Exploratory Factor Analysis

Exploratory factor analysis (EFA) is a statistical method used to investigate the underlying structure of relationships among a set of observed variables when that structure has not been sufficiently established in advance. It is commonly used to explore whether questionnaire items or other measured variables can be represented by a smaller number of underlying latent factors.

For example, a researcher developing a 25-item questionnaire about consumers’ attitudes toward sustainable products may suspect that several underlying dimensions exist but may not know precisely how many factors there are or which items belong to each factor. EFA allows the researcher to investigate that structure empirically rather than imposing a fully specified measurement model beforehand.

EFA, however, is not a single statistical procedure that produces an automatically correct answer. Researchers make consequential decisions about factorability, extraction, the number of factors retained, rotation and interpretation. Methodological literature has therefore long treated EFA as a multi-stage analytical process rather than simply a software command (Fabrigar et al., 1999; Costello and Osborne, 2005).

On this page:

• Exploratory factor analysis explained simply
• What exploratory factor analysis is
• EFA vs CFA
• EFA vs PCA
• When data are suitable for EFA
• Factor extraction
• How many factors to retain
• Parallel analysis
• Factor rotation
• Factor loadings, cross-loadings and communalities
• Interpreting and naming factors
• Dudovskiy EFA Decision Framework
• Application example
• Advantages and limitations
• Common mistakes
• EFA in business research
• EFA in the age of AI
• When to use EFA
• Dissertation example
• Exam tip

Methodological question	EFA decision
Is the latent structure already firmly specified?	If yes, CFA may be more appropriate
Are the variables sufficiently related?	Assess factorability before extraction
What is being modelled?	Common latent factors rather than simply components
How should factors be extracted?	Select an extraction method appropriate to the data and purpose
How many factors should be retained?	Use converging evidence rather than one automatic rule
Can factors plausibly correlate?	If yes, consider oblique rotation
Is an item acceptable?	Examine loadings, cross-loadings, communalities and theoretical meaning
Is the solution defensible?	Statistical evidence and theoretical interpretation should converge

Exploratory Factor Analysis Explained Simply

Imagine that a company wants to understand why employees choose to remain with the organization. Researchers create 18 questionnaire statements covering pay, relationships with managers, promotion opportunities, workplace atmosphere, recognition and professional development.

They suspect that these responses may reflect a smaller number of underlying dimensions, but they do not yet know exactly what those dimensions are.

EFA might reveal that several questions concerning promotion, training and career opportunities tend to move together, suggesting a factor that could be interpreted as Career Development. Questions concerning recognition, managerial communication and supervisor support might form another factor interpreted as Managerial Support, while salary and benefits items might form Financial Rewards.

The researcher has therefore moved from:

18 observed questionnaire items

to a possible structure involving:

Career Development + Managerial Support + Financial Rewards

The crucial word in EFA is exploratory. The researcher did not begin by specifying exactly which items must load on each factor and then test that predetermined structure. Instead, the relationships among the observed variables helped reveal a plausible latent structure.

Dudovskiy Research Assistant

Not sure if exploratory factor analysis is the correct methodology for your dissertation?

Enter your topic to receive a free, tailored methodology preview.

Dissertation topic

Generate Free Methodology Preview → Free · No account required

Learn more about Dudovskiy Research Assistant

What Is Exploratory Factor Analysis?

Exploratory factor analysis is a form of common factor analysis designed to identify latent dimensions that may account for patterns of relationships among observed variables. It is particularly appropriate when the researcher has insufficient evidence to specify the measurement structure completely in advance.

The underlying reasoning is that observed variables may correlate partly because they reflect common latent constructs. If responses to several questionnaire items concerning confidence, honesty and dependability are strongly related, for example, those relationships may partly reflect an underlying construct such as trust.

EFA estimates a factor structure that helps researchers understand such patterns. This normally involves examining how strongly variables relate to factors, whether variables relate substantially to more than one factor, how much of their variance is represented by the common-factor solution, and whether the resulting factors can be interpreted meaningfully.

Fabrigar et al. (1999) emphasize that EFA involves several important methodological choices, including decisions about extraction, number of factors and rotation. These choices can materially affect the solution, which is why simply accepting statistical-software defaults is difficult to defend methodologically.

Exploratory Factor Analysis vs Confirmatory Factor Analysis

EFA and confirmatory factor analysis (CFA) both investigate latent structure, but they address different methodological situations.

Exploratory Factor Analysis (EFA)	Confirmatory Factor Analysis (CFA)
Explores an uncertain factor structure	Evaluates a specified measurement structure
Loading structure is not fully imposed beforehand	Expected relationships between indicators and factors are specified
Often used during scale development	Often used during scale validation
Helps generate a plausible measurement structure	Tests whether a proposed structure is consistent with the data
Requires extraction, factor-retention and rotation decisions	Requires model specification and model-fit evaluation
Primarily exploratory	Primarily confirmatory

Suppose researchers create a new scale measuring consumer anxiety about artificial intelligence. Existing theory suggests that AI anxiety may have several dimensions, but there is insufficient evidence to specify precisely which questionnaire items represent which dimensions. EFA would be a reasonable way of investigating the structure.

If subsequent research proposes a specific four-factor model and identifies which items should measure each factor, researchers could collect new data and use CFA to evaluate that specified measurement structure.

The distinction can be summarized as:

EFA → What latent structure appears to characterize these variables?

CFA → Is this specified latent structure adequately represented in these data?

EFA is not atheoretical, however. Theory influences variable selection, analytical decisions and factor interpretation. The difference is that EFA permits substantially greater uncertainty about the structure being investigated.

Exploratory Factor Analysis vs Principal Component Analysis

EFA is also frequently confused with principal component analysis (PCA), partly because statistical software packages have historically made PCA easy to select within factor-analysis menus.

PCA and EFA can sometimes produce superficially similar outputs, but their objectives and statistical models differ. PCA constructs components from observed variables in order to account for their variance and is particularly useful for data reduction. Common factor analysis seeks to model shared variation among observed variables in terms of underlying latent factors.

This distinction becomes especially important when researchers make claims about latent constructs. If a dissertation states that the purpose of the analysis is to investigate underlying dimensions such as customer trust or employee engagement, the methodological logic of common factor analysis generally corresponds more directly to that objective.

Fabrigar et al. (1999) discuss this distinction as one of the important choices in exploratory analysis, while Costello and Osborne (2005) caution against treating PCA as the automatic default for research intended to investigate latent constructs.

PCA is not an inferior technique. It simply answers a different type of question.

Assessing Whether Data Are Suitable for EFA

EFA requires meaningful relationships among the observed variables. If the variables have little in common, attempting to explain them through common latent factors is unlikely to produce a useful solution.

Researchers therefore normally begin by examining the correlation structure. Variables that correlate neither with one another nor with theoretically related items may provide little evidence of common factors. Extremely high correlations can raise different concerns, including redundancy.

The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity are also frequently reported. KMO provides information about whether the correlation pattern is sufficiently compact for factor analysis, while a statistically significant Bartlett’s test indicates that the correlation matrix differs from an identity matrix.

Neither statistic should be interpreted as permission to proceed automatically. A significant Bartlett’s test can be relatively easy to obtain with a sufficiently large sample, while an acceptable overall KMO can conceal problems with particular variables. These diagnostics should therefore form part of a broader assessment of factorability.

Sample adequacy requires similar judgment. Rules such as “five respondents per variable” or “ten respondents per item” are convenient but overly simple. The stability of a factor solution also depends on features such as communalities, factor loadings, number of indicators per factor and complexity of the underlying structure.

Factor Extraction in EFA

Once EFA is justified, the researcher needs to decide how common factors will be extracted.

Two commonly encountered approaches are principal axis factoring (PAF) and maximum likelihood (ML) factor analysis. Maximum likelihood has useful inferential properties under appropriate distributional assumptions, while principal axis factoring can be used without the same reliance on multivariate normality.

The appropriate choice therefore depends on the purpose of the analysis, characteristics of the data and assumptions of the estimation method. There is no reason to select an extraction procedure merely because it happens to appear first in a software menu.

The distinction between extraction and factor retention is also important. Extraction concerns how the factor model is estimated. Factor retention concerns how many factors should remain in the solution.

These decisions interact, but they are not the same decision.

How Many Factors Should Be Retained?

Determining the number of factors is one of the most consequential stages of EFA. Retaining too few factors can combine conceptually distinct dimensions, while retaining too many can create unstable or trivial factors that represent noise rather than meaningful latent structure.

The familiar eigenvalue-greater-than-one rule, often called the Kaiser criterion, remains widely encountered. A scree plot provides another traditional approach by displaying eigenvalues and allowing researchers to inspect where their decline begins to level off.

Neither should normally be treated as an unquestionable decision rule. Research on factor retention has demonstrated limitations of simple heuristics, while methods such as parallel analysis can provide stronger evidence. Contemporary research continues to describe the eigenvalue-greater-than-one rule as overly simplistic and parallel analysis as an important alternative. (PubMed Central (PMC))

Theoretical interpretability also matters. A statistically suggested factor consisting of two weakly related items with no coherent conceptual meaning may not constitute a defensible construct simply because a numerical criterion retained it.

A strong factor-retention decision therefore draws on converging evidence rather than searching for one statistic that eliminates researcher judgment.

Parallel Analysis

Parallel analysis provides a more evidence-based way of deciding how many factors to retain. Originally proposed by Horn (1965), it compares eigenvalues obtained from the observed data with eigenvalues that would be expected from random data of comparable dimensions.

The reasoning is intuitive. Sampling variation can generate apparently meaningful eigenvalues even when no substantive latent structure exists. Parallel analysis asks whether the observed factors account for more structure than would reasonably be expected from random data.

Hayton, Allen and Scarpello (2004) describe factor retention as a critical EFA decision and identify parallel analysis as one of the more accurate approaches available for this purpose. (Sage Journals)

Suppose the first four observed eigenvalues are larger than the corresponding reference eigenvalues from the parallel analysis, while the fifth is not. This provides evidence supporting retention of four factors.

Even then, parallel analysis does not name the factors or establish their substantive validity. The resulting factor solution still needs to be estimated, examined and interpreted. Research using parallel analysis explicitly notes that researchers remain responsible for determining whether the resulting grouping of items is theoretically coherent. (PubMed Central (PMC))

Factor Rotation

An initial factor solution can be difficult to interpret because variables may show complex relationships with several factors. Rotation is used to obtain a factor representation that is easier to interpret.

A fundamental distinction is between orthogonal and oblique rotation.

Orthogonal rotations such as Varimax constrain the factors to remain uncorrelated. Oblique rotations such as oblimin and promax allow factors to correlate.

The choice should be theoretically defensible. Constructs investigated in business and social research often have plausible relationships. Employee satisfaction may relate to organizational commitment; perceived service quality may relate to customer trust; perceived usefulness may relate to technology adoption intentions.

Forcing such factors to be uncorrelated merely because Varimax is familiar can impose an unrealistic assumption on the solution. Costello and Osborne (2005) discuss oblique rotation as particularly relevant when correlations among underlying constructs are plausible.

Rotation should therefore serve interpretation without overriding substantive theory.

Factor Loadings, Cross-Loadings and Communalities

Once a factor solution has been extracted and rotated, researchers examine the pattern of relationships between variables and factors.

A factor loading indicates the relationship between an observed variable and a factor. Strong loadings can provide evidence that an item meaningfully represents the factor, although no loading threshold should be treated as universally decisive.

A cross-loading occurs when an item relates substantially to more than one factor. Suppose a questionnaire item loads 0.62 on Factor 1 and 0.55 on Factor 2. Assigning it unambiguously to Factor 1 simply because 0.62 is the larger number would ignore considerable evidence that the item reflects both dimensions.

Communality concerns the extent to which an observed variable’s variance is represented by the common factors. Very low communalities may indicate that an item is poorly represented by the factor solution.

These diagnostics should be interpreted together. A researcher might encounter an item with a moderately strong primary loading but also a substantial cross-loading and weak theoretical fit. That combination deserves more scrutiny than the primary loading alone would suggest.

Should Problematic Items Be Deleted?

Item deletion is one of the stages at which an exploratory analysis can become excessively mechanical.

Suppose an item has a low loading, substantial cross-loading or low communality. Removing it may improve the statistical clarity of the factor solution, but that does not automatically mean deletion improves the measurement instrument.

The item may represent an important aspect of the construct that other questions fail to capture. Its unexpected behaviour may also reveal ambiguous wording, a population-specific interpretation, an inadequately defined construct or an incorrect proposed factor structure.

Researchers should therefore ask two questions together:

What happens statistically if the item is removed?

and

What happens conceptually to the construct if the item is removed?

Repeatedly deleting inconvenient items until a visually clean loading matrix emerges can produce an instrument that is statistically tidy but conceptually narrow.

Interpreting and Naming Factors

EFA identifies patterns in the data; it does not independently determine what those patterns mean.

Suppose five items loading strongly on one factor concern managers providing feedback, recognizing achievement, listening to employees, communicating clearly and offering support. The researcher might interpret the common dimension as Supportive Leadership.

That interpretation should be based on the shared meaning of the items and relevant theoretical literature. Calling the factor “Supportive Leadership” merely because this was the construct the researcher hoped to discover would reverse the logic of interpretation.

Factor naming therefore involves an interaction between empirical structure and substantive knowledge.

A good factor label should summarize what the strongly associated items genuinely have in common without claiming more than the observed pattern supports.

Dudovskiy EFA Decision Framework

The Dudovskiy EFA Decision Framework synthesizes established principles of exploratory factor analysis into a sequence of methodological decisions. It does not propose a new statistical form of EFA. Its purpose is to help researchers move from an exploratory measurement problem to a factor solution they can explain and defend.

Research Purpose → Factorability → Extraction Method → Number of Factors → Rotation → Loading Pattern → Theoretical Interpretation → Defensible Factor Solution

The central principle is:

A defensible EFA solution emerges from converging statistical and theoretical evidence—not from accepting the software’s default settings.

1. Research Purpose

Start by defining why EFA is required. The researcher may be developing a new measurement instrument, adapting an existing scale to a substantially different context, or investigating a construct whose dimensional structure remains uncertain.

This step prevents EFA from becoming an exploratory exercise performed simply because many questionnaire items are available.

2. Factorability

Determine whether the observed variables contain sufficient common structure to make EFA meaningful. Correlation patterns, KMO, Bartlett’s test and characteristics of the sample can contribute to this assessment.

The purpose is not to pass a single statistical threshold but to establish that estimating common factors is methodologically reasonable.

3. Extraction Method

Choose an extraction approach that corresponds to the analytical objective and characteristics of the data. Researchers investigating latent constructs should distinguish common-factor extraction from PCA rather than accepting a software default without justification.

The methodology chapter should identify the selected method and explain the reasoning where the choice is consequential.

4. Number of Factors

Determine how many factors the evidence supports. Parallel analysis, scree information, statistical characteristics of the solution and theoretical interpretability can be considered together.

Factor retention is therefore treated as an evidence-integration problem rather than a mechanical eigenvalue rule.

5. Rotation

Choose a rotation consistent with plausible relationships among the factors. If the latent constructs can reasonably correlate, an oblique rotation may provide a more realistic representation than automatically imposing orthogonality.

The decision should follow theory rather than habit.

6. Loading Pattern

Examine primary loadings, cross-loadings, communalities and problematic items as a pattern rather than as isolated numbers.

The objective is not merely to produce the cleanest possible table. It is to determine whether the variables provide a coherent empirical representation of the proposed factors.

7. Theoretical Interpretation

Interpret and name the factors using the content of the variables, relevant literature and the research context. Statistical software can reveal loading patterns but cannot independently establish the conceptual meaning of a latent construct.

Theory therefore returns explicitly at the interpretation stage.

8. Defensible Factor Solution

The final solution should be statistically credible, theoretically coherent and transparently reported. Researchers should be able to explain why the factors were retained, how the analytical choices were made and what evidence supports their interpretation.

The framework therefore shifts EFA away from:

“Which options should I select in SPSS?”

toward:

“What combination of statistical and theoretical evidence supports this factor solution?”

Application of Exploratory Factor Analysis: an Example

Consider a researcher developing a new scale to examine consumer reluctance to use cashierless retail stores. Interviews and literature review produce 28 questionnaire items concerning privacy, payment security, technological confidence, loss of human interaction, perceived surveillance, reliability and convenience.

The researcher expects consumer reluctance to be multidimensional but does not have sufficiently established theory to specify the complete measurement structure in advance. EFA is therefore selected to explore how the items are empirically organized.

Before extraction, the researcher examines correlations and factorability diagnostics. A common-factor extraction method is selected, and parallel analysis is considered alongside the scree plot and theoretical interpretability when determining the number of factors. Because dimensions such as privacy concerns and perceived surveillance could plausibly correlate, an oblique rotation is used rather than automatically forcing factors to be independent.

The analysis supports four interpretable factors. Several privacy and surveillance items cluster together, technological-confidence items form another factor, concerns about loss of human interaction form a third, and reliability/security items form a fourth. Two questionnaire items cross-load substantially on multiple factors.

Instead of deleting those items immediately, the researcher returns to their wording and conceptual purpose. One is judged ambiguous and removed, while the other is retained because it captures a theoretically important aspect of the construct and its loading pattern remains interpretable.

The final solution is presented as an exploratory measurement structure, not as conclusive validation. A subsequent study could collect new data and evaluate the proposed four-factor structure using CFA.

Advantages and Limitations of Exploratory Factor Analysis

EFA is particularly valuable when researchers have good reason to expect latent structure but insufficient evidence to specify that structure completely in advance. It allows relationships among many observed variables to be represented through a smaller and potentially more meaningful set of underlying dimensions. This makes EFA especially useful in scale development, adaptation of measurement instruments and early investigation of emerging constructs.

The method also provides considerably more information than simple item-by-item analysis. Loadings, cross-loadings, communalities and factor correlations can reveal that apparently similar questionnaire items behave differently or that theoretically distinct concepts overlap empirically. Such findings can improve both measurement instruments and the conceptual understanding of the phenomenon being investigated.

The flexibility that makes EFA useful also creates vulnerability to researcher decisions. Extraction method, factor-retention criteria, rotation and treatment of problematic items can all affect the resulting solution. Different defensible analytical choices can sometimes produce different representations of the same data. This is one reason methodological guidance emphasizes careful decision-making rather than default settings (Fabrigar et al., 1999; Costello and Osborne, 2005).

An EFA solution is also sample-dependent. A clear structure in one dataset does not demonstrate that exactly the same structure will appear in another population, country, industry or period. When the purpose is to establish a measurement model rather than explore one, independent confirmatory evidence becomes important.

EFA should consequently be understood as a powerful method for discovering and refining plausible latent structure, not as an automatic procedure for proving that a construct has a particular structure.

Common Mistakes When Using Exploratory Factor Analysis

Software defaults can quietly become methodological decisions. Selecting PCA because it appears automatically, retaining every factor with an eigenvalue above one and applying Varimax because it is familiar may produce output, but the sequence does not constitute a reasoned EFA strategy.

Factor retention is particularly susceptible to this problem. The eigenvalue-greater-than-one rule is easy to apply but can misidentify the number of factors, whereas parallel analysis explicitly compares observed eigenvalues with those expected from random data. Methodological studies have repeatedly treated parallel analysis as a stronger factor-retention approach, although even it does not eliminate the need for theoretical interpretation. (PubMed Central (PMC))

Rotation creates another potential mismatch between theory and analysis. Automatically requiring orthogonal factors can be difficult to justify when the constructs under investigation are reasonably expected to correlate. Conversely, selecting an oblique rotation does not mean that factors must be strongly correlated; it allows the data to represent such relationships if they exist.

Repeated item deletion can also distort the measurement instrument. Removing every weak or cross-loading item until the output becomes visually clean may reduce conceptual coverage and create a scale optimized to one sample. Statistical diagnostics should trigger methodological investigation rather than automatic deletion.

The distinction between exploration and confirmation must also be respected. Discovering a factor structure through EFA and then treating the same analysis as confirmation of that structure exaggerates the strength of the evidence. Where the research objective requires confirmation, the exploratory solution should ideally be evaluated using independent data through CFA.

Exploratory Factor Analysis in Business Research

Business research frequently examines concepts that are theoretically important but impossible to observe directly. Customer trust, entrepreneurial orientation, perceived value, organizational commitment and resistance to technological change are examples of constructs typically measured through multiple observed indicators.

EFA becomes particularly valuable when researchers investigate emerging business phenomena for which measurement structures remain uncertain. Consider a study developing a scale for employee perceptions of algorithmic management. Potential items might concern perceived surveillance, fairness of automated decisions, autonomy, transparency, efficiency and trust.

Existing organizational theories may help generate those items without establishing precisely how employees organize these perceptions psychologically. EFA can investigate whether the observed responses support the proposed dimensions, combine some of them, or reveal a structure that differs from initial expectations.

This is particularly relevant for dissertations adapting established measures to substantially different business settings. Adaptation does not automatically require EFA, but meaningful changes in item wording, context, population or construct interpretation can create a legitimate exploratory measurement question.

Exploratory Factor Analysis in the Age of AI and Digital Research

Generative AI can now perform many practical parts of an EFA workflow remarkably quickly. It can generate statistical code, explain KMO and Bartlett’s test, produce parallel-analysis commands, compare rotation methods and summarize a factor-loading matrix. These capabilities reduce the technical barrier to sophisticated quantitative analysis.

The methodological danger is that EFA contains exactly the kinds of decisions that an AI system can make plausibly without necessarily making them appropriately. A model can recommend removing an item because its loading is low, for example, without appreciating that the item represents an essential part of the construct’s conceptual domain. It can label a factor convincingly even when that label is only weakly supported by the questionnaire items.

AI-generated questionnaires introduce an additional problem. Researchers can now create 30 or 50 polished Likert statements in seconds. EFA can subsequently organize responses to those statements, but statistical structure does not demonstrate that the original AI-generated item pool adequately represents the theoretical construct. Poor construct definition cannot be repaired merely by sophisticated factor analysis.

AI is therefore most useful as an analytical assistant rather than a methodological decision-maker. It can help researchers compare alternatives, check calculations, inspect code and challenge interpretations. The researcher remains responsible for explaining why EFA was appropriate, why particular analytical decisions were made and why the resulting factors have substantive meaning.

When to Use Exploratory Factor Analysis

EFA may be appropriate when:

• the latent structure of a set of variables is genuinely uncertain;
• a new multi-item measurement scale is being developed;
• an existing scale has been substantially adapted and its structure requires exploration;
• researchers expect observed variables to reflect a smaller number of underlying constructs;
• there is insufficient theoretical or empirical justification to specify a complete CFA model in advance;
• the variables exhibit sufficient shared structure for common-factor modelling;
• the available sample and data are adequate for the intended factor analysis;
• researchers are prepared to interpret the statistical structure in relation to substantive theory.

EFA is less appropriate when the researcher already has a clearly specified measurement model and the actual research question is whether that predetermined structure fits the data. In that situation, CFA may be more appropriate.

Dissertation Example

A dissertation titled “Understanding Employee Resistance to AI-Based Performance Management Systems” develops a 22-item questionnaire from previous literature and preliminary interviews. Although prior studies identify concerns involving fairness, surveillance, autonomy and transparency, no sufficiently established measurement model exists for the particular combination of constructs and population examined in the dissertation.

The methodology chapter therefore justifies EFA as an exploratory procedure for investigating the latent structure of the new scale. Before extraction, the researcher evaluates whether the correlation structure is appropriate for factor analysis and reports the relevant factorability diagnostics. A common-factor extraction method is selected, and factor retention is determined using parallel analysis alongside the scree plot, interpretability and theoretical reasoning rather than relying solely on the eigenvalue-greater-than-one rule.

An oblique rotation is used because the proposed dimensions of employee resistance are theoretically expected to relate to one another. The researcher evaluates factor loadings, cross-loadings and communalities and explains the reasoning behind any item removal rather than presenting deletion as an automatic numerical decision.

The resulting four-factor solution is interpreted using the content of the items and relevant organizational literature. The dissertation explicitly describes the structure as exploratory and identifies independent CFA as an appropriate direction for subsequent validation. Reliability of the resulting scale scores is then examined separately using an appropriate coefficient rather than being treated as established by EFA itself.

Exam Tip

When explaining EFA, do not describe it simply as a method that “reduces many variables into fewer variables.” That formulation obscures the common-factor model and risks confusing EFA with PCA.

A stronger answer explains that EFA investigates whether relationships among observed variables can be represented through underlying latent factors when the factor structure is not sufficiently established beforehand.

Remember the analytical logic:

Factorability → Extraction → Factor Retention → Rotation → Loading Pattern → Interpretation

The most important point is that EFA is not a software command. Extraction, number of factors, rotation and interpretation all require methodological judgment. Factor-retention research, for example, shows why relying automatically on eigenvalues greater than one can be problematic and why methods such as parallel analysis provide stronger evidence. (PubMed Central (PMC))

Build a methodology you can explain and defend

Exploratory factor analysis requires a connected series of decisions rather than a single statistical test. Dudovskiy Research Assistant can help you determine whether EFA fits your dissertation, structure the analytical sequence and explain the methodological reasoning behind your choices.

References

Fabrigar, L.R., Wegener, D.T., MacCallum, R.C. and Strahan, E.J. (1999). Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods, 4(3), 272–299.

Costello, A.B. and Osborne, J.W. (2005). Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Practical Assessment, Research & Evaluation, 10(7).

Horn, J.L. (1965). A rationale and test for the number of factors in factor analysis. Psychometrika, 30, 179–185.

Hayton, J.C., Allen, D.G. and Scarpello, V. (2004). Factor retention decisions in exploratory factor analysis: A tutorial on parallel analysis. Organizational Research Methods, 7(2), 191–205. (Sage Journals)

Brown, T.A. (2015). Confirmatory Factor Analysis for Applied Research. 2nd ed. New York: Guilford Press.