What Is PLS? How to Read PLS-SEM and Choose Evaluation Thresholds

What is PLS

PLS usually refers to Partial Least Squares, a predictive approach to structural equation modeling. In a thesis, you will often see the full term PLS-SEM, meaning Partial Least Squares Structural Equation Modeling. This method allows you to assess two parts at the same time: the measurement model, which covers the relationships between latent variables and observed variables, and the structural model, which covers the paths between latent variables.

When you run PLS-SEM in SmartPLS, the software does not return one single PLS number. You will read several groups of results, including Outer Loadings, Cronbach's Alpha, Composite Reliability, AVE, HTMT, VIF, Path Coefficients, R², f² and Q². The question “what PLS value is acceptable” therefore needs to be separated into specific indicators.

A short way to think about it is this: the measurement model answers whether the scale measures the construct properly, while the structural model answers how well the predictors explain and predict the dependent variable. PLS is useful for research with complex models, many latent variables, a strong focus on prediction, or a sample size that does not meet the requirements of some covariance-based SEM methods. You still need to check your data quality, theory and hypotheses before opening SmartPLS.

The role of PLS in quantitative research

PLS-SEM helps you test a hypothesized cause-and-effect model. For example, you may hypothesize that Perceived Usefulness and Perceived Ease of Use affect Intention to Use. In the model, the first two constructs are predictor variables, Intention to Use is the explained variable, and the arrows represent the directions of the relationships that need to be tested.

PLS results can support four main decisions. First, you check whether the observed variables load onto the correct latent variables. Second, you check whether the constructs have adequate reliability, convergent validity and discriminant validity. Third, you assess path coefficients, p-value or confidence intervals from bootstrapping to reach a conclusion about the hypotheses. Fourth, you examine R², f² and Q² to understand how the model explains, contributes to and predicts the dependent variable.

PLS commonly appears in studies using Likert questionnaires. You can also read what is a scale if you are still distinguishing between latent variables, observed variables and scales. Keep your data-column names consistent, such as PU1, PU2 and PU3 for the observed variables of Perceived Usefulness. SmartPLS reads .csv data, where each row is usually one respondent and each column is one observed variable.

PLS-SEM and CB-SEM serve different purposes. SmartPLS is often selected when the research emphasizes prediction, the model contains many constructs, or the data do not meet distributional requirements well. AMOS and other CB-SEM methods are commonly used when the main goal is to assess the overall fit of a theoretical model. The CFI, TLI, RMSEA and SRMR indicators from (Hu and Bentler, 1999) belong to CB-SEM. You should not place them in a SmartPLS criteria table.

If your model has a mediator or moderator, PLS can still handle it, but you need to read the role of each variable correctly. A mediator explains the mechanism of an indirect effect, while a moderator changes the strength or direction of a relationship. Read the separate guides on mediators, moderators and moderating before building the model.

What PLS value is acceptable

There is no single threshold for PLS as a whole. Read the results in sequence, starting with the measurement model and then moving to the structural model. The table below summarizes commonly used thresholds and their sources. These are reference points for explaining your decisions, not automatic reasons to delete a variable before checking the scale content and theoretical basis.

For the measurement model, do not look only at Outer Loading. A scale can have good loadings while failing HTMT, which means that two constructs are not sufficiently distinct. Conversely, a low loading needs to be considered together with the item content, AVE, Composite Reliability and the effect of deleting the item.

For the structural model, R² is not a p-value. R² shows the proportion of variance in the dependent variable explained by the predictors in the model. The path coefficient and confidence interval help you conclude the direction and statistical significance of each hypothesis. The R² thresholds of 0.75, 0.50 and 0.25 are presented according to (Hair et al., 2011), but an acceptable R² also depends on the field and research objective.

How to read PLS in SmartPLS output

Open your model results in SmartPLS 4 and read them in two layers. The first layer is Quality Criteria, where you check the scales. The second is the Bootstrapping output, where you test the paths. The table below is illustrative output, not the result of a real study.

In the Outer Loadings table, check which items have low loadings or unusual cross-loadings. Under Construct Reliability and Validity, read Cronbach's Alpha, rho_A, Composite Reliability and AVE. PLS may report several reliability coefficients, but you need to state your criteria consistently in the methods section. Do not choose the most favorable indicator only after seeing the results.

In Discriminant Validity, HTMT is an important criterion. If HTMT between two constructs exceeds the threshold, check how the items are worded, whether the constructs are too similar, and what the theory says. You can also report the Fornell-Larcker Criterion according to (Fornell and Larcker, 1981), where the square root of a construct's AVE is compared with the correlations between that construct and the others.

In Path Coefficients, a positive sign indicates a positive relationship in the sample, while a negative sign indicates a negative relationship. The coefficient sign alone is not enough to conclude that a hypothesis is supported. Check T Statistics, P Values or the bootstrap confidence interval. If the confidence interval for the coefficient does not contain 0, the conclusion that the effect is significant is supported under the bootstrap approach of (Preacher and Hayes, 2008).

How to write this in your thesis

You can write: “The PLS-SEM results show that the effect of [independent variable] on [dependent variable] has a path coefficient of β = [value], t = [value], p = [value]. Therefore, hypothesis H[ ] is [accepted, not accepted] in the study sample.” Replace every bracketed section with the actual values from your exported file.

What to do when PLS does not meet the thresholds

If Outer Loading is below the threshold, first check the variable coding and scale direction. An item that was not recoded correctly can have a sharply reduced loading. Check whether the data were entered incorrectly, whether there are many missing values, or whether a respondent selected the same response level for nearly the entire questionnaire.

If an item has a low loading but is theoretically important, do not delete it immediately. Check Composite Reliability and AVE first, then run the model with and without the item for comparison. Deleting an item needs both a methodological and substantive reason, and you should record the decision in the thesis. Repeatedly deleting items until every indicator looks good makes the model difficult to defend.

If HTMT does not meet the threshold, check whether the two constructs measure the same content. If the scales genuinely overlap, you may need to adjust the theoretical model, combine the constructs when there is a basis for doing so, or retain them and report the limitation. Do not rename a variable after seeing the results just to reduce HTMT.

If VIF is high, check whether the predictor variables are worded almost identically. Also review the correlation matrix, coding and hypothesis structure. Deleting an item because of high VIF should be based on theory, not on one run of the model.

If R² or Q² is low, the software has not automatically proved that the model is wrong. The dependent variable may be affected by factors not included in the model, the scale may not fit the context, or the research objective may focus on explaining a mechanism rather than prediction. Report the result accurately, discuss reasonable causes, and do not promise that adding a few variables will certainly increase R².

Distinguishing PLS from CB-SEM and AMOS

PLS-SEM and CB-SEM can both analyze structural models, but their evaluation logic is different. PLS focuses on explained variance and predictive ability. CB-SEM places more emphasis on how well the covariance matrix fits the theoretical model. Your choice between SmartPLS and AMOS should therefore begin with the research objective, data type and supervisor's requirements.

SmartPLS is useful when you need to assess the measurement model, paths, R², f², Q² and bootstrapping in a model with many constructs. AMOS is used for CB-SEM procedures and fit indicators such as CFI, TLI, RMSEA and SRMR. Do not take the AMOS CFI threshold and place it in a PLS criteria table, and do not use PLS R² as a substitute for the full fit assessment of CB-SEM.

PLS also differs from ordinary linear regression. Regression usually works directly with observed variables or composite scores, while PLS-SEM models latent variables and measurement error at the same time. If your model has one dependent variable and a few independent variables that have already been converted into composite scores, SPSS regression may be the more concise option. If it has several latent variables, a mediator or a measurement model that needs testing, PLS-SEM may be more appropriate.

Common mistakes when reading PLS

The first mistake is calling R², AVE or HTMT collectively “an acceptable PLS indicator”. Each indicator answers a different question, so your results table should clearly state the indicator name, value, threshold and source.

The second mistake is reporting a p-value of 0.000 as though the true probability were zero. This is how SmartPLS displays a small value after rounding. You can report p < 0.001 when that format fits the conventions of your thesis.

The third mistake is reporting only significant paths and omitting nonsignificant paths. The committee will often ask how many hypotheses the model contains, which hypotheses are supported and which are not. Keep both groups in the summary table.

The fourth mistake is changing the model repeatedly without saving versions. Save the original data, cleaned file, initial diagram and every item-deletion round. If you have outliers, read how to handle outliers, but do not delete respondents only because they make R² or the p-value look less favorable.

Frequently asked questions

What is PLS, and is PLS-SEM the same thing?

PLS is short for Partial Least Squares. PLS-SEM is the full term used when the method models structural equations, including a measurement model and a structural model. In a thesis using SmartPLS, the two terms often appear together, but PLS-SEM is more precise when you mean the complete procedure.

Does PLS require normally distributed data?

PLS-SEM is generally less dependent on the normality assumption than some CB-SEM methods. That does not mean incorrectly entered data, extensive missing data or poor-quality responses can be accepted. You still need to check coding, missing values, outliers and the reasonableness of each observed variable.

What PLS value is acceptable?

Ask which indicator you mean. Outer Loading is commonly referenced from 0.7, Composite Reliability from 0.7, AVE from 0.5, HTMT below 0.85 or 0.90 depending on how close the constructs are, VIF below 5, and Q² greater than 0. Each threshold needs its source and must be read in the context of the model. Do not use these numbers as automatic item-deletion rules.

Should I use SmartPLS or AMOS?

Choose SmartPLS when the objective emphasizes prediction, the model has many latent variables, or you need R², Q² and bootstrapping under PLS-SEM. Choose AMOS when the study follows CB-SEM and needs fit indicators such as CFI, TLI and RMSEA. Confirm the choice against your theoretical model and supervisor's requirements before running the analysis.

Does an item with an unacceptable Outer Loading have to be deleted immediately?

Check the coding direction, data quality, item content, and the item's effect on AVE, Composite Reliability and discriminant validity. If you delete it, state the reason, save the version before and after deletion, and avoid deleting items in bulk only to meet a threshold. An observed variable that is substantively important may need to be retained with a clear limitation.

Open your .csv file, check the variable names, and create a tracking table containing Outer Loading, AVE, HTMT, VIF, R², Q² and the bootstrapping results before writing Chapter 4. If you need to run PLS-SEM on real data and turn the results into tables and thesis paragraphs, you can use DoThesis M4 for SPSS and SmartPLS data analysis.