Psychology Dissertation Factor Analysis Checklist: EFA, CFA, PCA, Parallel Analysis and Omega
Psychology Dissertation Factor Analysis Checklist provides a practical framework for choosing, conducting and interpreting factor analysis in psychological research.
Psychologists often study constructs that cannot be observed directly, such as anxiety, depression, resilience, self-esteem and motivation. These are known as latent constructs. Researchers measure them using observable indicators, usually questionnaire items, and factor analysis helps determine whether those items reflect meaningful underlying dimensions. Factor analysis is therefore important in psychological scale development, construct validation and measurement research. Also Read psychology dissertation reliability and validity guide
What Is Factor Analysis?
Factor analysis examines patterns of correlations among observed variables to identify or test underlying latent factors.
For example, a 15-item academic stress questionnaire may include items relating to examination anxiety, workload pressure and fear of failure. Factor analysis can investigate whether these items form one general stress factor or several distinct dimensions. However, factor analysis does not prove that a psychological construct objectively exists. Results must be interpreted alongside theory, item content, sample characteristics and previous research.
Table of Contents
Exploratory Factor Analysis vs Confirmatory Factor Analysis
The first major decision is whether to use Exploratory Factor Analysis (EFA) or Confirmatory Factor Analysis (CFA).
| Feature | EFA | CFA |
|---|---|---|
|
Purpose |
Explore possible factor structure |
Test a proposed factor structure |
|
Prior theory |
Limited or uncertain |
Clearly specified |
|
Item-factor relationships |
Estimated from the data |
Defined before analysis |
|
Typical use |
Scale development |
Scale validation |
|
Key outputs |
Loadings, communalities, factor correlations |
Loadings, residuals, fit indices |
EFA is useful when the underlying structure is uncertain. A researcher creating a new loneliness scale, for instance, may use EFA to determine whether items cluster into social and emotional loneliness factors. CFA is appropriate when previous theory or research already proposes a structure. It tests whether that predefined model is compatible with the observed data.
Where possible, avoid performing EFA and CFA on exactly the same observations and describing the CFA as independent confirmation.
Factor Analysis vs Principal Component Analysis
Principal Component Analysis (PCA) and factor analysis are related but not equivalent. PCA primarily reduces a large number of observed variables into a smaller number of components. It works with total observed variance.
Common factor analysis focuses on shared variance among variables and assumes that latent constructs help explain their correlations. If the aim is data reduction, PCA may be suitable. If the research question concerns latent psychological traits such as anxiety, wellbeing or depression, common factor analysis is usually more theoretically appropriate.
Factor Extraction Methods
Extraction determines how factors are estimated.
- Principal Axis Factoring – focuses on shared variance and is often used when multivariate normality is questionable.
- Maximum Likelihood – provides model-based estimation and statistical tests when its assumptions are reasonably satisfied.
The extraction method should be chosen according to the research question and characteristics of the data rather than simply accepting the software default.
How Many Factors Should Be Retained?
Factor retention is one of the most important decisions in EFA.
Retaining too few factors may combine distinct psychological constructs. Retaining too many can produce unstable or meaningless factors.
Useful retention evidence includes:
- Parallel Analysis;
- scree-plot inspection;
- theoretical interpretability;
- factor strength;
- communalities; and
- competing factor solutions.
Why Parallel Analysis Is Preferred
The traditional eigenvalue-greater-than-one rule retains factors with eigenvalues above 1. Although simple, it can overestimate the number of factors.
Parallel Analysis compares eigenvalues from the real dataset with eigenvalues generated from random data.A factor is generally retained when its observed eigenvalue exceeds the corresponding random-data eigenvalue.
This approach provides stronger evidence than relying solely on the eigenvalue-greater-than-one rule. However, Parallel Analysis should still be interpreted alongside theory and factor interpretability rather than treated as an automatic decision rule.
Choosing a Factor Rotation
Rotation improves the interpretability of a factor solution.
- Orthogonal rotations, such as varimax, force factors to remain uncorrelated.
- Oblique rotations, including oblimin and promax, allow factors to correlate.
Because psychological constructs often overlap, oblique rotation is frequently more realistic. Anxiety, stress and depression, for example, may represent distinct constructs while still being meaningfully correlated.Researchers should justify the chosen rotation and report factor correlations when using an oblique method.
KMO and Bartlett’s Test
Before interpreting factor analysis results, researchers should assess whether the data are sufficiently factorable.
Kaiser-Meyer-Olkin Measure
The KMO statistic evaluates whether patterns of correlations are suitable for identifying common factors. Higher KMO values generally indicate better factorability.
Bartlett’s Test of Sphericity
Bartlett’s Test assesses whether the correlation matrix differs significantly from an identity matrix.
A significant result indicates that correlations exist among the variables. However, it does not guarantee that a useful or theoretically meaningful factor structure will emerge. KMO, Bartlett’s Test, correlation patterns and theoretical considerations should therefore be interpreted together.
Goodness-of-Fit in Confirmatory Factor Analysis
CFA evaluates how well a predefined measurement model represents the observed data.
Common goodness-of-fit indices include:
|
Index |
Main purpose |
|
χ² |
Tests discrepancy between model and data |
|
CFI |
Compares the model with a baseline model |
|
TLI |
Assesses comparative fit while considering complexity |
|
RMSEA |
Measures approximate model misfit |
|
SRMR |
Evaluates standardised residual differences |
No single index should determine whether a model is acceptable. Researchers should examine multiple fit indices together with factor loadings, residuals, theoretical coherence, sample size and estimation method.
McDonald’s Omega and Reliability
After establishing a defensible factor structure, researchers should assess whether the resulting scores are internally consistent.
McDonald’s Omega (ω) is a model-based reliability coefficient that uses factor loadings and measurement error. Unlike Cronbach’s alpha, omega can accommodate items that contribute differently to the construct.
However, omega is not automatically superior in every situation. Its accuracy depends on the quality of the underlying factor model. This is why dimensionality should normally be evaluated before reliability is interpreted.
Psychology Dissertation Factor Analysis Checklist
Before completing your dissertation analysis, confirm that you have:
- Defined the psychological construct;
- Justified EFA or CFA;
- Distinguished factor analysis from PCA;
- Assessed factorability;
- Selected an appropriate extraction method;
- Justified the number of retained factors;
- Considered Parallel Analysis;
- Chosen an appropriate rotation;
- Evaluated CFA using multiple fit indices; and
- Assessed reliability after establishing dimensionality.
Frequently Asked Questions
v Is EFA better than CFA?
Neither is inherently better. EFA explores an uncertain structure, while CFA tests a theoretically specified model.
v Is PCA the same as factor analysis?
No. PCA mainly reduces variables, while common factor analysis models shared variance through latent constructs.
v Is Parallel Analysis better than the eigenvalue rule?
Generally, yes. It provides stronger empirical evidence, although theoretical interpretation remains essential.
v Does a significant Bartlett’s Test mean factor analysis will work?
No. It only indicates that the variables are correlated sufficiently to reject an identity correlation matrix.
v Should I use McDonald’s Omega instead of Cronbach’s Alpha?
Omega is often preferable when item loadings differ, but its usefulness depends on an appropriate factor model.
Conclusion
A strong Psychology Dissertation Factor Analysis Checklist should connect statistical procedures to psychological measurement theory. Researchers should distinguish EFA from CFA and PCA, justify extraction and rotation choices, assess factorability, use defensible factor-retention methods such as Parallel Analysis and evaluate CFA using multiple fit indices. Reliability measures such as McDonald’s Omega should be interpreted only after establishing a defensible factor structure.
If you need dissertation support, seek guidance that helps you understand assumptions, analysis and reporting while preserving your ownership of the research. Ethical statistical support should clarify your findings, not manufacture data or alter results. You can also refer to our guide on psychology dissertation preregistration guide
