procedure (Henseler et al., 2016):
1. Step 1: Configural Invariance (identical indicators, identical model specifications, and identical data treatment)
2. Step 2: Compositional Invariance (permutation test verifying original correlation c equals 1 against the 5% permutation quantile)
3. Step 3: Equality of Composite Means and Variances (permutation test establishing full vs. partial measurement invariance)

MICOM is currently labelled as a beta feature as it undergoes continuous empirical validation and parallel performance optimization across releases.

Before comparing structural path differences across sub-populations (e.g., genders, countries, or customer segments), researchers must confirm that the constructs represent the same underlying phenomenon across all groups. Establishing compositional invariance via MICOM ensures that observed structural differences reflect genuine group disparities rather than measurement artifacts.

Yes. The MGA workflow enables 2-group comparisons, reporting group-specific parameters alongside statistical tests of path differences using both Henseler's non-parametric PLS-MGA permutation approach and parametric tests (e.g., Welch-Satterthwaite approximation) where appropriate.

The grouping variable must contain the distinct categories required for comparison and have sufficient sample size in each group. For 2-group MICOM and MGA, Metis verifies that the selected column yields exactly two valid comparison groups with adequate observations after missing values are filtered.

Different PLS-SEM programs can implement the same core method while utilizing different defaults or computational nuances. Numerical differences typically stem from:
• Data preprocessing: Missing-value imputation, standardization routines, and indicator scaling
• Algorithm settings: Weighting schemes (Path vs Factor vs Centroid), convergence stop criteria (e.g. 10^-7), and maximum iterations
• Sign orientation & correction: Individual indicator vs construct-level sign alignment
• Bootstrapping nuances: Random number generator seeds, resampling algorithms, and confidence interval formulas (BCa vs percentile)
• Cross-validation: Random partitioning folds and repetitions in PLSpredict

When comparing Metis with software like SmartPLS or ADANCO, ensure identical datasets, model configurations, weighting schemes, and random seeds are used.

Not exclusively. While seminr provides the statistical estimation engine, Metis adds an entire native desktop architecture: visual model design canvas, interactive dataset editing, model validation rules, real-time result exploration, publication-ready Tark report generation, parallel computing orchestration, and custom post-hoc modules.

Yes. Metis enables researchers to export or copy the complete R replication script. This allows full code inspection, automated audit trails, and replication in RStudio or high-performance clusters for open-science compliance.

Validation is a core, continuous part of Metis engineering. Analytical components are tested against established benchmark datasets (e.g., Corporate Reputation, ECSI), published SEM literature examples, and cross-verified against R packages like cSEM and seminr under matched parameter configurations.

Yes. Researchers should always inspect diagnostic metrics, collinearity indicators, and methodological assumptions before finalizing publications. Metis champions research transparency by providing inspectable settings, diagnostic summaries, and reproducible R scripts rather than operating as a black box.

No. Viewing results from an exploratory run does not automatically commit them to permanent project storage. Click Save Results when you wish to preserve a specific model estimation for later inspection, historical comparison, or inclusion in Tark reports.

Tark is the built-in automated reporting assistant in Metis. It compiles saved model results into formatted, journal-ready APA 7th edition tables and vector path diagrams, which can be exported directly into editable Microsoft Word (.docx) documents, eliminating hours of manual formatting work.

To generate a complete, publication-ready research report:
1. Select Tark it → Create Report from the top navigation or Results View.
2. Choose the saved PLS-SEM base model and accompanying Bootstrapping, PLSpredict, MICOM, or MGA runs you wish to include.
3. Configure vector path diagram options (loadings, path coefficients, p-values, R²).
4. Click Create Tark Report to generate an editable Microsoft Word (.docx) file featuring standard APA 7th edition table borders, proper variable alignment, bold headers, and automatic statistical note blocks.

No. Tark generates formatted statistical tables, parameter summaries, and diagrams. The researcher remains responsible for interpreting the findings, contextualizing hypotheses within theoretical frameworks, discussing implications, and addressing study limitations.

Tark formats tables according to standard APA 7th edition academic guidelines, which are accepted by major publishers (Elsevier, Springer, Wiley, Taylor & Francis, Emerald). Researchers should always review the specific author guidelines of their target journal for any unique stylistic adjustments.

Metis supports extensive export capabilities:
• Clipboard: 1-click copy formatted tables for Word or Google Docs
• Spreadsheets: Download tables as CSV or Excel (.xlsx) files
• Full Word Reports: Complete structured .docx reports via Tark
• HTML Reports: Standalone browser-viewable analysis summaries
• R Scripts: Complete reproducible R code
• Vector Diagrams: High-resolution path models in SVG and PNG formats

Yes. Metis is built specifically for academic and scientific research workflows. Researchers must ensure that chosen statistical methods are suitable for their research questions and clearly report all analytical configurations, sample sizes, and diagnostic thresholds.

No. Metis is actively developed and focuses on delivering validated, robust PLS-SEM procedures. When a specific model configuration is incompatible with an analysis, Metis prevents execution and explains why, avoiding erroneous or unvalidated results.

A beta tag denotes features (such as MICOM) that are fully functional but undergoing active validation, user feedback, and optimization across edge-case configurations.

Resampling procedures such as Bootstrapping (5,000+ resamples), PLSpredict (100+ re-estimations), MICOM permutation tests, and MGA permutations involve thousands of iterative model estimations. Execution time depends on sample size, indicator count, and CPU cores. While calculations run locally in the background, you can pass the time with the built-in 2048 mini-game in the desktop app.

First, review model specifications, indicator assignments, missing value treatments, and calculation settings. If unexpected behavior persists, reproduce the issue using the exported R script and report it via our GitHub Issues channel with reproducible steps (excluding confidential data).

Submit bug reports or feature requests via our GitHub Issues tracker or our web Feedback Form. Include the Metis version (v0.3.1), operating system, analysis type, and error message.

Consult the Metis User Guide and established methodological textbooks (e.g., Hair et al., A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM), 3rd Edition; Sarstedt et al., 2022).

You can rate Metis directly inside the desktop application by opening Help in the application menu and selecting Rate Metis. Your feedback guides our feature development and statistical enhancements. You can also submit reviews via our Feedback Form.

If you cannot find an answer in this FAQ or the Documentation, feel free to reach out directly:

• Open a GitHub Issue
• Submit Feedback Form
• Contact Support