Because there have been so many questions here about developing scales that use Likert-scored items, I have compiled a set of resources on this topic -- including several that discuss EFA.
Do you mean exploratory or explanatory? The latter would usually be referred to as confirmatory in factor analytic contexts and basically amounts to structural equation modelling (SEM). So any answer as well as any useful orientation in David Morgan's very helpful compilation will depend on that.
Factor analysis is done to further refine the scale and reduce the set of observed variables to smaller numbers. One critical assumption underlying the appropriateness of factor analysis is to ensure that the data matrix has sufficient correlations to justify its application (Hair et al., 1995). The next step involves assessing the overall significance of the correlation matrix with Bartlett test of sphericity, which provides the statistical probability that the correlation matrix has significant correlations among at least some of the variables. The decision to include a variable in a factor should be based on factor loading greater than ±0.3 (Hair et al., 1995), and all factors whose eigenvalues are greater than 1.0 should be retained in the factor solution (Tabachnick and Fidell, 1989). The choice regarding factor loading greater than ±0.3 should not be based on any mathematical proposition but should relates more to practical significance.
Thank you, Dr. Morgan, Dr. Horvath and Dr. Abbas for the insights. Dr. Horvath, you are right. By mistake, I put it explanatory, it would be exploratory. Thank you for the correction.
in this case, I think it is the best idea to take some time and look at the post/thread provided by David Morgan.
The whole idea of factor analysis for scale construction is rather simple: (i) to make sure that the items measure the same thing (resp. how many different things the items measure) and (ii) to evaluate how well they do their job. For item selection, the usual strategy will be to exclude items that either have low factor loadings (e.g.