Your supervisor said “use structural equation modeling”, and the advice then split: one colleague insists AMOS is the standard, another says SmartPLS is faster and works with your 160 responses. Who is right? Both, for different studies. The AMOS vs SmartPLS choice is really a choice between two estimation methods, covariance-based SEM (CB-SEM) and partial least squares SEM (PLS-SEM), and that choice follows from four features of your thesis: what the model is for, how large and how normal your sample is, whether your constructs are reflective or formative, and what your examiners expect to see reported.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.
This AMOS vs SmartPLS comparison sets out the two methods, a decision table, the sample-size rules, the reporting each program requires, a worked example of two theses that reached different decisions, and the licensing facts we could verify on the vendors’ pages.
What is the difference between AMOS and SmartPLS? IBM SPSS Amos is a program for covariance-based structural equation modeling. It estimates a model, usually by maximum likelihood, so that the model-implied covariance matrix reproduces the observed one as closely as possible, and it judges the result with global fit indices. SmartPLS is the most widely used program for partial least squares SEM, a variance-based method that maximizes the explained variance of the dependent constructs and tests path coefficients by bootstrapping; SmartPLS 4 also includes CB-SEM, regression and factor analysis modules. Settling AMOS vs SmartPLS is therefore mostly choosing a method.
Why is AMOS vs SmartPLS a methods decision before a software decision?
CB-SEM and PLS-SEM answer different questions. CB-SEM asks whether a theoretically specified model is consistent with the data: the chi-square statistic and the fit indices describe how well the implied covariance matrix reproduces the observed one, and the method is at its best when a mature theory is being confirmed or two competing models are being compared. PLS-SEM asks how much variance in the target constructs the model explains and how well it predicts new cases: the output is dominated by R², effect sizes and predictive measures, and there is no generally accepted global test of fit.
Hair, Risher, Sarstedt and Ringle (2019) frame the underlying choice in the AMOS vs SmartPLS decision as confirmation versus prediction and exploration; Kline (2016) treats CB-SEM as the default for theory testing with adequate samples. Decide the method first, justify it in the methodology chapter with at least two references, and only then open the program.
Which features of your thesis point to each program?
| Feature of your thesis | Points to AMOS (CB-SEM) | Points to SmartPLS (PLS-SEM) |
|---|---|---|
| Purpose | Confirming an established theory; comparing competing models | Predicting outcomes, exploring new relationships, extending a theory to a new context |
| Sample size | 200 or more; about 10 cases per free parameter (Kline, 2016) | Smaller samples justified by the inverse square root method or a power analysis (Hair et al., 2022) |
| Distribution | Approximately multivariate normal, or a bootstrap correction available | Skewed or ordinal data; the estimator makes no distributional assumption |
| Constructs | Reflective: items are interchangeable effects of the construct | Formative or composite constructs whose indicators define the construct; higher-order constructs |
| Model size | Small to medium, three to six constructs | Large models with many constructs, mediators and moderators |
| Evidence of model quality | χ², CFI, TLI, RMSEA, SRMR | R², f², Q², PLSpredict, HTMT; SRMR only as a rough indicator |
| Discipline conventions | Psychology, education, health sciences | Management, marketing, information systems, tourism |
If one row points one way and the others the opposite way, settle AMOS vs SmartPLS by the purpose row. The method should match the research question; sample size is a constraint to be justified, not the reason for the method.
📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.
How do the two methods treat sample size and distribution?
Maximum likelihood estimation in AMOS needs enough cases to produce a stable covariance matrix. The usual benchmark is 200 cases or more, with Kline’s (2016) ratio of about 10 cases per free parameter: a model with 38 free parameters calls for roughly 380 cases. Smaller samples bring non-convergence, inflated chi-square values and unstable estimates. Maximum likelihood also assumes multivariate normality; when the data are skewed, AMOS offers the Bollen-Stine bootstrap for the model test, and the asymptotically distribution-free estimator needs samples far larger than most theses have.
PLS-SEM makes no distributional assumption and relies on bootstrapping for inference, but it is not a license to analyze any sample. The old “10 times rule” (10 cases per arrow pointing at the most complex construct) is now widely criticized because it ignores effect size.
Kock and Hadaya’s (2018) inverse square root method gives the minimum sample as (2.486 ÷ |p|)² at a 5% significance level and 80% power, where p is the smallest path coefficient you expect to be significant: .30 requires 69 cases, .20 requires 155, and .15 requires 275. A thesis with 160 responses can therefore detect paths of about .20 and larger, and should say so.
Reflective or formative: what are your constructs made of?
A reflective construct causes its indicators: job satisfaction produces agreement with “I enjoy my work” and “I would recommend this employer”, the items are interchangeable, and dropping one does not change the construct. Internal consistency measures (alpha, composite reliability, AVE) apply. A formative construct is defined by its indicators: “digital readiness” built from infrastructure, skills, culture and leadership is the sum of those parts, the indicators need not correlate, and dropping one changes the meaning. Formative constructs are evaluated by indicator weights, their significance and collinearity (VIF below 5, preferably below 3), not by alpha.
CB-SEM can model formative constructs, but identification requires additional reflective indicators or outcomes, and the specification is a frequent source of errors. PLS-SEM estimates composites directly, so if a central construct in your model is formative, the AMOS vs SmartPLS question usually resolves in favour of SmartPLS, the simpler and better-documented route.
What does each side of the AMOS vs SmartPLS choice require you to report?
| Component | AMOS (CB-SEM) | SmartPLS (PLS-SEM) |
|---|---|---|
| Reflective measurement model | Standardized loadings of .50 or more (ideally .70), AVE of .50 or more, composite reliability of .70 or more, discriminant validity by Fornell-Larcker and HTMT | Outer loadings of .708 or more, composite reliability between .70 and .95 with alpha and rho_A, AVE of .50 or more, HTMT below .85 (or .90 for conceptually close constructs) with bootstrap confidence intervals (Henseler, Ringle & Sarstedt, 2015) |
| Formative constructs | Weights and their significance; identification strategy | Outer weights with bootstrap significance, VIF below 5 (preferably below 3), redundancy analysis |
| Model quality | χ² with df and p, χ²/df, CFI and TLI of .90 or more (.95 for good fit), RMSEA of .08 or less (.06 for good fit) with its 90% CI, SRMR of .08 or less (Hu & Bentler, 1999) | No global fit test; SRMR below .08 reported as approximate only; R² and adjusted R², f² (.02, .15, .35 for small, medium, large), Q² above zero, PLSpredict |
| Structural model | Standardized paths with p values, R² per endogenous construct, indirect effects with bias-corrected bootstrap CIs from 2,000 to 5,000 resamples | Path coefficients with bootstrap t and p values and percentile CIs from 10,000 subsamples, specific indirect effects, multigroup analysis where relevant |
| Sample justification | Cases per parameter; normality check | Inverse square root method or power analysis |
Hair et al. (2019) provide a reporting checklist for PLS-SEM that examiners in business schools increasingly expect to see followed and cited.
Worked example: two theses, one decision each
Thesis A (education) tests an established model in which transformational leadership predicts job satisfaction and commitment through trust, with four reflective constructs measured by 16 items among 420 teachers. Skewness and kurtosis are within ±1, the purpose is confirmatory, and the committee expects fit indices.
Decision: AMOS. The CFA returns χ²(98) = 212.4, p < .001, χ²/df = 2.17, CFI = .96, TLI = .95, RMSEA = .053, 90% CI [.044, .062], SRMR = .041; loadings range from .62 to .88, AVE from .51 to .68 and composite reliability from .81 to .92. In the structural model, leadership predicts satisfaction, β = .46, p < .001, with R² = .38, and the indirect effect through trust is 0.12, 95% bias-corrected bootstrap CI [0.06, 0.19] from 5,000 resamples.
Thesis B (business) explores whether digital readiness, a formative construct with four indicators, predicts firm performance through two mediators and under one moderator, with 160 managers and clearly skewed data. The smallest path expected to matter is .20, so the inverse square root method sets the minimum at 155 cases, which the sample meets.
Decision: SmartPLS. Outer weights range from .21 to .38 and are all significant, with VIF between 1.4 and 2.3; the reflective outcome constructs show loadings of .72 or more and HTMT values of .78 or less. The key path is β = .32, t = 4.21, p < .001, 95% CI [.17, .46], with R² = .41 for performance, f² = .18 and Q² = .27 from 10,000 bootstrap subsamples.
Neither student is the better analyst. The AMOS vs SmartPLS decisions differ because the purpose, the constructs and the sample differ, and the justification paragraph in each methodology chapter says exactly that.
What are the verified licensing and platform facts?
The IBM product page for SPSS Amos (accessed October 2026) describes it as a standalone application available on Windows, included in SPSS Statistics Premium and purchasable with the Base, Standard and Professional editions, with a free trial; it shows no prices. The SmartPLS website (accessed October 2026) presents SmartPLS 4, advertises PLS-SEM, CB-SEM, regression and factor analysis, and offers a free 30-day trial; licence types and prices are on its pricing page and change over time, so we do not reproduce them.
Two practical consequences for the AMOS vs SmartPLS decision: Mac users need a Windows machine or virtual machine for AMOS, and most universities already hold an SPSS licence that may or may not include Amos, so ask your IT service before paying.
Free alternatives that examiners accept when the package and version are cited: lavaan in R (also inside JASP) for CB-SEM, and seminr or cSEM in R for PLS-SEM.
Which mistakes do examiners flag in SEM chapters?
- Choosing PLS-SEM only because the sample is small while the purpose is confirmatory, and not saying so.
- Reporting CFI and RMSEA from SmartPLS output, or presenting R² and Q² in AMOS as if they were fit indices.
- Assessing discriminant validity in PLS-SEM with Fornell-Larcker alone, when HTMT is the recommended criterion.
- Deleting many items to reach acceptable fit without a theoretical reason.
- Bootstrapping with 500 resamples; current practice is 5,000 for mediation tests in AMOS and 10,000 subsamples in SmartPLS.
- Reading a high R² in PLS-SEM as proof of the theory, or a non-significant chi-square in a small sample as proof of fit.
- Omitting the sample-size justification for the AMOS vs SmartPLS choice altogether.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
Can I run a confirmatory factor analysis in SmartPLS?
PLS-SEM evaluates the measurement model with loadings, AVE and HTMT, and consistent PLS (PLSc) approximates a common factor model, but it does not produce the chi-square-based CFA with global fit indices that instrument-validation studies expect. If your purpose is to validate a scale, use AMOS or lavaan.
My sample is 150 with five reflective constructs from an established theory. Which program?
This is borderline for AMOS. If the model has few free parameters and the distributions are acceptable, AMOS with a stated limitation is defensible; if the model is complex or the data skewed, PLS-SEM with an inverse square root justification is the safer choice. Agree the AMOS vs SmartPLS decision with your supervisor and document it.
Do examiners in the UK and Europe accept PLS-SEM?
Widely in business, management, marketing and information systems, where the Hair et al. (2019) guidelines are the reference standard. Psychology and education departments still lean toward CB-SEM. Whichever side of AMOS vs SmartPLS you choose, follow and cite the reporting guidelines for its method.
Should I prepare the data in SPSS before either program?
Yes. Cleaning, missing-value handling, outliers, descriptives and Cronbach’s alpha are done in SPSS; AMOS reads the .sav file directly, and SmartPLS imports a clean numeric data file with no text codes and an explicit missing-value marker.
Can I report model fit for a PLS-SEM model?
Only SRMR, described as an approximate indicator, with .08 as the commonly used bound. Do not report CFI, TLI or RMSEA from a PLS model. The quality evidence in PLS-SEM is R², f², Q² and PLSpredict.
Sources
- Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (3rd ed.). SAGE.
- Hair, J. F., Risher, J. J., Sarstedt, M., & Ringle, C. M. (2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2–24. https://doi.org/10.1108/EBR-11-2018-0203
- Kline, R. B. (2016). Principles and Practice of Structural Equation Modeling (4th ed.). Guilford.
- Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6(1), 1–55. https://doi.org/10.1080/10705519909540118
- Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115–135. https://doi.org/10.1007/s11747-014-0403-8
- Kock, N., & Hadaya, P. (2018). Minimum sample size estimation in PLS-SEM: The inverse square root and gamma-exponential methods. Information Systems Journal, 28(1), 227–261. https://doi.org/10.1111/isj.12131
- SmartPLS GmbH. SmartPLS 4. https://www.smartpls.com/ (accessed October 2026); IBM. IBM SPSS Amos. https://www.ibm.com/products/structural-equation-modeling-sem (accessed October 2026)
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.