Written and maintained by CASRAI Editorial Board
Published · Last updated
On this page
SmartPLS is a software package for estimating structural equation models with partial least squares (PLS-SEM), which the SmartPLS documentation describes as a composite-based SEM method that estimates construct scores and the structural relationships between them. In practice you import a data file, draw a model of constructs and indicators, run the PLS-SEM algorithm, check the quality criteria (including discriminant validity with the HTMT ratio) and then run bootstrapping to obtain standard errors, t-values and confidence intervals. This guide follows that sequence using the menu names in the current SmartPLS documentation (read 2026-11-14) and lists the settings you should report so a reader can reproduce your run.
What SmartPLS is used for in research
SmartPLS is used when a study has latent constructs measured by several indicators and the research question is about relationships among those constructs: which drivers matter, how strong the paths are, and how well the model predicts.
What is PLS-SEM?
PLS-SEM is the method; SmartPLS is one program that implements it. The SmartPLS guidance on choosing PLS-SEM says the method suits prediction and explanation, identifying key drivers, models with formative constructs and complex models, and it supports prediction-oriented checks such as PLSpredict and CVPAT. The same guidance says covariance-based SEM (CB-SEM) is typically the better fit when the aim is theory confirmation, common-factor models, confirmatory factor analysis or when global model fit is central. It also states that a small sample or non-normal data are not sufficient reasons on their own to choose PLS-SEM, and it advises a multimethod approach where both are plausible. For sample size, the documentation lists power analysis, the inverse square root method, the gamma-exponential method and Monte Carlo power analysis as options rather than a single rule of thumb.
PLS-SEM is not the same as PLS regression. Partial least squares regression is a prediction technique for correlated predictors, covered in What Is Partial Least Squares Regression?; PLS-SEM estimates a path model with latent constructs. If you only need direct, indirect and total effects among observed variables, path analysis is the simpler relative, and for the covariance-based alternative see SEM fit indices and cut-offs.
How to install or access it
Get the program only from the vendor site, smartpls.com. The documentation hub (read 2026-11-14) has sections for Get started, Methods, Practice, Guides, Literature, Courses and an FAQ, and its navigation lists Downloads, system requirements, a free 30-day trial and a “How to cite SmartPLS” page. Check the system requirements page for your operating system before installing, and read the vendor’s current licence and trial terms there, since this guide does not cover them. When you publish, cite the software using the vendor’s own “How to cite SmartPLS” instructions, including the version you used.
Step-by-step: from data file to bootstrapped results
The steps below follow the official tutorials “Import a Data File” and “First PLS path model” (both read 2026-11-14).
1. Import the data
- Select a project in the Project list.
- Right-click to open the context menu and choose Import Data File.
- Locate the file and click Open. The documentation lists these supported types: comma-separated value files (.csv), text files (.txt, optionally tab-separated), SPSS files (.sav) and Excel files (.xls, .xlsx).
- In the dialog that appears, describe your data: set the scale types and minimum and maximum values, and mark missing values. You can reopen the setup dialog later to rename or remove indicators.
The first line of the file supplies the indicator names, so keep column headers short and unique. If your data came from a survey export, check them first; run basic data screening before modelling.
2. Create a PLS-SEM model
- In the Project list, select PLS-SEM in the Main toolbar.
- Enter a model name and save. The Model editor opens.
- Pick indicators from the Indicators list and drag them onto the Modeling canvas. Name the construct and press ENTER.
- For a formative construct, right-click the construct and choose Invert measurement model.
- Use the Connect tool in the Main toolbar: click the source construct and release on the target construct to draw each structural path.
3. Run the PLS-SEM algorithm
- Choose Calculate in the Main toolbar and select the PLS-SEM algorithm.
- Leave the defaults for a first run and make sure Open report is selected.
- Click Start calculation. The Results view opens, with report navigation on the left; the report can be saved or exported to Excel or HTML.
4. Bootstrap for significance
- Choose Edit, then Calculate, then Bootstrapping.
- Set the options described in the data table below (subsamples, confidence interval method, test type, seed).
- Click Start calculation.
The SmartPLS bootstrapping page says the procedure tests path coefficients, outer weights, outer loadings, Cronbach’s alpha, HTMT and R² and returns standard errors, t-values and 95% confidence intervals. For the resampling idea behind it, see Bootstrapping in Statistics.
Output and how to read it
Results open in a report you navigate from the left-hand panel. Two items of the documentation are worth singling out because they decide whether your constructs are separable at all.
Discriminant validity with HTMT
The SmartPLS discriminant validity page says an HTMT value below 0.90 indicates discriminant validity between reflective constructs. HTMT is computed on absolute correlations (the HTMT+ version, since SmartPLS 3.2.1) and appears under Quality Criteria in the PLS and PLSc results. For a bootstrapped HTMT test, the page points to the Complete (slower) bootstrapping option, a one-tailed test, and the upper bound of the bias-corrected confidence interval staying below 0.9, citing Franke and Sarstedt (2019). It describes the Fornell-Larcker criterion and cross-loadings as outdated and cites Henseler, Ringle and Sarstedt (2015) for HTMT. Note that this page gives 0.90, not a stricter 0.85; if your field or journal asks for a different cut-off, say which you applied.
For the conceptual background, see convergent and discriminant validity and the companion page on average variance extracted.
What each check answers
| Question | Where to look in SmartPLS | What the documentation says |
|---|---|---|
| Are two reflective constructs empirically distinct? | Quality Criteria, HTMT | Below 0.90 indicates discriminant validity; bootstrap with Complete (slower), one-tailed, upper bound of the bias-corrected interval below 0.9 |
| Are path coefficients, weights and loadings reliably different from zero? | Bootstrapping report | Standard errors, t-values and 95% confidence intervals are returned |
| How much variance is explained? | R² in the results (also testable by bootstrapping) | R² is among the parameters bootstrapping can test |
| Is the model built for prediction? | PLSpredict, CVPAT | Listed as prediction-oriented assessments supporting the PLS-SEM choice |
Report your settings
A PLS-SEM result is not reproducible unless the run settings are stated. Report the software name and version, the number of subsamples, the confidence interval method, whether the test was one- or two-sided, the significance level, and whether a fixed or random seed was used. The SmartPLS documentation says a random seed applies only in SmartPLS 3, so in the current version state whether you fixed the seed. Also state how many cases and indicators the model used and the rule you applied for sample size (for example power analysis or the inverse square root method), because the documentation says small samples alone do not justify the method.
Troubleshooting
- Nothing opens after calculating. Confirm that Open report was selected before you clicked Start calculation in the algorithm dialog.
- A formative construct is estimated as reflective. Right-click the construct and choose Invert measurement model, as in the first-model tutorial.
- Indicators are missing or misnamed. Indicator names come from the first line of the imported file; reopen the data file setup dialog to rename or remove indicators.
- Imported scale types look wrong. Reopen the setup dialog and correct the scale types, minimum and maximum values and missing-value markers.
- Bootstrapping is very slow. The documentation recommends 1,000 subsamples for initial assessment and 10,000 for the final report, and it offers a parallel processing option; the Complete (slower) HTMT option is slower by design.
Alternatives on CRAN
If you prefer scripted, reproducible analysis, two R packages estimate PLS path models. The details below come from their CRAN pages (read 2026-11-14); check each package’s licence on its page before use. For R basics see R.
| Package | What CRAN says it does | Version and date on CRAN |
|---|---|---|
| seminr | Building and estimating structural equation models by PLS path modeling or covariance-based SEM, including CB confirmatory factor analysis | 2.6.0, published 2026-10-13 |
| plspm | Partial least squares path modeling (PLS-PM), for metric and non-metric data, including REBUS analysis | 0.6.0, published 2025-09-26 |
Frequently asked questions
What is SmartPLS used for?
SmartPLS is used to estimate PLS-SEM models: you define constructs and indicators, run the PLS-SEM algorithm and then bootstrap to test path coefficients, weights, loadings and other parameters.
What is PLS-SEM?
PLS-SEM is a composite-based structural equation modelling method that estimates construct scores and the relationships between constructs. SmartPLS describes it as suited to prediction, explanation, key-driver analysis, formative constructs and complex models.
How many bootstrap subsamples should I use in SmartPLS?
The documentation lists 10,000 as the default, recommends 1,000 for an initial assessment and 10,000 for the final results.
What HTMT value shows discriminant validity in SmartPLS?
The SmartPLS page states that an HTMT value below 0.90 indicates discriminant validity between reflective constructs.
Is PLS-SEM the same as PLS regression?
No. PLS regression is a prediction technique; PLS-SEM estimates a structural model with latent constructs.
Sources
- SmartPLS documentation hub, https://www.smartpls.com/documentation (read 2026-11-14)
- SmartPLS tutorial, Import a Data File (Raw Data or Covariance Matrix), https://www.smartpls.com/documentation/tutorials/import-datafiles (read 2026-11-14)
- SmartPLS tutorial, First PLS path model, https://www.smartpls.com/documentation/tutorials/first-pls-path-model (read 2026-11-14)
- SmartPLS, Bootstrapping, https://smartpls.com/documentation/algorithms-and-techniques/bootstrapping (read 2026-11-14)
- SmartPLS, Discriminant validity assessment, https://www.smartpls.com/documentation/algorithms-and-techniques/discriminant-validity-assessment (read 2026-11-14)
- SmartPLS, More insights on when to use PLS-SEM and when not, https://www.smartpls.com/documentation/choosing-pls-sem/more-insights-on-when-to-use-pls-sem-and-when-not (read 2026-11-14)
- CRAN, seminr, https://cran.r-project.org/web/packages/seminr/index.html (read 2026-11-14)
- CRAN, plspm, https://cran.r-project.org/web/packages/plspm/index.html (read 2026-11-14)
Data: SmartPLS bootstrapping settings: defaults, options and documented recommendations
Bootstrapping options in SmartPLS as described on the vendor's bootstrapping documentation page, compiled into one table of setting, default, available options and the documented recommendation. Compiled from the SmartPLS documentation read on 2026-11-14.
Period covered: 2026.
| Setting | Default | Options | Documented recommendation |
|---|---|---|---|
| Subsamples | 10,000 | Number of bootstrap subsamples; parallel processing option available | 1,000 for initial assessment; 10,000 for final results |
| Confidence interval method | Percentile | Percentile, studentized, bias-corrected and accelerated (BCa) | None stated; BCa suggested if concerned about the bootstrap distribution |
| Test type | n/a | One-sided or two-sided | One-tailed test for the bootstrapped HTMT check |
| Seed | n/a | Fixed seed, or random seed (random seed in SmartPLS 3 only) | State whether the seed was fixed |
| Parameters tested | n/a | Path coefficients, outer weights, outer loadings, Cronbach's alpha, HTMT, R2 | Report standard errors, t-values and 95% confidence intervals |
Sources
- SmartPLS documentation: Bootstrapping (accessed 2026-11-14)
- SmartPLS documentation: Discriminant validity assessment (accessed 2026-11-14)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







