Sometimes we do not have a predefined model in our study, i.e., although we have Q1, Q2, …, etc., we do not know how they should be grouped into different constructs. CFA does not work here because it requires a predefined model. In this case, we use the exploratory factor analysis (EFA) instead to explore how the items should be categorized into the factors.
This is also called dimension reduction. For example, if we could group our five items into two constructs, then we have reduced the dimension of the data from five to two. Hence the name.
Note #1: The dimension reduction in EFA is completely empirical, i.e., it results from the data, but it may not be theoretically sound. That means, while you may group the items as suggested by EFA, such grouping may not make any theoretical sense at all. For this reason, EFA is usually used only in the development stage of a scale or theory when we have items at hand but we do not know how they should be grouped. We then use EFA to make some suggestions (hence the name “exploratory”), so that we can use it as a reference when we consider the theories behind the model. On the other hand, if you already have a model to start with, then you should use CFA instead.
Note #2: Check this out if you want to read further about EFA: https://stats.oarc.ucla.edu/spss/output/factor-analysis/.
Note #3: EFA is not the same technique as PCA (principal components analysis). See here for their differences: https://www.theanalysisfactor.com/the-fundamental-difference-between-principal-component-analysis-and-factor-analysis/.
Analyze -> Dimension Reduction -> Factor: Put all the key variables into the Variables box on the right hand side.
Click Descriptives -> Check KMO and Barlett's test of sphericity -> Continue.
Click Extraction and select the extraction methods. I would usually use Maximum likelihood but you should follow the literature if you want to compare with other studies. In the same dialog under Extract, select "Based on Eigenvalue" and set 1 as the threshold. Click Continue.
Click Rotation and choose the rotation method. I would usually use Direct Oblimin but again you should follow the literature in your own context. Click continue.
The Rotated Component Matrix table shows us how we can group the items into factors.
The % of Variance column under Extraction Sums of Squared Loadings in the Total Variance Explained table shows the percentages of variances explained by each of the factors.
In the KMO and Bartlett's Test table, you should also check that the assumptions of EFA are satisfied. Desirably:
The p-value of Bartlett’s Test of Sphericity should be <0.05.
The overall KMO should be >0.5.
See here for the meanings of these measures.
Analyses -> Factor -> Exploratory Factor Analysis: Put all the key variables on the right hand side.
Under Method, select the extraction and rotation methods. I would usually use Maximum likelihood for Extraction and Oblimin for rotation, but you should follow the literature if you want to compare with other studies.
Under Assumption Checks, select both Bartlett’s test of sphericity and KMO measure of sampling adequacy.
Under Number of Components, select Based on eigenvalue and set 1 as the threshold.
Under Additional Output, select Factor summary.
Note: If you get the error message saying that "This analysis has terminated, likely due to hitting a resource limit" after you change some settings in EFA, undo the change and do it again.
The Factor Loadings table shows us how we can group the items into factors.
The % of Variance column in the Factor Statistics table shows the percentages of variances explained by each of the factors.
In the Assumption Checks tables, you should also check that the assumptions of EFA are satisfied. Desirably:
The p-value of Bartlett’s Test of Sphericity should be <0.05.
The overall KMO should preferrably be >0.8.
See here for the meanings of these measures.