Overview
Exploratory factor analysis (EFA) asks what structure underlies a set of items when you are not yet sure; confirmatory factor analysis (CFA) tests whether a structure you specify in advance fits the data. Researchers often run EFA in SPSS on an established scale and then report it as validation, or run CFA on the same sample used to build the model. Both weaken the argument.
We help you choose the right approach for your situation — a new instrument, a translated or adapted scale, or a well-established measure used in a new population — and make the decisions that matter: how many factors to retain, which extraction and rotation to use, how to treat ordinal items, and what to do with weak or cross-loading items.
You receive the analysis with syntax, tables of loadings, reliability and validity statistics, and a record of every item-level decision, written so you can explain it in a methods chapter or viva.
Who it’s for and the problems it solves
This is for you if you are…
- Researchers developing, translating or adapting a questionnaire
- Master's and doctoral students who must validate scales before testing hypotheses
- Authors asked by reviewers for CFA, measurement invariance or reliability evidence
- Educational and health researchers validating assessment instruments
Common problems we help with
- SPSS suggests more factors than your theory predicts
- Several items load on two factors, or on none
- Your CFA fit indices (CFI, RMSEA, SRMR) fall short of common benchmarks
- You are unsure whether to use Cronbach's alpha or McDonald's omega
- Your scale was translated and you need to show it works the same way in the new population
What’s included — and what isn’t
Included
- Advice on EFA versus CFA (or both, on split or separate samples) for your design
- Data screening: item distributions, missing data, KMO and Bartlett's test for factorability
- Factor retention using parallel analysis, scree plots and interpretability, rather than eigenvalues above 1 alone
- Extraction and rotation choices (principal axis or ML; oblique rotations such as oblimin or promax where factors correlate)
- Polychoric correlations and appropriate estimators when items are ordinal
- CFA model specification, estimation (ML, MLR or WLSMV) and fit assessment with chi-square, CFI, TLI, RMSEA and SRMR
- Reliability: Cronbach's alpha and McDonald's omega, with item–total statistics
- Convergent and discriminant validity evidence (AVE, composite reliability, HTMT)
- Measurement invariance testing across groups (configural, metric, scalar) when comparisons are planned
- Annotated syntax in SPSS, R (psych, lavaan), Mplus, AMOS or JASP/jamovi
Not included
- Deleting items or adding error covariances solely to reach a fit cut-off without theoretical justification
- Presenting EFA results on an established scale as confirmation of its structure
- Writing the psychometric sections of your thesis for you — we explain the evidence so you can present it
- Promising that any particular fit index will reach a given threshold
What you receive
- Measurement plan
- Whether EFA, CFA or both are used, on which sample, with which estimator and retention criteria, agreed before analysis.
- Factor solution tables
- Pattern and structure loadings (EFA) or standardized loadings and fit indices (CFA), with communalities and factor correlations.
- Reliability and validity summary
- Alpha, omega, composite reliability, AVE and discriminant validity statistics with plain notes on any construct that falls short.
- Item decision log
- Each item that was retained, revised or dropped, the evidence, and the effect on content coverage.
- Syntax and output
- Commented code and cleaned output files so the analysis can be reproduced by you or an examiner.
How it works
Understand the instrument
We review the scale's origin, intended dimensions, item wording and response format, and how it was administered.
Choose the approach
We recommend EFA, CFA or a sequence of both, the estimator and sample split, and confirm it with you.
Screen and analyze
We check item distributions and factorability, then extract or fit the model and assess the solution.
Refine transparently
Any item problems are discussed with you; changes are made only with a documented rationale.
Report and explain
You receive tables, figures and syntax, followed by a walkthrough of the results.
Decisions that shape a factor analysis
- Parallel analysis
- Compares observed eigenvalues with those from random data; more accurate than the eigenvalue-greater-than-one rule, which tends to over-extract factors.
- Oblique rotation
- Oblimin or promax allow factors to correlate, which is realistic for most psychological and social constructs; varimax assumes they are unrelated.
- Ordinal indicators
- Items with few response options are analyzed with polychoric correlations and WLSMV or similar estimators rather than treated as continuous.
- CFA fit assessment
- Global fit (chi-square, CFI, TLI, RMSEA with its confidence interval, SRMR) read together with local fit (residuals, modification indices) and theory.
- Reliability coefficients
- McDonald's omega is reported alongside or instead of Cronbach's alpha when loadings differ across items, since alpha assumes they are equal.
- Measurement invariance
- Multi-group CFA with increasingly constrained models; changes in CFI and RMSEA guide whether groups can be compared on the same scale.
When to explore and when to confirm
Use EFA when the structure is genuinely uncertain: a new instrument, a substantially modified scale, or a population where the original structure is doubtful. Use CFA when you have a clear, theory-based expectation, such as an established three-factor scale. If you need both, the best practice is to run EFA and CFA on different samples, or split a large sample at random, so the confirmation is not circular.
For translated instruments we usually start with CFA of the original structure and move to EFA only if it clearly fails, since that preserves comparability with prior studies.
Handling problem items honestly
Low loadings, cross-loadings and Heywood cases (impossible estimates such as negative variances) are signals worth understanding before acting. An item may be poorly worded, badly translated, reverse-coded incorrectly or genuinely measuring something else. We look at wording and content first, then statistics, and document the decision.
Modification indices can suggest error covariances that improve fit, but adding them without justification amounts to fitting noise. We only recommend changes you can defend — for example, two items with near-identical wording — and we show fit before and after so readers can judge.
What factor analysis cannot tell you
A clean factor structure is evidence of internal structure, one part of validity. It does not show that the scale measures the intended construct rather than a related one, or that it predicts relevant outcomes. We help you present factor analysis as one strand of validity evidence alongside content, convergent and criterion evidence, and note limitations such as single-source, self-report data and sample-specific results.
What the work can look like
Illustrative example — fictional data
Sample excerpt: validating an adapted teacher-wellbeing scale
A doctoral researcher adapted an 18-item, three-factor wellbeing scale for teachers in a new country and collected 640 responses. Numbers are invented for illustration.
CFA of the original structure
Using WLSMV for five-point items, the three-factor model showed CFI = .951, TLI = .943, RMSEA = .061, 90% CI [.055, .067], SRMR = .049. Standardized loadings ranged from .52 to .84, except item W11 (.29).
Item decision
W11's translation used an idiom with a different meaning locally. It was removed with this rationale recorded; the revised model fit slightly better and the construct remained covered by five items.
Reliability
McDonald's omega was .86, .82 and .79 for the three factors. Omega was preferred because loadings varied considerably within factors.
Illustrative example with invented numbers. Real reports follow your instrument, data and reporting standard.
What you’ll need to provide
- The questionnaire with item wording, response scales and intended dimensions
- The source of the scale (original paper or manual) and any translation or adaptation notes
- The anonymized dataset, with reverse-coded items identified
- Group variables if you intend to compare groups
- Any supervisor, reviewer or journal requirements for psychometric reporting
You don’t need everything ready to ask for a quote — tell us what you have and we’ll say what’s missing.
What affects price and turnaround
Price depends on
- Number of items, factors and scales
- Whether EFA, CFA or both are needed, and whether invariance testing is included
- Data condition, including reverse coding and missing data
- Delivered analysis versus guided sessions
- Level of reporting support required
Turnaround depends on
- Availability of the full questionnaire and scale documentation
- Time needed for decisions about problem items
- Number of groups and models to compare
Every project is quoted individually. How we price our work.
Questions people ask before starting
Can I run EFA and CFA on the same data?
It is common but weak, because the CFA then confirms a model derived from the same data. If your sample is large enough we split it randomly; otherwise we explain the limitation clearly so you can acknowledge it.
How many participants do I need?
It depends on the number of items, the strength of loadings and how well-defined the factors are. Rules like '10 per item' are rough. We assess adequacy for your specific model and say honestly if the sample is thin.
What if my fit indices are just below the recommended cut-offs?
Cut-offs such as CFI ≥ .95 or RMSEA ≤ .06 are guidelines, not laws, and depend on model size and estimator. We look at local misfit to understand why, and help you report fit transparently rather than chase a number.
Do you work in JASP or jamovi?
Yes, along with SPSS, R, Mplus and AMOS. JASP and jamovi are good choices if you want a free, menu-driven tool with reproducible output.
Will you write my methods chapter section on validity?
We provide the analysis, tables and explanations, and can review your draft for accuracy. You write the section yourself, which also prepares you for examiner questions.