AVE in SmartPLS: How to Run, Read, and Evaluate It

What is AVE in SmartPLS used for

AVE, short for Average Variance Extracted, is used to assess the convergent validity of a latent variable in the measurement model. It shows, on average, how much variance a construct explains in its associated indicators. You will usually find AVE in the Construct Reliability and Validity table after running the PLS Algorithm in SmartPLS.

When a construct has high AVE, its indicators tend to converge and reflect the same concept. AVE does not tell you whether one construct affects another. That question belongs to Path Coefficients, p-value, t-statistics, and the bootstrapping results.

The general AVE formula is the average of the squared outer loading values for the indicators belonging to a construct. AVE is therefore directly affected by outer loading. One indicator with a low loading can pull down the AVE of the entire construct, but you should not delete an indicator simply to make one number look better.

If this concept is new to you, you can also read What is AVE. To place AVE within the full PLS-SEM workflow, PLS-SEM will help you distinguish the measurement model from the structural model.

Preparing the model and data

Before running AVE in SmartPLS, check the two layers of your model. The first is the measurement model, which contains the constructs and the indicators measuring them. The second is the structural model, which contains the arrows between constructs and the research hypotheses. AVE belongs to the measurement model, so a path between two variables does not directly change the meaning of AVE.

Indicator names should match your questionnaire and data file. For example, TC1, TC2, and TC3 may belong to the TinCay construct, while HL1, HL2, and HL3 belong to HaiLong. If your names are inconsistent, matching the output to the questionnaire and Chapter 3 will take longer.

SmartPLS 4 usually imports data from a .csv file. Each row represents one respondent, and each column represents one indicator. The first row should contain variable names, not long question wording or special symbols. Data cells should contain numbers, such as values on a 1 to 5 Likert scale. Check missing values, reverse-coded variables, and unused columns before importing the file.

You also need to check whether the construct is modeled as reflective or formative. AVE, outer loading, Composite Reliability, and HTMT are mainly used when evaluating reflective constructs. For a formative construct, the assessment focuses on outer weights, statistical significance, collinearity, and theoretical fit. Do not apply the AVE threshold for a reflective construct to a formative construct.

Check that you are using the intended SmartPLS version. This article follows the menus in SmartPLS 4. SmartPLS 3 places some tables and menus differently, although the logic of the PLS Algorithm and bootstrapping is similar. If you need an overview of the software, read SmartPLS before following the steps below.

Steps for running AVE in SmartPLS 4

Step 1: Create a project and import the data file

Open SmartPLS 4, create a new project, and import the .csv file. Check the variable list after the software reads the data. If variable names are truncated, Vietnamese characters are corrupted, or a column is detected as text, return to the original file and correct it before building the model.

At this stage, do not replace missing values with an arbitrary number just to make SmartPLS accept the file. State your data-handling rule in the methods section. If the missing-data rate is high or one respondent left too many items blank, review your respondent-retention and exclusion criteria before running the model.

Step 2: Create the measurement model

Drag the constructs onto the SmartPLS 4 canvas, then drag the corresponding indicators onto each construct. Check the arrow direction between the construct and its indicators. In a reflective measurement model, the arrows usually run from the construct to the indicators. Name each construct consistently with the research model and results tables.

If a construct contains reverse-coded indicators, recode the data before running the analysis. For a 1 to 5 scale, for example, the new value of a reverse-coded variable may be calculated as 6 minus the old value, but you must follow the design of your questionnaire. SmartPLS does not automatically recognise a negatively worded item and reverse its score for you.

Step 3: Run the PLS Algorithm

In SmartPLS 4, select the model, click Calculate, choose PLS-SEM Algorithm, and first run it with the default settings. The preliminary output shows outer loadings on the diagram. Then open the detailed results to review the reliability and validity tables.

The path you need is the Quality Criteria results group, followed by the Construct Reliability and Validity table. This table usually contains the columns Cronbach's alpha, rho_A, Composite reliability (rho_C), and Average variance extracted (AVE).

Do not read AVE in isolation. Review outer loading, CR, and discriminant validity at the same time. A construct may meet the AVE threshold while one indicator has a very low loading that still needs to be examined in light of the content and theory.

Step 4: Run Bootstrapping when inferential testing is needed

AVE describes the measurement-model results from the PLS Algorithm. When you need to test the statistical significance of an outer loading or path coefficient, choose Calculate, then choose Bootstrapping in SmartPLS 4. Set the number of bootstrap subsamples according to your research procedure and record the settings you used.

A widely cited PLS-SEM guideline uses 5,000 bootstrap subsamples (Hair et al., 2022). Bootstrapping does not turn a low AVE into an acceptable AVE. It helps you assess the stability and statistical significance of the related estimates.

Reading the results table

After running the PLS Algorithm, open Quality Criteria > Construct Reliability and Validity. The Average variance extracted (AVE) column is where you read the AVE for each construct. Each row corresponds to a latent variable, not an indicator.

The table below is illustrative output, not the result of a real study. The column names follow the way SmartPLS commonly displays them so you can compare them with the file open on your screen.

For example, the TinCay row has an AVE of 0.636. You can state that the TinCay construct explains, on average, about 63.6% of the variance in the indicators belonging to it, according to the AVE calculation. This does not mean that TinCay increases loyalty by 63.6% or produces a dependent outcome.

For GiaTri, the AVE is 0.498, which is very close to the 0.5 benchmark. Check each indicator's outer loading, the wording of the item, and CR before deciding whether to delete an indicator. Rounding may cause 0.498 to appear as 0.50 in some tables, but use the original value when evaluating the result.

You also need to read AVE alongside discriminant validity. In SmartPLS 4, you can open Discriminant Validity, which includes HTMT and the Fornell-Larcker Criterion. HTMT below 0.85, or below 0.90 for constructs that are conceptually close, is a commonly used reference threshold (Henseler et al., 2015). The criterion that the square root of AVE should be greater than the correlations between constructs is the Fornell-Larcker criterion (Fornell and Larcker, 1981).

Evaluation thresholds

The threshold commonly used for AVE in a reflective construct is 0.5 or above. This means that the construct explains at least half of the average variance in its indicators. The criteria of CR at 0.7 or above and AVE at 0.5 or above are presented in (Fornell and Larcker, 1981). Some sources also use Bagozzi and Yi as a reference for assessing reliability and convergent validity (Bagozzi and Yi, 1988).

Do not turn the threshold table into an automatic deletion rule. Consider the number of indicators remaining, the content meaning, and the construct's fit with the model. The table below lists criteria from PLS-SEM and includes the sources you can use when explaining your decisions in the thesis.

CFI, TLI, RMSEA, and their corresponding fit-index thresholds belong to CB-SEM, which is commonly associated with AMOS or covariance-based SEM software. Do not place these indices in the AVE assessment for SmartPLS. To distinguish the two approaches, read What is SEM and What is Partial Least Squares.

A reasonable sequence is to check outer loading, Cronbach's Alpha or rho_A, Composite Reliability, and AVE, then assess discriminant validity with HTMT and Fornell-Larcker. Cronbach's Alpha at 0.7 or above is commonly accepted (Nunnally, 1978), while alpha from 0.6 may be considered for a new scale or exploratory research (Hair et al., 2010). These criteria are related, but each measure answers a different question.

How to write this in your thesis

In Chapter 4, report the measurement model first and the structural model afterward. Your first table can summarise Cronbach's Alpha, rho_A, Composite Reliability, and AVE for each construct. A following table can present outer loadings or the indicators removed, if any.

You can use the template below, then replace the placeholders with the actual values in your output:

How to write this in your thesis: “The results in Table [table number] show that the AVE values of the constructs range from [minimum value] to [maximum value]. The values for [construct name] are 0.5 or above, meeting the criterion for convergent validity proposed by Fornell and Larcker (1981). The Composite Reliability values of the constructs range from [value range], indicating that the composite reliability of the scales is at an acceptable level.”

If a construct has low AVE, report the value and your decision honestly. For example: “The AVE of [construct name] is [value], below the reference benchmark of 0.5. The research team reviewed the outer loading, indicator content, and theoretical basis before deciding whether to retain or remove [variable code].” Write that the scale meets the criteria only when the indicators and reasoning actually support that conclusion.

In the structural-model section, move to VIF, R², f², Q², path coefficients, and bootstrapping. AVE does not replace R² and does not tell you whether hypotheses H1 and H2 are supported. Keeping these two sections separate shows the committee that you understand the different questions answered by the measurement model and the structural model.

Common mistakes

Using AVE to conclude that a hypothesis is supported

AVE assesses the convergent validity of a construct. To conclude that one variable affects another, read Path Coefficients and the bootstrapping results, including the path coefficient, t-statistics, p-value, or confidence interval according to your analysis settings.

Deleting an indicator only because its loading is low

A low-loading indicator can reduce AVE, but deleting it changes the content of the scale. Review the loading, CR, AVE after deletion, theoretical meaning, and number of indicators remaining. Save each model version so you can explain the decision to your supervisor.

Mixing CB-SEM criteria into SmartPLS

Putting CFI, TLI, or RMSEA in a SmartPLS assessment table is a method-classification error. For PLS-SEM, focus on outer loading, CR, AVE, HTMT, VIF, R², f², Q², and path coefficients according to your research design.

Confusing AVE with outer loading

Outer loading belongs to an individual indicator, while AVE belongs to the construct as a whole. A construct with four indicators has one AVE value in the Construct Reliability and Validity table, while each indicator has its own loading in the outer-loadings results.

Running the model repeatedly without recording the reason

EFA or PLS-SEM rarely produces a clean table on the first run. You may need to check the data, indicators, and model. Every indicator removal should have a methodological reason and should be recorded, rather than running the model until the numbers cross a threshold and reporting only the final version.

Frequently asked questions

What AVE value is acceptable in SmartPLS?

An AVE of 0.5 or above is commonly used to assess the convergent validity of a reflective construct (Fornell and Larcker, 1981). You still need to review outer loading, Composite Reliability, and discriminant validity together. Do not reach a conclusion from the AVE cell alone.

What should I do if AVE is below 0.5?

First, check the outer loading of each indicator, reverse coding, data coding, and scale content. If you have a clear theoretical basis, you may consider removing a weak indicator and running the model again, but report the reason and reassess the construct after deletion.

If AVE is only slightly below 0.5, such as 0.498, do not round it up to create an acceptable result. Discuss the research context, CR, and number of remaining indicators with your supervisor before deciding.

What is the difference between AVE and Composite Reliability?

AVE reflects how much variance the construct explains in its indicators, so it relates to convergent validity. Composite Reliability reflects the consistency of the indicators within the same construct. A construct can meet the CR threshold while its AVE still requires separate review for convergent validity.

What should I do if SmartPLS 4 does not show the AVE table?

Check whether you have run PLS-SEM Algorithm, whether the model contains constructs and indicators, and whether the indicators are recognised as numeric variables in the .csv file. After running the algorithm, open Quality Criteria and find Construct Reliability and Validity. If the construct is formative, the relevant table may be different because AVE is not the primary criterion for that measurement type.

Do I need to run Bootstrapping to obtain AVE?

You can view AVE after running the PLS Algorithm. Bootstrapping is used when you need to test the statistical significance and stability of estimates such as outer loading or path coefficient. It does not replace the AVE assessment and does not automatically correct a construct with weak convergent validity.

Can I use AVE in SmartPLS to assess an SEM model?

You can use AVE for the measurement model in a PLS-SEM model, mainly for reflective constructs. AVE does not assess the entire quality of the structural model. You still need to review VIF, R², f², Q², path coefficient, and bootstrapping results according to your research question.

Open your SmartPLS 4 file, run PLS-SEM Algorithm, record the Construct Reliability and Validity table, and compare each construct using AVE, CR, outer loading, and HTMT instead of looking at one number alone. If you want to run the full procedure on a .sav or .csv file and turn the output into analysis tables, M4 analysis is the module for this step.