Recently I've been conducting factor analyses in order to reduce my data to more usable bits. I've been reviewing articles and books that describe FA and recently reread Costello and Osborne's article in Practical Assessment, Research, and Evaluation ( ) regarding best practices. Their article suggests the following four best practices with respect to FA, many of which go against the default options in statistical software packages such as SPSS!
- Use factor analysis instead of principal components analysis in order to ensure that variance is not inflated. They note that principal components analysis, the most commonly used method and default in SPSS, is not a true factor analysis method but only used for data reduction.
- Use maximum likelihood (if your data is normally distributed) or principal axis factors (if your data are not normal) as the factor extraction method. As the default is principal components analysis which we are recommended NOT to use, use of factor analysis means we have many factor extraction methods to choose from. The above have been shown to work with all types of data.
- Use the scree test to determine the number of factors that should be extracted. This is easily selected in SPSS. In cases where the elbow is hard to assess, they suggest simply running multiple factor analyses and setting the number of factors to retain manually – once at the projected number based on the a priori factor structure, again at the number of factors suggested by the scree test if it is different from the predicted number, and then at numbers above and below those numbers.
- Use oblique rotation as opposed to orthogonal rotations. They make this suggestion because orthogonal rotations produce factors that are uncorrelated and oblique methods allow the factors to be correlated.
Once I've followed these best practices, I am always left wondering how to best utilize the variables in each factor to produce a factor estimate or score. That will be the topic of my next post!
