What Is SmartPLS? How to Read the Output and Evaluation Thresholds

What Is SmartPLS

SmartPLS, which is the software's correct name, is a tool for running PLS-SEM, a variance-based structural equation modeling method. You use it when your model contains latent variables measured by multiple indicators and you want to assess both the scales and the relationships between constructs.

In SmartPLS, you import survey data in .csv format, create constructs, draw arrows between constructs, and then run the PLS algorithm. You usually read the results in two groups. The first is the measurement model, which includes Outer Loadings, Cronbach's Alpha, Composite Reliability, rho_A, AVE, and HTMT. The second is the structural model, which includes Path Coefficients, R², f², Q², VIF, and the Bootstrapping results.

You can think of the SmartPLS workflow as follows:

Survey data → measurement model → structural model → Bootstrapping → hypothesis conclusions

If you are new to this method, see what is partial least squares to distinguish PLS from the covariance-based SEM approach. The PLS-SEM article is useful when you need the full process, from model design to reporting the results.

The Role of SmartPLS in Quantitative Research

SmartPLS helps you answer two questions in a quantitative thesis. First, do the indicators measure the construct you have named correctly. Second, do the constructs affect one another in the directions stated by your research hypotheses.

Suppose your model proposes that Perceived Usefulness affects Behavioral Intention. Each construct is measured by three or four indicators. Before you conclude anything about the hypotheses, you need to check outer loading, reliability, AVE, and HTMT. If the measurement model is not acceptable, interpreting the path coefficient has a weak foundation.

Once the measurement model reaches an acceptable level, you examine Path Coefficients and run Bootstrapping. A positive path coefficient shows a positive direction of association in the sample. A positive sign alone does not establish that the hypothesis is supported. You also need to examine T Statistics, P Values, or the bootstrap confidence interval.

SmartPLS is designed for PLS-SEM, whereas AMOS is commonly used for CB-SEM. For that reason, CFI, TLI, and RMSEA should not be placed in a SmartPLS report as though they were the main PLS-SEM criteria. If you are confusing the two SEM approaches, what is SEM can help you distinguish their areas of use.

You also need to remember that the software does not decide whether your model is theoretically reasonable. SmartPLS calculates results from the data and structure you enter. A path with a small p-value does not prove that the model fits every context. You still need to rely on the theoretical foundation, the scale, the sampling method, and the research questions.

What Counts as Acceptable in SmartPLS

There is no single number that determines whether an entire SmartPLS model is acceptable. Each group of indicators has a different purpose. The table below presents thresholds commonly cited in PLS-SEM theses. Use them as reference points when reading the output, not as mechanical rules for deleting indicators.

For outer loading, 0.70 is commonly used in PLS-SEM. An indicator below this threshold does not automatically have to be deleted. Check AVE, reliability after deletion, the indicator's content, and the theoretical reason for keeping it. If you delete an indicator only to make the statistics look better, the committee may ask why it was removed.

For HTMT, a value below 0.85 is commonly used when the constructs are clearly different. A threshold of 0.90 may be considered when two concepts are closely related, but you need to explain the context. For AVE, a threshold of 0.50 means that a construct explains a substantial portion of the variance in its indicators under the commonly used criterion described in the foundational literature and in what is AVE.

R² is not the percentage of a particular hypothesis that is correct or incorrect. R² indicates how much of the dependent construct is explained by the independent variables in the model. For example, R² of 0.52 can be reported as the model explaining 52 percent of the variance in the dependent construct, followed by an interpretation appropriate to the context.

How to Read SmartPLS Output

In SmartPLS 4, you usually begin by selecting the model, running PLS-SEM Algorithm, and then opening the results in Results. To assess the statistical significance of the paths, select Bootstrapping and check the relevant result tables. The interface in SmartPLS 3 and SmartPLS 4 may place buttons in different locations, but the indicator groups and reading logic remain similar.

The following is illustrative output, not the result of a real study. The column names follow the format commonly displayed in SmartPLS output so you can compare them with the file currently open.

In the PU → BI row, Original Sample (O) is 0.428, the estimated path coefficient from the original sample. T Statistics is 5.284, and P Values is displayed as 0.000 because the software rounds the value. In your thesis, do not interpret 0.000 as a probability equal to zero. You can write p < 0.001 if that matches your institution's reporting convention.

The TR → BI row has a positive coefficient but a p-value of 0.109 in the illustrative output. If your test threshold is 5 percent, this result does not provide sufficient statistical evidence to support the effect of TR on BI. State the conclusion that follows from the data. Do not rename the hypothesis to make the result look more favorable.

For the measurement model, open Outer Loadings to examine each indicator. Part of an illustrative output can be read as follows:

This is also illustrative output. All outer loadings exceed 0.70, Composite Reliability exceeds 0.70, and AVE exceeds 0.50. When you read your actual file, check every construct instead of looking at one summary row. For discriminant validity, open Discriminant Validity and examine the HTMT table. If a construct exceeds the threshold, do not immediately delete the indicator with the lowest loading. The problem may be overlapping content between two constructs.

What to Do When SmartPLS Results Do Not Meet the Criteria

Start with the data, then check the indicators, and only then consider adjusting the model. Reopen the .csv file and check whether the columns have the correct variable names and numeric format, whether there are unusual characters, whether rows are misaligned, and whether reverse-coded variables have been coded correctly. A column entered as text instead of numbers can produce incorrect results or be excluded from the analysis.

Check the Original Data

Check missing values, identical responses across all indicators, and cases that suggest careless responding. If you defined cleaning rules in the methods chapter, apply them consistently. Do not delete a record simply because it reduces R².

Check the Indicators

If outer loading is low, check whether the indicator was reverse-coded correctly and whether its content genuinely belongs to the construct. You can rerun the model after removing one indicator, but save every run. Deleting an indicator should have a measurement or theoretical reason, rather than being based only on the wish to reach a particular number.

Check HTMT and VIF

A high HTMT value may indicate that two constructs measure content that is too similar. Re-read the definitions, the wording of the indicators, and the theoretical model. Check high VIF among predictors of the same construct. Do not merge two constructs only because HTMT does not meet the criterion if the merger has no conceptual basis.

Run Bootstrapping Again

After the model decisions have stabilised, run Bootstrapping with 5,000 subsamples according to the recommendation commonly cited in PLS-SEM (Hair et al., 2022). Save the settings, number of subsamples, confidence interval, and final results. If a hypothesis is not supported, report that result and discuss possible reasons within the limits of the data.

If the problem appears to be software-related, you can cross-check the analysis in R or another SEM tool. Do not run several programs and then select only the table that gives the most convenient result. SmartPLS has an accessible interface for students who are new to the method. R is flexible and free, but you must manage code, packages, and reproducibility. AMOS is suited to CB-SEM and uses a different fit-assessment logic. Paying for a SmartPLS service may save operating time, but the main risk is that you will not understand the output when the committee asks about it during the defence.

Distinguishing SmartPLS from Commonly Confused Indicators

SmartPLS is software, whereas PLS-SEM is a method. You can say that you used SmartPLS to perform a PLS-SEM analysis. Calling SmartPLS an indicator makes the sentence inaccurate.

Outer loading is the relationship between an indicator and a construct in the measurement model. Path coefficient is the coefficient for a relationship between constructs in the structural model. Both values can be positive or negative, but they answer different questions.

AVE assesses convergent validity at the construct level. HTMT assesses discriminant validity between constructs. R² assesses the proportion of variance in the dependent variable explained by the predictors. Therefore, a high R² cannot replace HTMT, and an acceptable AVE does not prove that a structural hypothesis is statistically significant.

P Values and T Statistics in the Bootstrapping table are used to assess the statistical significance of a path. They are not measures of effect size. A small coefficient can have a small p-value when the standard deviation is low, while a large coefficient in a small sample may not reach statistical significance.

SmartPLS also differs from a SEM model in how you assess the model. When reading PLS-SEM, focus on indicators appropriate to PLS, such as outer loading, CR, AVE, HTMT, VIF, R², f², and Q². Do not place CFI, TLI, and RMSEA in the same criteria table simply because you have seen them in CB-SEM literature.

Common Errors

The first error is importing .xlsx data and expecting SmartPLS to read it unchanged. Export the data table to .csv, keep the first row as the variable names, and check the delimiter, Vietnamese characters, and blank cells before importing.

The second error is using inconsistent variable names between the data table and the model. If the column is named PU_1 but the model calls it PU1, check the mapping. Construct and indicator names should be clear enough for you to trace the output back to the questionnaire.

The third error is going straight to Path Coefficients and skipping the measurement model. When a construct has low AVE or high HTMT, the path conclusions may be questioned. Save the outer loading, reliability, convergent validity, and discriminant validity tables before interpreting the hypotheses.

The fourth error is writing p = 0.000. A safer format is p < 0.001 when the software table rounds to three decimal places. Also report the path coefficient, t-value, or confidence interval so the reader has enough information.

The fifth error is deleting indicators repeatedly until every statistic looks acceptable. This changes the content of the scale and may leave the model inconsistent with its original theoretical foundation. Save each data version, record the reason for every adjustment, and discuss the decision with your supervisor when removing an indicator affects the scale structure.

Frequently asked questions

Is SmartPLS the same as PLS-SEM?

SmartPLS is software, while PLS-SEM is an analytical method. You use SmartPLS to build the model, run PLS-SEM, and check the measurement and structural models. In your thesis, state the software name and the method name clearly in the research methods section.

What counts as acceptable in SmartPLS?

There is no single threshold for all SmartPLS results. Check each group of indicators, including outer loading, CR, AVE, HTMT, VIF, R², f², and Q². The thresholds in the criteria table are reference points. Your final decision must also consider the scale content and theoretical model.

What should I do if SmartPLS gives an import error?

Check whether the file is .csv, whether the first row contains the variable names, and whether numeric columns contain text or special characters. Also check the number of rows, blank cells, and the coding of reverse-coded variables. After correcting a copy of the data, import it again and compare the number of observations with the original file.

Should I delete an outer loading below 0.7 immediately?

Do not delete it immediately. Check the construct's reliability and AVE before and after removing the indicator, and reread the indicator's content. If the indicator has important theoretical meaning, retaining it may be more defensible than deleting it to reach a numerical threshold. Record the reason for the decision so you can explain it later.

Are SmartPLS 3 and SmartPLS 4 different?

The two versions may have different interfaces and place some menus in different locations, but the main process still revolves around the model, PLS-SEM Algorithm, Bootstrapping, and the result tables. When writing procedural notes, state which version you are using. If you use SmartPLS 4, check the button names on the actual screen instead of relying on screenshots from an older version.

Open your data file, check the variable names, and create a table recording each SmartPLS indicator before writing Chapter 4. If you need to run the analysis on your own .sav or .csv file and read the output through the same process, see M4 data analysis.