PLS-SEM tutorial: a step-by-step guide

What PLS-SEM is

Partial least squares structural equation modeling (PLS-SEM) estimates relationships between latent constructs, such as trust, satisfaction or intention, that are measured with several survey items. It combines a measurement model, which links each construct to its items, with a structural model, which links constructs to one another through hypothesised paths.

PLS-SEM builds each construct as a weighted composite of its items and chooses the weights so that the model explains as much variance as possible in the outcome constructs. That makes it a prediction-oriented, variance-based method, in contrast to covariance-based SEM (CB-SEM), which estimates common factors and tests how well the model reproduces the covariance matrix.

When to use PLS-SEM

Researchers in management, marketing, information systems, tourism and education choose PLS-SEM when:

• the goal is to explain and predict key outcome constructs, or to extend an existing theory;
• the model is complex, with many constructs, items and paths, or includes mediation and moderation;
• the model contains formative constructs, whose items form the construct rather than reflect it;
• the sample is modest, or the data are not normally distributed;
• latent variable scores are needed for follow-up analyses such as fsQCA.

If the aim is to confirm a theory through overall model fit, CB-SEM may be the better choice. Many studies report why they chose PLS-SEM in one or two sentences, citing these reasons.

Step 1: Prepare the data

Arrange the data with one row per respondent and one column per item. Give items short names with a shared prefix, such as TR1, TR2 and TR3 for trust; PLS-SEM Studio uses those prefixes to build the constructs for you.

• Missing data. Remove cases with missing values or, when fewer than about 5% of values per item are missing, replace them with the item mean.
• Straight-lining and outliers. Remove respondents who gave the same answer to every item and check extreme values.
• Distribution. PLS-SEM does not assume normality, but report skewness and kurtosis; the app computes them when you load the file.
• Sample size. The inverse square root method is a common way to justify the minimum sample. A rough rule of thumb is ten times the largest number of arrows pointing at a construct.

The free data cleaning and coding tool handles these steps in your browser: set missing-value codes, reverse-code negatively worded items, remove duplicates and straight-liners, and download a clean CSV for PLS-SEM Studio.

Step 2: Specify the model

Decide for each construct whether it is reflective or formative. Reflective items are interchangeable expressions of the same underlying concept, so they should correlate highly. Formative items each capture a different facet, and dropping one changes what the construct means.

Then draw the paths that represent your hypotheses. In PLS-SEM Studio you choose Draw path, click the cause and then the effect. Indirect routes, such as X → M → Y, are found and tested automatically, and moderators are added by choosing the path they change. If you prefer text, the same model can be written as:

TR =~ TR1 + TR2 + TR3
SAT =~ SAT1 + SAT2 + SAT3
LOY =~ LOY1 + LOY2 + LOY3
SAT ~ TR
LOY ~ SAT + TR

Step 3: Assess the measurement model

Reflective constructs

The Fornell–Larcker criterion and cross-loadings are still often reported, but simulation studies show HTMT detects discriminant validity problems far more reliably (Henseler, Ringle and Sarstedt, 2015). Report HTMT as the main test.

Formative constructs

Check collinearity among the items with the variance inflation factor (VIF below 3, and certainly below 5), then test the significance of each outer weight with bootstrapping. A non-significant weight with a loading of 0.50 or more is usually retained, because the item still contributes absolutely to the construct.

Step 4: Assess the structural model

1. Collinearity. Inner VIF values below 3 show the predictors of each construct are not too strongly related.
2. Path coefficients. Run bootstrapping with 5,000 subsamples and report each standardised coefficient β with its t value, p value and 95% confidence interval. A hypothesis is supported when the interval excludes zero.
3. Explanatory power. R² values of 0.75, 0.50 and 0.25 are often described as substantial, moderate and weak, although what counts as high depends on the field.
4. Effect sizes. f² of 0.02, 0.15 and 0.35 indicate small, medium and large effects.
5. Predictive power. Report PLSpredict: Q²predict above zero, and compare the RMSE (or MAE) of PLS with a linear-model benchmark. If PLS beats the benchmark on all or most indicators, the model has high or medium predictive power (Shmueli et al., 2019).
6. Model fit. Some journals ask for the SRMR; values below 0.08 are commonly read as good fit, though fit indices remain debated in PLS-SEM.

For indirect and interaction effects, see the mediation and moderation guide.

Step 5: Report the results

A typical results section has three parts: the measurement model tables (loadings, reliability and validity, HTMT), the structural model table (hypotheses, β, t, p, confidence interval, f², decision) together with R² and Q², and a path diagram. PLS-SEM Studio exports all of these to Word, Excel or LaTeX and drafts a methods paragraph that names the algorithm settings and bootstrap procedure you used.

Report the software, the weighting scheme, the number of bootstrap subsamples, the type of confidence interval and whether the tests were one- or two-tailed. Reviewers look for all five.

References

• Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A primer on partial least squares structural equation modeling (PLS-SEM) (3rd ed.). Sage.
• Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115–135.
• Shmueli, G., Sarstedt, M., Hair, J. F., Cheah, J.-H., Ting, H., Vaithilingam, S., & Ringle, C. M. (2019). Predictive model assessment in PLS-SEM: Guidelines for using PLSpredict. European Journal of Marketing, 53(11), 2322–2347.
• Dijkstra, T. K., & Henseler, J. (2015). Consistent partial least squares path modeling. MIS Quarterly, 39(2), 297–316.

Related guides

Related free statistical tools