What Is AVE in SPSS and How Do You Read the Results?
What Is AVE in SPSS
AVE stands for Average Variance Extracted. It shows how much of the variance in the observed variables used to measure a latent variable is explained by that latent variable. For example, if the AVE of the Satisfaction construct is 0,62, the construct explains an average of 62% of the variance in its observed variables.
In quantitative research, AVE is used to assess the convergent validity of a scale. The formula is usually written as:
AVE = sum of squared outer loadings / number of indicators
Here, outer loading is the loading of each indicator on the latent variable. With an SPSS-only workflow, you usually will not see a table called AVE in Cronbach's Alpha or EFA output. SPSS provides results such as Rotated Component Matrix, Component Matrix, or Factor Matrix, from which you can use factor loadings to calculate an approximate value. AVE is usually reported directly by SmartPLS, AMOS, or another SEM software package.
That is why the phrase “AVE in SPSS” can be confusing. If you are running Cronbach's Alpha and EFA in SPSS, you first assess the foundation of the scale. When you need to assess a measurement model containing latent variables, you usually move to CFA in AMOS or PLS-SEM in SmartPLS. You can also read What is convergent validity to connect AVE with the related measures.
Why AVE Matters in Quantitative Research
AVE answers a specific question: do the indicators in the same scale actually converge on the same construct? When the indicators have high loadings on the same factor, the variance explained by that factor is larger and AVE tends to be higher. When the loadings are low, AVE decreases.
You should not read AVE separately from the other measures. A scale is usually assessed through two groups of indicators:
- Reliability: Cronbach's Alpha, Composite Reliability, or rho_A.
- Validity: AVE for convergent validity, and HTMT or the Fornell-Larcker criterion for discriminant validity.
AVE only addresses convergent validity. It does not tell you whether two different constructs overlap. For example, a Service Quality scale can have an acceptable AVE while still failing discriminant validity against the Satisfaction scale. In that situation, you need to check HTMT or the Fornell-Larcker criterion. This is directly related to What is construct validity.
In the analysis workflow, AVE is usually read after you have identified the scale structure and checked the loadings. With PLS-SEM, the sensible order is to assess the measurement model first and then move to the structural model. This two-step approach is presented in (Anderson and Gerbing, 1988). If AVE is below the usual threshold, interpreting path coefficients, R², or mediation effects lacks a sound measurement foundation.
You also need to distinguish AVE from total variance explained in EFA. Total variance explained appears in the Total Variance Explained table and describes the variance explained by all retained factors. AVE describes the variance explained by each latent variable. Both measures contain the word “variance”, but they are not interchangeable.
What AVE Value Is Acceptable in SPSS?
The commonly used threshold is AVE of 0,50 or above. When AVE reaches 0,50, the latent variable explains at least half of the average variance in its indicators. This criterion is based on (Fornell and Larcker, 1981). For PLS-SEM, you can also refer to the current assessment procedure in (Hair et al., 2022).
The table below summarises the thresholds that commonly appear when you read the output. Each row includes a source so that you can explain the criterion in your thesis.
| Measure or criterion | Common threshold | Meaning when reading the output | Source |
|---|---|---|---|
| AVE | 0,50 or above | The latent variable meets the commonly required level of convergent validity | (Fornell and Larcker, 1981) |
| Composite Reliability | 0,70 or above | The scale meets the composite reliability criterion | (Fornell and Larcker, 1981) |
| Outer loading in PLS-SEM | 0,70 or above | The indicator makes a strong contribution to the latent variable | (Chin, 1998) |
| HTMT | Below 0,85, or below 0,90 for closely related constructs | Discriminant validity meets the HTMT criterion | (Henseler et al., 2015) |
| VIF | Below 5 | There is no serious sign of multicollinearity among indicators or predictor variables | (Hair et al., 2019) |
An AVE of 0,50 does not mean that every indicator must have exactly the same loading. You need to consider the full model, the loadings, reliability, and the theoretical rationale when deciding whether to remove an indicator. A low-loading indicator can pull AVE down, but removing an indicator only to make the number look better can remove important content from the construct.
For EFA in SPSS, Hair et al. are commonly used as a reference for factor loadings of 0,50 or above and total variance explained of 50% or above (Hair et al., 2010). These are EFA criteria, not another name for AVE. When you write Chapter 4, use the exact table name and measure so that the committee does not have to ask you to clarify what you reported.
How to Read AVE in the Output
SPSS does not automatically create an AVE table in Analyze > Dimension Reduction > Factor. Therefore, if you open a .sav file, run EFA, and search for “AVE” in the Output Viewer, you may not find it. Identify the software and the type of model before you interpret the output.
If you run EFA in SPSS, check the KMO and Bartlett's Test, Total Variance Explained, and Rotated Component Matrix tables. A KMO of 0,50 or above is used as a benchmark for data adequacy according to (Kaiser, 1974). The eigenvalue-above-1 rule is associated with (Kaiser, 1960). These tables help you assess the factor structure, but they are not an AVE table.
If you run PLS-SEM in SmartPLS, open the measurement model results and check the Outer Loadings, Construct Reliability and Validity, and Discriminant Validity tables. In Construct Reliability and Validity, you will usually see the columns Cronbach's Alpha, rho_A, Composite Reliability, and Average Variance Extracted (AVE).
The table below is illustrative output, not the result of a real study. The column names follow the way the software commonly displays them so that you can identify where to look and how to read the values.
| Construct | Cronbach's Alpha | rho_A | Composite Reliability | Average Variance Extracted (AVE) | Illustrative conclusion |
|---|---|---|---|---|---|
| Service Quality | 0,842 | 0,851 | 0,893 | 0,676 | Convergent validity achieved |
| Trust | 0,781 | 0,790 | 0,861 | 0,608 | Convergent validity achieved |
| Satisfaction | 0,704 | 0,719 | 0,819 | 0,532 | Achieved, check loadings further |
| Continuance Intention | 0,681 | 0,695 | 0,806 | 0,512 | Achieved, close to the threshold |
In this example, the AVE of Satisfaction is 0,532. It is higher than 0,50, so it meets the commonly used convergent validity criterion. You still need to check the loading of each indicator and discriminant validity. The AVE of Continuance Intention is 0,512, only slightly above the threshold, so check the quality of each indicator rather than making a broad claim from AVE alone.
If you calculate AVE manually from the loadings, suppose a construct has three loadings of 0,80, 0,75, and 0,70. The illustrative calculation is (0,80² + 0,75² + 0,70²) / 3, which gives approximately 0,57. This only demonstrates the formula. When SmartPLS already provides AVE, use the software result and save the version of the model that produced it.
What to Do When AVE Does Not Meet the Threshold
When AVE is below 0,50, do not delete an indicator immediately. Work through the checks in order so that you preserve both statistical quality and the content meaning of the scale.
Check the Outer Loading or Factor Loading
Open the Outer Loadings table in SmartPLS or the Rotated Component Matrix in SPSS. Look for indicators with low loadings, negative loadings caused by reverse coding, or high loadings on more than one factor. If an indicator was entered incorrectly, reverse-coded incorrectly, or assigned to the wrong group, correct the data and run the analysis again using the proper procedure.
Check Whether an Indicator Needs Reverse Coding
An indicator with a negative meaning needs to be reverse-coded before you run the reliability analysis. For example, with a Likert scale from 1 to 5, the new value of a reverse-coded indicator can be calculated as 6 - old value. Check the coding scheme in the questionnaire. Do not reverse-code an indicator only because its loading is low.
Consider Removing a Low-Loading Indicator
If an indicator has a clearly low loading, you can run a test model after removing it, but record the results before and after removal. The decision should consider the construct content, the number of remaining indicators, and the methodological rationale. For an exploratory scale, an Alpha of 0,60 can be used as a reference according to (Hair et al., 2010), but this threshold does not automatically turn a weak scale into a good one.
Run the Model Again and Check Every Measure
After removing one indicator, AVE may increase while Composite Reliability, HTMT, or content validity changes. Check Outer Loadings, Construct Reliability and Validity, and Discriminant Validity again. Do not stop as soon as AVE moves above 0,50.
Return to the Scale and Data When the Problem Is Systematic
If several constructs all have low AVE, the cause may be the questionnaire, the translation of the indicators, sample size, missing data, or incorrect grouping of variables. Software cannot repair a poorly designed scale. If low AVE reflects the actual data and the indicators do not converge, report it honestly and discuss it with your supervisor instead of running repeated rounds until you obtain the number you want.
Distinguishing AVE from Similar Measures
AVE and Cronbach's Alpha are both related to scales, but they answer different questions. Alpha assesses whether the indicators have internal consistency. AVE assesses how much variance in the indicators is explained by the latent variable. A high Alpha does not guarantee a high AVE.
AVE is also different from Composite Reliability. Composite Reliability assesses overall reliability, while AVE assesses convergent validity. A construct can meet the Composite Reliability criterion while still having an AVE below the threshold, and it then needs further assessment. You can also read What is the Fornell-Larcker criterion, since this criterion commonly appears alongside AVE when you assess discriminant validity.
AVE differs from Total Variance Explained in EFA. Total Variance Explained shows the total variance explained by the retained factors across the full set of variables. AVE is calculated for each latent variable in the measurement model. In Chapter 4, place the two results in their appropriate sections and avoid writing “AVE achieved 50%” based on the EFA table.
AVE is also not cross-loading. Cross-loading shows how an indicator loads on other factors. If you see an indicator loading highly on two factors, check What is cross-loading. AVE can meet the threshold while the model's discriminant validity remains unconvincing.
Common Errors
The most common error is looking for AVE in the SPSS output from Cronbach's Alpha. The Reliability Statistics and Item-Total Statistics tables do not provide AVE. Distinguish reliability analysis, EFA, CFA, and PLS-SEM before you choose the table to report.
The second error is using an AVE threshold without giving the source. In your thesis, you can write: “The AVE values of all latent variables were greater than 0,50, meeting the convergent validity criterion according to (Fornell and Larcker, 1981).” If a construct is close to the threshold, report its exact value instead of writing that every construct is very good.
The third error is confusing AVE with total variance explained. EFA requires total variance explained of 50% or above according to (Hair et al., 2010), while AVE is expected to be 0,50 or above according to (Fornell and Larcker, 1981). These criteria must be used in the correct context.
The fourth error is deleting several indicators at once without saving each version of the output. Keep the original data file, record the name of each removed indicator, the reason for removal, and the result after each round. If you use SmartPLS, save the model by version so that you can explain why AVE changed when your supervisor asks.
Frequently asked questions
Do you need to run AVE separately in SPSS?
If you only run Cronbach's Alpha and EFA in SPSS, SPSS does not directly create an AVE measure. AVE is usually calculated in a CFA or PLS-SEM measurement model. You can use EFA loadings to illustrate the formula, but you should not call total variance explained in EFA AVE.
What AVE value is acceptable?
An AVE of 0,50 or above is commonly considered to meet convergent validity according to (Fornell and Larcker, 1981). You still need to read it alongside outer loading, Composite Reliability, and discriminant validity. An acceptable AVE value alone is not enough to conclude that the entire measurement model is sound.
Does AVE below 0,5 mean that you should delete an indicator immediately?
No. First check the loading, reverse coding errors, incorrect data entry, and the content of the indicator. Then consider removing an indicator with a low loading, run the model again, and report the reason. If several constructs have low AVE at the same time, the problem may be in the scale design or the data.
Can SPSS calculate AVE from the Rotated Component Matrix?
SPSS does not automatically display AVE in the Rotated Component Matrix. You can use the loadings for an illustrative calculation, but EFA factor loading and PLS-SEM outer loading belong to different procedures. Use the correct measure name and software name in your report.
Are AVE and convergent validity the same thing?
AVE is an important criterion for assessing convergent validity, but the two concepts are not identical. When you conclude that convergent validity has been established, you usually consider AVE together with outer loading and Composite Reliability. Your wording should follow the model and method you selected.
How to write this in your thesis
Open the data file again, identify whether you are running EFA in SPSS or PLS-SEM in SmartPLS, and then check the correct Construct Reliability and Validity table or the corresponding EFA tables before writing the results. If you need to run AVE and the full set of tests on your .sav or .csv file, you can use the M4 data analysis module.