What to Do When Asked Why You Chose PLS-SEM
The problem you are facing
You have run your model in SmartPLS and have the Outer Loadings, Reliability and Validity, Discriminant Validity, Path Coefficients, and Bootstrapping tables in front of you. During your presentation, your supervisor or the committee asks, “Why did you choose PLS-SEM instead of CB-SEM, SPSS regression, or AMOS?” If your only answer is “because many previous studies used SmartPLS” or “because my sample is small,” your justification will be weak.
This question does not usually mean that you selected the wrong software. The examiner wants to know whether your method followed from your research question and data characteristics, or whether you selected it out of habit. Go back to four points: the research objective, the structure of the model, the prediction requirement, the characteristics of the data, and the testing procedure you actually completed.
PLS-SEM is more appropriate when your study focuses on explaining the variance of dependent variables, developing or extending a model, working with several latent variables, and assessing the measurement model and structural model in the same overall procedure. That is a methodological argument. Saying that SmartPLS is easy to run is a technical reason and should not be your main justification.
If the question comes with doubts about sample size, review the guidance in what to do when the committee asks about sample size. Sample size does not automatically prove that PLS-SEM is correct, and it does not automatically make the choice wrong.
Common reasons the question comes up
There are four common groups of problems that lead to deeper questions about your PLS-SEM justification. Compare them with your proposal, questionnaire, and .csv file before changing your answer.
| Sign during questioning | Common reason | How to check your research records |
|---|---|---|
| You say “PLS-SEM works with a small sample” but do not explain the research objective | You selected the method from one sample-size threshold | Re-read the objective, hypotheses, and main dependent variable |
| The model has several latent variables, but the questionnaire does not explain the source of the scales | The measurement model was not prepared properly | Check the variable codes, scale sources, reflective indicators, and formative indicators |
| Almost every path in the results is significant | You may have modified the model repeatedly according to p-values | Open the history of each run and record every deleted variable and reason for deletion |
| You use CFI, TLI, and RMSEA to explain SmartPLS | You have mixed CB-SEM criteria into a PLS-SEM analysis | Recheck the methods chapter and results tables |
| The committee asks whether the data are real | You did not preserve the cleaning process and original data | Compare the survey file, .csv file, coding, and processing log |
| The model changed after you viewed the results | You may have adjusted the model to fit the data without reporting it | Compare the initial hypothesised model with the reported model |
Another common problem is using the statement “PLS-SEM does not require normally distributed data” as an absolute answer. That sentence is not enough to defend your choice. You still need to check coding, missing values, duplicate responses, outliers, indicator validity, and the match between each scale and its construct.
If your supervisor asks you to change the model, the issue may concern the theoretical basis rather than SmartPLS. You can review what to do when your supervisor asks you to change the model, then separate theoretical changes from technical changes.
How to handle the issue in priority order
1. Turn your reason for choosing PLS-SEM into a structured argument
Organise your answer in this order: what the study aims to do, what the model looks like, what characteristics the data have, and what PLS-SEM allows you to test. For example, if your objective is to assess factors affecting intention to use a service, you can explain that the model contains several latent variables measured by multiple indicators and that you need to assess the measurement model and structural relationships within one procedure.
You may choose PLS-SEM when the study emphasises explaining and predicting the dependent variable, the model is exploratory or extends an existing model, and the data do not meet conditions commonly required in CB-SEM particularly well. You still need to connect the reason to your own study. Do not turn it into a claim that PLS-SEM is always better than AMOS or regression.
A safer answer is: “I chose PLS-SEM because the objective of my study is to explain the effects of latent variables on the outcome variable. The model contains several constructs measured by indicators, and the procedure allows me to assess the measurement model first and then the structural model. I checked the quality of the scales before interpreting the paths.”
2. Distinguish PLS-SEM from CB-SEM and SPSS regression
You do not need to memorise a long comparison table. You need to understand which option answers your research question better.
SPSS regression is suitable when the variables entered into the model have already been combined into observed variables or mean scores. PLS-SEM is suitable when you want to model latent variables and their indicators. AMOS and CB-SEM are commonly used when the focus is testing the fit of an established theoretical model. PLS-SEM places strong emphasis on explained variance, prediction, and the assessment of relationships in the structural model.
Do not say that PLS-SEM is suitable for every small-sample study. The fact that the software is free, has an easy interface, or runs quickly is also not a primary academic reason. Software cost is a practical consideration, but it cannot replace a methodological argument.
3. Recheck the measurement model before your defence
In SmartPLS 4, check the variable names, arrow directions, measurement type, and the relationship between each construct and its indicator. For reflective scales, you usually review Outer Loadings, Cronbach's Alpha, rho_A, Composite Reliability, and AVE. Formative scales require a different assessment, so do not apply every reflective-measurement criterion mechanically.
The commonly reported reference thresholds are below. Each number needs a source in the methods chapter, and you should report the actual value from your study as well.
| Measure | Reference level | Canonical source |
|---|---|---|
| Outer loading | 0.70 or above | (Chin, 1998) |
| Composite Reliability | 0.70 or above | (Fornell and Larcker, 1981) |
| AVE | 0.50 or above | (Fornell and Larcker, 1981) |
| HTMT | Below 0.85, or below 0.90 for closely related concepts | (Henseler et al., 2015) |
| VIF | Below 5 | (Hair et al., 2019) |
| R² | 0.75 substantial, 0.50 moderate, 0.25 weak | (Hair et al., 2011) |
| f² | 0.02 small, 0.15 medium, 0.35 large | (Cohen, 1988) |
| Q² | Greater than 0 | (Stone, 1974) |
| Bootstrap | 5,000 resamples is commonly used | (Hair et al., 2022) |
These thresholds are interpretation benchmarks, not permission to delete indicators until the table looks better. If you delete an indicator, consider its loading, conceptual content, reliability, and theoretical reason together. An indicator with a low loading may still represent important content and should not be removed only to increase AVE.
4. Preserve the data and every run so you can answer doubts about the results
When someone questions whether the data are genuine, provide a process that can be checked. Keep the raw dataset, cleaned dataset, .csv file imported into SmartPLS, variable codebook, response-exclusion criteria, and screenshots or exported result files. Do not edit the original file directly and submit only the final results table.
You also need to explain how invalid responses were handled. For example, a row with exactly the same answer for every item, an extremely short completion time, or many blank indicators may be reviewed against criteria established in advance. Record the initial number of responses, the number removed, and the final number retained. The figures must match across the methods chapter, sample description, and data file.
If the results look unusually clean, check them yourself before the committee does. Confirm that the paths were not entered in the wrong direction, indicators were not copied, rows are not exact duplicates, reverse-coded scales were recoded correctly, and the SmartPLS file is using the correct final dataset.
5. Rerun bootstrapping and export the correct results table
In SmartPLS 4, run the model, check the measurement model, and then select Bootstrapping to assess the statistical significance of Path Coefficients, Outer Loadings, and the relationships you need to report. State the number of bootstrap subsamples, the confidence-interval method, and the basis for your conclusion. According to (Hair et al., 2022), 5,000 bootstrap subsamples is commonly used in the PLS-SEM procedure.
The table below is illustrative output, not the result of a real study. The column names follow the format you commonly see in SmartPLS output. In your thesis, replace the values with those from your own file.
| Relationship | Original Sample (O) | Sample Mean (M) | Standard Deviation (STDEV) | T Statistics | P Values |
|---|---|---|---|---|---|
| Service quality → Satisfaction | 0.382 | 0.385 | 0.071 | 5.380 | 0.000 |
| Satisfaction → Continuance intention | 0.447 | 0.449 | 0.064 | 6.984 | 0.000 |
| Perceived value → Continuance intention | 0.126 | 0.129 | 0.059 | 2.136 | 0.033 |
| Perceived cost → Continuance intention | -0.081 | -0.078 | 0.052 | 1.558 | 0.119 |
In this illustrative table, the final relationship has P Values of 0.119, so there is insufficient evidence to conclude that an effect exists at the 5 percent significance level. Do not turn a non-significant result into an accepted hypothesis simply because the coefficient has the expected sign.
6. Prepare a short answer for direct questioning
Your answer should take about 30 to 60 seconds. Then stop and let the examiner ask the next question. One version you can adapt is: “My study uses several latent variables measured by indicators, and its main objective is to explain and predict the outcome variable. I therefore chose PLS-SEM to assess the measurement model first and the structural model afterward. I did not choose it only because of sample size. The decision was based on the research objective, model structure, and testing procedure. I preserved the raw data, cleaning criteria, and analysis steps so that the results can be checked.”
If the examiner asks why you did not use AMOS, answer within the scope of the study: “AMOS or CB-SEM could also be considered when the main focus is testing the fit of an established theoretical model and the data meet the relevant conditions. My study focuses on explanation and prediction, so PLS-SEM is more consistent with the stated objective.”
When you need to collect more data
Software cannot repair an unrepresentative sample, fabricated responses, a questionnaire sent to the wrong population, or a high proportion of missing data. Consider collecting more data when the number of valid observations is insufficient for the proposed model, important respondent groups are missing, or widespread collection errors caused you to remove many responses.
Justify sample size through the model and method, rather than saying only that “SmartPLS can run.” If the study includes several comparison groups, a mediator, or a moderator, consider the required sample size for each analysis. When the committee asks about SPSS sample size, the common mistake when the committee asks about SPSS sample size can help you distinguish a calculation error from a reporting error.
Collecting more data is not the default solution for a large p-value. If a hypothesis is not significant, report that result and discuss it appropriately. You can also review what to do when your result contradicts the hypothesis instead of changing the data to make the result look better.
How to explain the analysis to your supervisor
Bring a short set of documents to the meeting: the initial model, variable codebook, reason for choosing PLS-SEM, data-cleaning procedure, measurement model assessment table, structural model assessment table, and log of every run. With those documents in front of you, the discussion moves from “I think this software is easy to use” to “these are the decision and the evidence for each step.”
How to write this in your thesis: “The study selected PLS-SEM because the model contains latent variables measured by multiple indicators and the research objective focuses on explaining [dependent variable] through the relationships between [independent variables]. The analysis consisted of assessing the measurement model and the structural model. The assessment criteria were selected based on [methodological source], while the data were checked for [missing data, invalid responses, and variable coding] before analysis.”
Replace the bracketed sections with real information. Do not write “the data were suitable” unless you have a table showing the checks. Do not write that “the model met all criteria” when one measure remains below the threshold. The committee often asks the next question precisely where the wording makes an overly broad claim.
How to prevent the problem during study design
State in the proposal why you are using PLS-SEM, which variables are constructs, which indicators belong to each construct, which hypotheses need testing, and which assessment criteria you expect to use. If you change the model after examining the data, record that it was a post-analysis adjustment. Do not present it as though it had been specified from the beginning.
The questionnaire needs the scale source, translation and adaptation procedure, Likert scale, screening items, and response-exclusion criteria. Keep variable names consistent from the questionnaire to Excel, .csv, and SmartPLS. A code such as SAT1, SAT2, SAT3 is easier to check than long column names containing diacritics or names that change between files.
You should also keep a run log containing the date, data version, change made, deleted indicator, reason, and approving person. If your topic overlaps with an earlier study or the model is too similar to previous research, deal with it early using what to do when your topic overlaps with another study, rather than changing the model hastily after data collection.
Common mistakes when fixing the analysis
The first mistake is memorising the answer that PLS-SEM is suitable for small samples. Sample size is only one part of the argument. The second mistake is reporting every index found in SEM materials, including CFI, TLI, and RMSEA, in a SmartPLS analysis. These indices belong to CB-SEM assessment under (Hu and Bentler, 1999), so keep the method's scope clear.
The third mistake is deleting indicators repeatedly until Alpha, AVE, and the p-values all look good. This can remove important content from the construct and produce a model that is difficult to explain. Every deletion needs a reason, a record, and consideration alongside the theory.
The fourth mistake is saying that the data were “collected completely at random” when you did not use a probability-sampling design. Describe the actual approach, target respondents, questionnaire distribution channel, and selection criteria. The final mistake is submitting result tables that do not match the data file. One variable with a different number of observations across tables can increase doubts about reliability.
Frequently asked questions
What should I do if I am asked why I chose PLS-SEM and my sample is not large?
Present sample size as a condition that must be checked, rather than your only reason. State the prediction objective, number of latent variables, model structure, number of paths leading to the dependent variable, and assessment procedure. Then provide the valid sample-size table and explain how you excluded responses.
Should I say that PLS-SEM is easier to run than AMOS?
You can say that its interface or operating procedure fits your skills and available time, but that is only a practical reason. The main reason must relate to the research objective, model characteristics, measurement model, structural model, and type of result that needs to be interpreted.
What should I do if someone doubts that my data are genuine?
Open the raw data, cleaned data, codebook, and processing log. You need to show the initial number of responses, number excluded, exclusion reasons, final number retained, and agreement between the data file and results tables.
If you cannot explain the source of a data row or have edited the original file repeatedly, state that limitation clearly and review the data before submission. Never create additional rows to make the results look better.
Does PLS-SEM require a normality test?
You still need to check data quality and problems that may affect the results. Saying that “PLS-SEM does not require normally distributed data” does not replace checks for missing values, outliers, reverse coding, duplicate responses, and scale appropriateness.
If a hypothesis is not significant, do I need to switch to AMOS?
No. A non-significant path can be a valid empirical result. Consider changing the method only when the original objective, model, or testing requirement is genuinely more consistent with another method, rather than because you want a smaller p-value.
Can I use both SPSS and SmartPLS in the same thesis?
Yes, when each tool serves a clearly defined purpose. SPSS can be used for data cleaning, sample description, or preliminary checks, while SmartPLS can be used for the measurement model and structural model. State the input data, processing order, and reason for keeping the assessment criteria of the two methods separate.
Open your data file now and prepare a one-page justification covering the research objective, reason for choosing PLS-SEM, valid sample size, cleaning procedure, and main SmartPLS tables. Check every figure against the file before your meeting. If you need to run the analysis on the actual .sav or .csv file and preserve the workflow, see M4 data analysis by DoThesis.