Factor Analysis
Factor analysis is a family of statistical techniques used to investigate patterns of relationships among observed variables and represent those relationships through a smaller number of underlying factors. It is particularly useful when researchers measure concepts that cannot be observed directly, such as customer satisfaction, brand trust, employee engagement or organizational commitment.
Suppose a questionnaire contains 20 statements intended to measure different aspects of employee experience. Rather than treating all 20 responses as unrelated variables, factor analysis can investigate whether groups of items appear to reflect underlying constructs such as job satisfaction, organizational commitment and workplace support.
A fundamental distinction is between exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). EFA is generally appropriate when the researcher needs to explore the underlying factor structure, whereas CFA is used when a theoretically specified measurement structure is being tested. The distinction matters because factor analysis involves consequential methodological decisions rather than a single automatic statistical procedure. Fabrigar et al. (1999), for example, show that choices made during exploratory factor analysis can materially affect the resulting solution. (DOI)
On this page:
- Factor analysis explained simply
- What factor analysis is
- Factors and observed variables
- Factor loadings
- Exploratory vs confirmatory factor analysis
- Factor analysis vs principal component analysis
- When data are suitable for factor analysis
- KMO and Bartlett’s test
- Factor extraction
- Deciding how many factors to retain
- Factor rotation
- Communalities and cross-loadings
- Interpreting and naming factors
- Dudovskiy Factor Analysis Decision Framework
- Application example
- Advantages and limitations
- Common mistakes
- Factor analysis in business research
- Factor analysis in the age of AI
- When to use factor analysis
- Dissertation example
- Exam tip
| Question | Short answer |
|---|---|
| What does factor analysis investigate? | Patterns of relationships among observed variables that may reflect underlying factors |
| What is a factor? | An unobserved or latent dimension represented by relationships among observed variables |
| What is a factor loading? | A parameter indicating the relationship between an observed variable and a factor |
| What is EFA primarily used for? | Exploring uncertain factor structure |
| What is CFA primarily used for? | Testing a theoretically specified factor structure |
| Is PCA the same as factor analysis? | No |
| Should factor number be decided only by eigenvalues >1? | No |
| Must factors always be uncorrelated? | No |
| Does a clean factor structure establish reliability? | No |
| Does factor analysis establish construct validity by itself? | No |
Factor Analysis Explained Simply
Imagine a retailer surveys customers using twelve statements about their shopping experience. Some questions ask whether employees are helpful and knowledgeable. Others concern product quality and reliability, while another group concerns delivery speed and convenience.
The researcher does not initially know whether these twelve responses represent twelve separate aspects of the customer experience or whether they can be understood through a smaller set of underlying dimensions.
Factor analysis examines patterns of relationships among the twelve variables. Suppose the analysis suggests that they cluster around three factors:
Factor 1: Service Quality
→ helpful employees
→ knowledgeable employees
→ responsive employees
→ effective problem resolution
Factor 2: Product Quality
→ product reliability
→ product performance
→ perceived durability
→ consistency of quality
Factor 3: Delivery Convenience
→ delivery speed
→ delivery reliability
→ tracking convenience
→ delivery flexibility
Instead of describing twelve largely separate questionnaire items, the researcher now has evidence suggesting a three-factor representation of the observed responses.
The factors are not directly observed in the same way as questionnaire answers. They are latent constructs inferred from patterns in the observed variables.
What Is Factor Analysis?
Factor analysis refers to statistical models designed to represent relationships among observed variables through a smaller number of latent factors. The underlying idea is that correlations among measured variables may partly arise because those variables share common underlying influences.
For example, responses to statements such as “I intend to purchase from this company again,” “I would recommend this company to others,” and “I consider this company my preferred provider” may be related because they reflect an underlying construct such as customer loyalty.
Factor analysis attempts to model this common structure rather than merely describing pairwise correlations among the individual variables.
This makes it particularly important in research involving latent constructs. Researchers cannot directly observe concepts such as trust, loyalty, engagement or perceived value. They operationalize them through observable indicators and then investigate whether the relationships among those indicators are consistent with the proposed measurement structure.
Factor analysis therefore sits at the intersection of statistics and measurement theory. The calculations matter, but so does the theoretical meaning assigned to the resulting factors.
Observed Variables and Latent Factors
An observed variable is something measured directly. In questionnaire research, individual Likert-scale items are observed variables.
A latent factor is an underlying construct inferred from relationships among those variables.
Consider four questionnaire statements:
Item 1: I trust this brand to keep its promises.
Item 2: This brand is dependable.
Item 3: I believe information provided by this brand.
Item 4: I feel confident buying products from this brand.
The researcher directly observes participants’ responses to these four items. Brand trust, however, is not directly observed. It is theoretically proposed as a latent factor that helps account for their shared variation.
A simplified representation is:
Brand Trust
↓
Item 1
Item 2
Item 3
Item 4
The important methodological point is that the existence and interpretation of a factor cannot be justified simply because statistical software produces a column of loadings. Factors need to make substantive and theoretical sense.
Factor Loadings
A factor loading represents the relationship between an observed variable and a latent factor within the specified factor model. Larger absolute loadings generally indicate a stronger relationship between an item and the factor.
A simplified factor-loading matrix might look like this:
| Item | Service Quality | Product Quality |
|---|---|---|
| Staff are helpful | 0.81 | 0.14 |
| Staff are knowledgeable | 0.76 | 0.11 |
| Problems are resolved effectively | 0.72 | 0.18 |
| Products are reliable | 0.12 | 0.84 |
| Products perform consistently | 0.19 | 0.79 |
| Products are durable | 0.16 | 0.73 |
This is a relatively clear pattern. The first three items load strongly on Service Quality, while the remaining items load strongly on Product Quality.
Researchers sometimes apply fixed rules such as treating loadings above 0.40 or 0.50 as meaningful. Such thresholds can be useful heuristics, but factor interpretation should not be reduced to a universal cutoff. Sample size, research context, theoretical expectations, cross-loadings and the overall pattern of results also matter.
The central question is not simply:
“Is the loading above 0.50?”
It is:
“Does the observed loading pattern provide a coherent and defensible representation of the construct being investigated?”
Exploratory Factor Analysis vs Confirmatory Factor Analysis
The distinction between exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) is fundamental.
| Exploratory Factor Analysis | Confirmatory Factor Analysis |
|---|---|
| Explores possible latent structure | Tests a specified measurement model |
| Appropriate when structure is uncertain | Appropriate when strong theoretical expectations exist |
| Researchers do not fully specify the loading pattern beforehand | Researchers specify which indicators are expected to relate to which factors |
| Factor number and item relationships are investigated | A proposed factor structure is evaluated against observed data |
| Common in early scale development | Common in scale validation and theory testing |
| Requires extraction and rotation decisions | Requires model specification and model-fit assessment |
Suppose a researcher develops a new questionnaire measuring attitudes toward AI-enabled workplaces and does not yet know how the items are structured. EFA can investigate whether the responses suggest dimensions such as perceived usefulness, perceived threat and trust in AI.
Now suppose previous research already proposes those three dimensions and specifies which questionnaire items measure each one. The researcher wants to determine whether that established measurement model is supported in a new sample. CFA is more consistent with that objective.
The difference can therefore be summarized as:
EFA → What factor structure appears to characterize these data?
CFA → How well does this specified factor structure correspond to these data?
This distinction is deliberately simplified. EFA still requires methodological judgment and theory, while CFA is not purely mechanical hypothesis testing. But the explore versus test distinction provides a useful starting point.
Factor Analysis vs Principal Component Analysis
Principal component analysis (PCA) and factor analysis are frequently confused because statistical packages often place them close together and their outputs can appear superficially similar.
They are not identical methods.
Principal component analysis transforms observed variables into components designed to account for variance in those variables. It is fundamentally a data-reduction technique.
Common factor analysis models shared variation among observed variables in terms of underlying latent factors, distinguishing common variance from variance not shared through the common factors.
This distinction matters when the research question concerns latent constructs. If a researcher wants to investigate whether questionnaire items reflect underlying theoretical constructs such as organizational commitment or brand trust, common factor analysis is generally more directly aligned with that purpose than PCA.
Fabrigar et al. (1999) discuss the distinction between common factor analysis and component analysis as one of the important analytical decisions researchers must make. (DOI) Costello and Osborne (2005) similarly caution against simply relying on statistical-software defaults when conducting exploratory factor analysis. (Legacy File Share)
PCA remains a valuable technique. The mistake is not using PCA; it is using PCA while describing the analysis and interpretation as though a latent-factor model had been estimated.
Is the Data Suitable for Factor Analysis?
Not every collection of variables should be subjected to factor analysis. There must be sufficient meaningful relationships among variables for a latent-factor representation to be useful.
Researchers therefore commonly examine the correlation matrix and use diagnostics such as the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity.
Sample size also matters, but simple universal rules such as “factor analysis requires ten respondents per item” can be misleading. The adequacy of a sample depends on characteristics including the strength of loadings, communalities, number of variables associated with each factor and complexity of the factor structure. Methodological literature has therefore moved away from treating a single observations-per-variable ratio as universally sufficient.
The correct question is not:
“Have I reached the magic minimum sample size?”
but:
“Is the information in these data sufficient to estimate the intended factor structure with adequate stability?”
KMO and Bartlett’s Test
The Kaiser-Meyer-Olkin measure assesses aspects of whether the pattern of correlations is suitable for factor analysis. Values closer to 1 generally indicate that the variables share enough common variance structure for factor analysis to be potentially useful.
Researchers often encounter rules such as:
| KMO | Common interpretation |
|---|---|
| ≥ 0.90 | Excellent |
| 0.80–0.89 | Very good |
| 0.70–0.79 | Good |
| 0.60–0.69 | Mediocre/moderate |
| 0.50–0.59 | Poor |
| < 0.50 | Usually problematic |
These should be treated as orientation rather than substitutes for examining the data.
Bartlett’s test of sphericity tests whether the observed correlation matrix differs from an identity matrix. A statistically significant result is commonly taken as evidence that sufficient correlations exist to justify investigating a factor structure.
However, statistical significance can become easier to obtain in large samples. KMO and Bartlett’s test therefore should not become another automatic pair of boxes that students tick before pressing “factor analysis.” Their role is to contribute evidence about whether the correlation structure is suitable for the intended analysis.
Factor Extraction
Once the researcher decides that exploratory factor analysis is appropriate, another decision concerns how the factors will be extracted.
Common approaches include principal axis factoring and maximum likelihood factor analysis. These procedures are based on different estimation principles and assumptions. The appropriate choice depends on the characteristics of the data and analytical objectives.
A frequent mistake is to leave whatever extraction method the software package selects by default without understanding what has actually been estimated. This is particularly problematic when software defaults to principal component analysis and the dissertation subsequently describes the procedure as common factor analysis.
Fabrigar et al. emphasize that extraction method is one of several consequential decisions in EFA and that analytical choices should correspond to the purpose and characteristics of the research. (DOI)
The methodology chapter should therefore name the extraction method and, where relevant, explain why it was appropriate.
How Many Factors Should Be Retained?
Determining the number of factors is one of the most important decisions in exploratory factor analysis.
Historically, researchers have frequently used the eigenvalue-greater-than-one rule, sometimes called the Kaiser criterion. Another familiar method is visual inspection of a scree plot.
Neither should automatically determine the solution.
Researchers can also use methods such as parallel analysis, which compares eigenvalues from the observed data with those expected from appropriately generated random data. Multiple sources of evidence can be considered together, including statistical criteria, interpretability and theoretical coherence.
Costello and Osborne identify the number of factors to retain as one of the major decisions researchers need to make in EFA, alongside extraction, rotation and sample considerations. (Legacy File Share)
Suppose the eigenvalue rule suggests four factors, the scree plot appears to suggest three, and parallel analysis supports three. The researcher should not quietly select whichever solution looks most convenient. The disagreement itself deserves examination.
A statistically possible factor solution is not necessarily a theoretically meaningful one.
Factor Rotation
After factors have been extracted, rotation is often used to make their structure easier to interpret.
Two broad families are:
Orthogonal rotation → factors are constrained to remain uncorrelated.
Oblique rotation → factors are permitted to correlate.
Varimax is a familiar orthogonal rotation, while methods such as oblimin and promax allow correlated factors.
The choice should reflect substantive expectations rather than habit. Constructs in business and social research frequently have plausible relationships with one another. Customer satisfaction and customer loyalty, for example, need not be statistically independent merely because they are conceptually distinguishable.
Costello and Osborne recommend considering oblique rotation because correlated factors are common in social-science research; if factors prove essentially uncorrelated, an oblique solution can still reveal that pattern. (Legacy File Share)
The question is therefore not:
“Which rotation is most popular?”
but:
“Does theory permit the underlying constructs to be related?”
Communalities and Cross-Loadings
A communality represents the proportion of an observed variable’s variance accounted for by the common factors in the model. Very low communalities can indicate that an item is not well represented by the extracted factor solution.
A cross-loading occurs when an item has meaningful loadings on more than one factor.
Suppose an item stating “This retailer responds quickly when I have a problem” loads 0.58 on Service Quality and 0.49 on Customer Support. The item does not provide a clean distinction between the two proposed factors.
That does not mean it must automatically be deleted. The overlap may reveal ambiguous wording, theoretically overlapping constructs, an inappropriate factor solution or a genuinely complex relationship.
Mechanical item deletion can produce a statistically cleaner result while reducing the conceptual coverage of a construct. As with Cronbach’s alpha, researchers should investigate why an item behaves unexpectedly before removing it simply to improve statistical output.
Interpreting and Naming Factors
Statistical software can estimate factor loadings. It cannot determine the substantive meaning of a factor independently of the research context.
Suppose five items loading on one factor concern supervisor support, feedback, recognition, communication and accessibility. The researcher might label the factor Managerial Support.
That label should reflect the common conceptual meaning of the items, the theoretical literature and the research question. It should not be selected merely because it sounds plausible.
Researchers should also avoid circular reasoning. If a factor is labelled “employee engagement” simply because the researcher hoped to find employee engagement, the interpretation has not actually been justified. The observed item pattern must support the proposed meaning.
Factor naming is therefore an interpretive stage in which statistical structure and theoretical reasoning meet.
Dudovskiy Factor Analysis Decision Framework
The Dudovskiy Factor Analysis Decision Framework synthesizes established methodological principles into a practical decision process. It does not propose a new form of factor analysis. Its purpose is to help researchers decide what kind of factor-analytic question they are asking before selecting statistical procedures.
Research Purpose → Theoretical Maturity → Explore or Test? → EFA or CFA → Evaluate Factor Structure → Interpret and Justify
The central principle is:
Use factor analysis according to the research question and theoretical maturity: explore uncertain latent structure with EFA; test a theoretically specified measurement structure with CFA.
1. Research Purpose
Begin with the methodological problem. Is the objective to discover possible dimensions underlying a set of variables, develop a measurement scale, investigate the structure of an unfamiliar construct, or evaluate an existing measurement model?
The answer determines what factor analysis is expected to accomplish.
2. Theoretical Maturity
Consider how much is already known about the measurement structure. A newly developed questionnaire addressing an emerging phenomenon presents a different methodological situation from a well-established scale supported by previous factor-analytic research.
Strong prior theory does not guarantee that the proposed structure will fit a new dataset, but it changes the type of question the researcher can reasonably ask.
3. Explore or Test?
The central decision is whether the factor structure remains sufficiently uncertain that it needs to be explored or whether a specific structure has already been defined and should be evaluated.
This prevents a common methodological reversal: conducting an exploratory analysis and then describing the discovered structure as though it had been specified in advance.
4. EFA or CFA
If the primary objective is to investigate an uncertain latent structure, EFA is generally appropriate.
If theory and previous evidence support a specific measurement structure that the researcher wishes to test, CFA is generally appropriate.
The decision should emerge from the research purpose rather than from which statistical software the researcher happens to know.
5. Evaluate Factor Structure
The researcher then evaluates the relevant evidence. In EFA this can involve factor retention, loadings, cross-loadings, communalities, rotation and interpretability. CFA instead focuses on the specified measurement model, parameter estimates, model fit and theoretically defensible model evaluation.
These analyses answer related but different questions and should not be treated as interchangeable procedures.
6. Interpret and Justify
The final factor solution must be interpreted in relation to theory and the research context. Statistical results alone do not name constructs or establish their substantive meaning.
The researcher should therefore be able to explain not only what factor structure emerged or was supported, but also why the selected analytical approach was appropriate for the methodological question.
The framework transforms the starting question from:
“How do I run factor analysis?”
into:
“What do I know about the measurement structure, what do I need to learn about it, and which factor-analytic approach answers that question?”
Application of Factor Analysis: an Example
Consider a researcher developing a questionnaire to investigate consumer resistance to autonomous delivery services. Based on interviews and literature review, the researcher develops 24 questionnaire items covering concerns about safety, loss of human interaction, privacy, reliability, technological complexity and employment consequences.
Although theory suggests several possible dimensions, the structure of this new scale has not yet been established. The researcher therefore uses exploratory factor analysis rather than imposing a predetermined measurement model.
Initial diagnostics indicate that the correlation structure is suitable for factor analysis. The researcher selects an appropriate common-factor extraction method and uses an oblique rotation because dimensions of consumer resistance are theoretically expected to correlate.
The analysis suggests four interpretable factors rather than the six broad themes initially considered during questionnaire development. Several privacy and technological-control items cluster together, while safety and reliability items form another factor. Two items show substantial cross-loadings and are examined for conceptual ambiguity rather than being automatically deleted.
The resulting four-factor structure provides an empirically informed model that can subsequently be evaluated in an independent sample using CFA.
This sequence illustrates an important principle:
Qualitative/theoretical item development → EFA → proposed measurement structure → independent CFA
EFA is not being used merely because the researcher has questionnaire data. It is being used because the latent structure of a newly developed measure remains uncertain.
Advantages and Limitations of Factor Analysis
Factor analysis allows researchers to investigate complex relationships among many observed variables without interpreting each relationship independently. In measurement research, this provides a way of examining whether sets of indicators appear to represent theoretically meaningful latent constructs. It can therefore contribute substantially to questionnaire development, measurement refinement and evaluation of theoretical models.
Its ability to reduce apparent complexity is especially valuable in business research, where concepts such as service quality, organizational culture and customer experience can involve numerous observed indicators. Factor analysis can reveal whether those indicators exhibit a more manageable underlying structure.
The apparent mathematical objectivity of factor analysis can nevertheless disguise considerable researcher judgment. Decisions concerning variables, sample, extraction, factor retention, rotation, treatment of cross-loadings and interpretation can change the resulting solution. Fabrigar et al. demonstrate why questionable analytical decisions can lead to problematic factor-analytic results. (DOI)
Results are also dependent on the sample and measurement context. A factor structure identified in one population does not automatically become a universal property of the questionnaire. Replication and confirmatory evaluation may be necessary before strong claims about measurement structure are justified.
Factor analysis can therefore provide powerful evidence about latent structure, but the quality of that evidence depends on the quality of the measurement design, data, statistical decisions and theoretical interpretation.
Common Mistakes When Using Factor Analysis
Treating PCA as synonymous with factor analysis can produce a mismatch between research purpose and statistical method. PCA is valuable for data reduction, but researchers investigating latent constructs should understand the distinction between components and common factors rather than relying automatically on software defaults.
Selecting factors solely because their eigenvalues exceed one similarly turns a methodological decision into an automatic rule. Factor retention should consider stronger statistical evidence, the pattern of loadings and theoretical interpretability rather than one criterion in isolation.
Using Varimax automatically can impose uncorrelated factors even when the constructs are theoretically expected to relate. Rotation should correspond to plausible relationships among the factors rather than to familiarity with a particular menu option.
Deleting every item with an inconvenient loading can gradually transform the measurement instrument into something different from the construct originally defined. Statistical diagnostics should identify issues for investigation; they should not automatically dictate theoretical decisions.
A further problem occurs when EFA is used to discover a factor structure and the same dataset is then treated as strong confirmation of that structure. Exploration and confirmation are logically different activities. Where feasible, a structure developed through EFA should subsequently be evaluated with new data or an appropriately independent sample.
Finally, a clear factor solution does not prove that a scale is reliable or valid. Reliability coefficients such as Cronbach’s alpha or McDonald’s omega address related but distinct measurement questions, while construct validity requires a broader body of evidence.
Factor Analysis in Business Research
Factor analysis is particularly important in business research because many variables of interest are latent. Researchers cannot directly observe brand loyalty, entrepreneurial orientation, perceived service quality, employee engagement or organizational commitment in the same way that they can observe revenue or employee headcount.
Questionnaires therefore frequently operationalize these constructs through multiple indicators. Factor analysis helps researchers investigate whether the observed responses correspond to the proposed conceptual structure.
A researcher studying digital banking adoption, for example, might measure perceived usefulness, perceived ease of use, trust and perceived risk using several items for each construct. EFA may be appropriate during the development of a new instrument, while CFA can subsequently evaluate whether the proposed four-factor measurement structure is supported.
Factor analysis is consequently more than a technique for simplifying large datasets. In much business and management research, it forms part of the methodological bridge between abstract theoretical constructs and observable empirical measurements.
Factor Analysis in the Age of AI and Digital Research
AI has made factor analysis considerably easier to perform. Researchers can generate R or Python code, receive explanations of factor-loading matrices, compare rotations and ask AI systems to interpret unfamiliar statistical output. This reduces technical barriers that previously made multivariate analysis difficult for many dissertation students.
The corresponding risk is automated methodological overconfidence. An AI system can produce a factor solution even when the researcher has not adequately defined the construct, justified the sample, selected an appropriate correlation matrix or understood the difference between PCA and common factor analysis. Computational correctness does not guarantee methodological appropriateness.
AI can also suggest factor names after examining groups of questionnaire items. This can be genuinely useful for brainstorming interpretations, but the resulting label should be checked against the wording of the items, relevant theory and the conceptual definition of the construct. Factor naming is a substantive research judgment, not merely a language-generation task.
Generative AI also makes questionnaire creation extremely easy. A researcher can generate dozens of apparently plausible Likert items in seconds. That convenience increases rather than decreases the need for rigorous measurement evaluation because linguistic plausibility does not demonstrate that generated items represent the intended latent structure.
The most productive use of AI is therefore to support analytical reasoning—checking code, explaining alternatives, identifying assumptions and challenging interpretations—while leaving the researcher responsible for the theoretical and methodological argument.
When to Use Factor Analysis
Factor analysis may be appropriate when:
- the research involves multiple correlated observed variables;
- the researcher is interested in underlying latent constructs;
- a questionnaire contains multiple indicators intended to represent one or more constructs;
- the dimensional structure of a newly developed instrument needs to be explored;
- an established theoretical measurement structure needs to be evaluated;
- the researcher wants to investigate whether observed variables can be represented by fewer underlying dimensions;
- the data and sample provide an adequate basis for estimating the intended factor model;
- the researcher can theoretically interpret the resulting factors rather than relying solely on statistical output.
Factor analysis is less appropriate when there is no meaningful theoretical or empirical reason to expect common latent structure among the variables.
Dissertation Example
A dissertation titled “The Influence of Digital Workplace Experience on Employee Retention Intentions” develops a questionnaire containing 18 items intended to capture different aspects of employees’ digital working environment. Although previous literature identifies possible dimensions such as technology usability, digital collaboration and organizational technology support, the precise factor structure of the adapted instrument has not been established for the population studied.
In the methodology chapter, the researcher therefore justifies exploratory factor analysis as an appropriate procedure for investigating the latent structure of the adapted scale. The chapter explains how factorability was evaluated, identifies the extraction and rotation procedures, and states the criteria used to determine the number of factors rather than simply reporting that “factor analysis was conducted.”
The resulting analysis supports three interpretable factors broadly corresponding to technology usability, digital collaboration and technology support. Factor loadings, communalities and cross-loadings are examined alongside theoretical considerations when evaluating individual items. The researcher does not claim that this exploratory result conclusively validates the measurement model.
The methodology subsequently distinguishes this structural evidence from reliability evidence. Cronbach’s alpha and/or McDonald’s omega are calculated for the relevant scale scores rather than being treated as substitutes for factor analysis. The dissertation therefore presents a coherent measurement argument connecting construct definition, factor structure, reliability and subsequent hypothesis testing.
Exam Tip
If asked to explain factor analysis, avoid defining it merely as “a technique for reducing many variables into fewer variables.” That description is closer to the objective commonly associated with principal component analysis and overlooks the latent-variable logic central to common factor analysis.
A stronger answer explains that factor analysis investigates patterns of relationships among observed variables in order to model underlying latent factors. It should also distinguish between EFA, which explores uncertain factor structure, and CFA, which evaluates a theoretically specified measurement model.
For EFA, remember that factor analysis is not one button or one decision. Extraction, number of factors, rotation and interpretation all matter; methodological research has long shown that poor decisions at these stages can produce problematic solutions. (DOI)
The principle to remember is:
Do not begin by asking which factor-analysis button to press. Begin by asking whether you are trying to discover a measurement structure or test one.
Build a methodology you can explain and defend
Factor analysis involves interconnected decisions about measurement, dimensionality and statistical analysis. Dudovskiy Research Assistant can help you determine whether EFA or CFA fits your dissertation, connect factor analysis with reliability testing, and explain the methodological choices you need to justify.
References
Brown, T.A. (2015). Confirmatory Factor Analysis for Applied Research. 2nd ed. New York: Guilford Press.
Hair, J.F., Black, W.C., Babin, B.J. and Anderson, R.E. (2019). Multivariate Data Analysis. 8th ed. Cengage.