EFA vs CFA SPSS – Which Factor Analysis Does Your Dissertation Actually Need

EFA vs CFA SPSS confusion is, hands down, the question I get asked most by dissertation students, and I have watched it cost people months of rework when they get it wrong.

Here is the short version. If you are developing a new scale or questionnaire with no fixed structure in the literature, run EFA. If you are testing a scale that other researchers have already validated, run CFA directly.

And here is the part most guides mention only in passing, almost as an afterthought – SPSS alone can only run EFA. CFA needs AMOS, R, or SmartPLS.

	EFA	CFA
Research stage	Early, scale development	Later, scale validation
Purpose	Discover an unknown structure	Test a known structure
Runs in plain SPSS	Yes	No, needs AMOS, R, or SmartPLS

Exploratory vs Confirmatory Factor Analysis – The Core Difference

These two methods sit at opposite ends of the same question. EFA lets the data tell you what structure exists. CFA makes you declare the structure first, then checks if your data agrees with it.

What EFA Actually Does to Your Data

EFA (Exploratory Factor Analysis) is a data-driven technique that identifies how many underlying factors exist in a set of items and which items belong together, without assuming the structure in advance. Say you have written 20 new items for a “workplace trust” scale. You genuinely do not know yet if trust splits into two dimensions or five.

Before extracting anything, SPSS also runs a KMO and Bartlett’s test of sphericity, and most committees expect KMO above 0.60 with a significant Bartlett’s result before they will accept any EFA output at all. EFA looks at how your items correlate with each other and groups them into factors using tools like the scree plot and parallel analysis, without you telling it what the answer should be in advance. As a quick rule, most dissertations extract factors with an eigenvalue greater than 1, though parallel analysis is considered the stronger method, and items are usually retained only if their factor loading is 0.40 or above, with 0.30 as the bare minimum many committees still accept. Items that load above 0.30 on more than one factor at the same time, known as cross-loadings, are a common reason to drop or reword a question.

Most textbooks praise EFA for its flexibility, but they rarely admit how much the number of factors can shift just because you switched rotation method from an orthogonal rotation like varimax to an oblique rotation like oblimin, the two broad rotation families every EFA eventually has to choose between. In my own practice, I tell every student to run both rotations and check if the conclusion survives. If it does not, that instability belongs in your limitations section, not hidden from your reader.

What CFA Actually Does to Your Data

CFA (Confirmatory Factor Analysis) is a theory-driven technique that tests whether a specific, pre-defined factor structure actually fits your observed data, the standard definition used across the wider confirmatory factor analysis literature. CFA runs the opposite way to EFA. You already have a theory, either from a validated scale, prior research, or an EFA on a pilot sample, and you test whether your data fits that exact structure. If you are using the Utrecht Work Engagement Scale, for example, its three factor structure is already fixed in the literature.

You do not rediscover it, you confirm it, by checking fit statistics like CFI, TLI, RMSEA, and SRMR against established thresholds. If the fit is poor, standard practice is to inspect modification indices before concluding the scale fails outright, though any change must be justified by theory, not just by what improves the numbers. A second-order CFA may also be worth considering if your construct itself has recognised sub-dimensions, though that is usually a decision for a later stage of your analysis.

Why SPSS Alone Cannot Run CFA

SPSS’s factor analysis module, found under Analyze, Dimension Reduction, Factor, only performs EFA. For CFA you need AMOS (IBM’s own add-on, often bundled with university SPSS licences), R with the lavaan package, or SmartPLS. This single fact answers most of the EFA vs CFA SPSS software questions that land in my inbox every week.

Most articles on this topic mention the software gap almost as a footnote, buried deep after a long comparison table. I think that placement is backwards. For a student halfway through their analysis, discovering that an entirely different piece of software is now required is the single biggest reason factor analysis chapters get delayed.

Decisions about which correlation matrix to run EFA on also matter here, and this is the same choice that comes up when picking between Pearson and Spearman correlation for your item data. A peer-reviewed walkthrough of these same software and matrix decisions, with SPSS and LISREL syntax shown side by side, is available in this open-access factor analysis guideline, if you want the fuller technical picture beyond what any single blog post can cover.

EFA vs CFA SPSS – Full Comparison by Stage and Purpose

The table below is the one I wish someone had shown me early in my own research career. It lines up EFA and CFA side by side, by research stage, by purpose, and by what SPSS can and cannot do for you.

Criterion	EFA	CFA
Research stage	Early, scale development	Later, scale validation or SEM prep
Purpose	Discover an unknown factor structure	Confirm a specified factor structure
Theory needed beforehand	Little to none	Strong, from literature or a prior EFA
Number of factors	Decided by the data	Fixed in advance by the researcher
Rotation	Required (varimax, oblimin, promax)	Not applicable
Fit statistics	Not formally evaluated	CFI, TLI, RMSEA, SRMR, chi-square
SPSS support	Yes, Analyze, Dimension Reduction, Factor	No, needs AMOS, R (lavaan), or SmartPLS
Typical sample size	Around 5 to 10 respondents per item, 200 or more overall	Similar baseline, though model complexity can push it higher
Common dissertation use	New or heavily adapted questionnaire	Established scale applied to a new sample or population

By Research Stage

EFA belongs in the early, messy part of your research, when you are still figuring out what your items are actually measuring. CFA belongs later, once a structure already exists in the literature, or once your own EFA has given you one to test on a fresh sample.

By Purpose and Prior Theory

This is the distinction that trips up the most students. EFA assumes you do not know the answer yet. CFA assumes you do, and your whole job is to check if the data agrees with you. If you walk into your analysis already knowing how many factors you expect and which items belong where, you have no business running EFA.

You are in CFA territory.

By Software and Output

Planning your sample size correctly matters for both methods, and it is worth working this out before data collection starts, not after. A proper sample size calculator built for factor analysis will save you from collecting too few responses and finding out only at the analysis stage.

When to Use CFA Instead of EFA

Is Your Scale Established or Brand New

Settling the EFA vs CFA SPSS question for your own dissertation starts here, with your literature review, not your software menu. The honest test is simple. If 70 to 80 percent of your items are recognisably the same as a scale already published and validated by someone else, your committee almost always expects CFA, even if you have added two or three contextual items of your own. A properly validated scale usually comes with more than a published factor structure alone.

It typically also carries acceptable reliability, commonly reported as Cronbach’s alpha above 0.70, along with evidence of convergent and discriminant construct validity, so check for all of this in the original source, not just the factor count.

What Your Literature Review Already Tells You

Your literature review has usually already answered this question for you. You just have not noticed. If you spent three pages describing a scale’s prior factor structure from earlier studies, you have implicitly committed yourself to CFA. Going back and running EFA afterwards contradicts your own writing.

Questions to Ask Before You Run Anything

1. Check your literature review first. Has the scale you are using already been validated in a peer reviewed study with a clear, published factor structure? If yes, lean towards CFA.
2. Ask if you wrote the items yourself or adapted them heavily for your population or industry. If this exact set of items has never been tested together before, you need EFA first, as part of a proper survey research design.
3. Check your own department’s track record. If your committee has a history of accepting only CFA for established instruments, do not fight that norm with EFA just because EFA feels easier to run.
4. Decide your sample split before you collect any data, not after. If you plan EFA followed by CFA, you need two independent samples, not one sample analysed twice. This two-stage sequence is recognised best practice for developing a new scale, not needless repetition, so present it in your write-up as a deliberate methodological choice rather than an afterthought.
5. When genuinely in doubt, ask your supervisor directly, in writing, before you run anything. A single line such as “Based on Scale X being established in the literature, I plan to proceed with CFA, please confirm” can save you weeks of rework later.

A Real Case Study – Wrong Method, Right Fix

I will call her Ananya. She was finishing an MBA dissertation on customer loyalty in retail, and she had adapted a brand loyalty scale originally published in 2014, keeping roughly 80 percent of the original items and tweaking the wording for her retail context.

What Went Wrong

Ananya ran EFA, mostly out of habit, because that is the template most of her batchmates had followed the previous year. Her internal review flagged it immediately. Since she had retained the large majority of a previously validated scale, her committee considered it established, and they expected CFA, not an exploratory run.

How It Was Fixed

She did not need to restart data collection. The dataset she already had was perfectly usable. She simply had not applied the right test to it.

She ran CFA in R using the lavaan package on her full sample, since this was a validation exercise rather than a two-stage EFA-then-CFA sequence, so using one sample here was not the same mistake as running both methods on the same data. She rewrote her methodology narrative, added a proper fit indices table with CFI, TLI, RMSEA, and SRMR, and her committee approved it with only minor revisions.

The Lesson

Most dissertation-help articles frame a story like this as a disaster. I do not think it is one. Ananya lost about ten days reworking her analysis and rewriting a few paragraphs, not ten months, because she caught the problem before submission, not during her viva.

The pattern to remember is simple. If your scale is 70 to 80 percent recognisably the same as something already validated, your committee will expect CFA almost every time, even with a handful of added items. If you are even slightly unsure which side of that line your own scale falls on, it is worth working out whether you need a second pair of eyes on your methodology chapter before you submit it, not after.

Common Factor Analysis Dissertation Mistakes

Running EFA and CFA on the Same Sample

This is the mistake that methodology papers call out most often, and for good reason. It artificially inflates how well your model fits. Academic articles repeat this warning constantly, yet most of them stop short of telling you what to actually do about it.

My own advice is practical. Split your sample roughly 50/50 before you collect data, or gather two separate waves, and check whether your split remains defensible given your sampling approach.

Picking CFA for a Scale That Is Not Actually Established

Some methodologists even have a name for what happens when CFA is forced onto a structure too early. They call it scale destruction, because the result you get bears no resemblance to the real factor structure underneath your data. If your scale has not genuinely been validated elsewhere, forcing CFA on it does not make your research look more rigorous. It just produces a poor fit for the wrong reasons.

Ignoring Poor Model Fit Instead of Reporting It

A poor fitting CFA model is not something to quietly drop from your write-up or fix by deleting items without disclosure. Report the fit honestly, explain any modifications you made and why, and let your committee see the full picture. Before any of this, it is worth checking your data for basic assumption violations, including multicollinearity, using something like a VIF check, because a messy correlation matrix will distort both EFA and CFA results regardless of which one you choose.

How to Justify Your Method Choice in Writing

Your methodology chapter should state your choice plainly and defend it in a sentence or two, not bury it, following the same APA reporting conventions your committee already expects elsewhere in your dissertation. Here are three templates you can adapt, not copy word for word, since your own citations and sample sizes need to go in.

• “Because the Job Satisfaction Scale (Smith, 2012) has an established four factor structure, confirmatory factor analysis was used to test whether this structure held in the present sample (N = 250).”
• “As no prior validated measure existed for this construct within this population, exploratory factor analysis was conducted to identify the underlying factor structure before proceeding to further analysis.”
• “Exploratory and confirmatory factor analyses were conducted on independent subsamples (n1 = 150, n2 = 150) to first identify and then confirm the factor structure of the newly developed scale.”

Whichever template fits your situation, pair it with your actual citation, your actual sample size, and a one line reason tied to your own literature review.

FAQ

About the Dissertation Expert Behind This Guide

I have spent the last 12 years guiding PhD and Master’s students through exactly this kind of methodology decision, across psychology, business, and education dissertation programmes. EFA vs CFA SPSS questions land in my inbox every single week, usually from someone who is one supervisor meeting away from submission. You can find more of my work, or get in touch directly, through my LinkedIn profile.