Exploratory Factor Analysis (EFA) in SPSS

This step-by-step tutorial will walk you through doing an exploratory factor analysis (EFA) in SPSS to come-up with a clean pattern matrix to be used in confirmatory factor analysis (CFA) part of structural equation modeling (SEM) in SPSS-AMOS. It uses the maximum likelihood extraction as it is the algorithm used in AMOS. This has two parts, which are: (A) Setting up the Analysis, and (B) Interpreting the Outputs.

A. Setting up the Analysis

Dataset to use: No missing values dataset

1. Go to Analyze — Dimension reduction — Factor
2. Throw all reflective measures (from latent factors in the model)
3. In the Descriptives tab, set the following:
• Statistics: Leave at default (Initial solution); and
• Correlation matrix: Reproduced and KMO and Bartlett's test of sphericity
4. In the Extraction tab, set the following:
• Method: Maximum likelihood;
• Display: Unrotated factor solution;
• Extract: Based on Eigenvalue -- greater than 1
• Initially, you can do this to see if your expected number of factors will show up.
• If otherwise, you can choose fix number of factors and specify the number of factors you expect based on your hypothesized model.
• Maximum iterations: Leave it at default, which is 25.
5. In the Rotation tab, set the following:
• Method: Promax, Kappa 4;
• Display: Rotated solution; and
• Maximum iterations: Leave it at default, which is 25.
• (Skip the Scores tab. No need to configure.)
6. In the Options tab, select the following:
• Missing values: Leave at default (Exclude cases listwise)
• Suppress small coefficients
• Set the absolute value below at 0.30

Steps in Exploratory Factor Analysis in SPSS

B. Interpreting the Outputs

1. In the KMO and Barlett's Test:
• Values above 0.80 are good; less than 0.60 is questionable.
• Interpretation:
• This test determines if the variables are correlated or not.
• Low KMO means you don't have highly correlated indicators.
• It means that you have more formative factors.
• Each item represents a separate dimension of that factor which may or not be correlated.
2. In the Communalities:
• Look at the extraction column.
• Higher values are better. There should be no values of less than 0.40.
• Interpretation:
• This determines the adequacy of the correlations between variables.
• Low values mean that variables may struggle to load significantly on any factor.
• Variables with low values may be removed after you examine the pattern matrix.
3. In the Total Variance Explained,
• Look at the cumulative % under the Extraction Sums main column.
• A value above 50% is good; above 60% is better.
• Interpretation:
• This shows the cumulative variance explained.
4. Ignore the Factor Matrix.
5. In the Goodness of fit test:
• You can also ignore this one, but:
• The general rule is that Chi-square/df = somewhere between 1.00 and 3.00
6. In the Redundancy or Reproduced Correlations table:
• Ignore the table.
• Look at the bottom and see the % non-redundant residuals.
• Less than 5% us good.
• Interpretation:
• This tells as to what extent is the variance explained by error.
7. In the Pattern matrix:
• Items must be loading on all factors.
• Loadings should be above 0.50 at every item and averaging out above 0.70 at every construct.
• Loadings less than 0.50 are not desirable.
• Interpretation:
• This tells how every item loads into groups.
• This is a measure of convergence.

In summary, conducting an EFA is important before doing a CFA in SPSS-AMOS. Its ultimate goal is to come up with a clean pattern matrix where acceptable values of KMO, communalities, factor loadings, and factor correlation matrix, etc., are satisfied. This avoids data reliability and validity concerns in the CFA and actual SEM. Hope this helps.

Big thanks to Dr. James Gaskin for helping me learn on this topic. You may check this YouTube video SEM Series (2016) 3. Exploratory Factor Analysis (EFA), where I based these steps together with other videos on his YouTube Channel.