Exploratory Factor Analysis (EFA)

Factor analysis refers to a family of statistical techniques that range from confirmatory techniques to pure exploratory procedures. The objective is to explain the variability among a larger number of observed random variables in terms of a smaller number of unobserved random variables or factors. Spearman (1904) pioneered the technique when he attempted to explain whether a general intelligence factor underlies individual performance on tests. Spearman’s belief in a unifactor solution led the path to multifactor solutions, but the purpose of factor analysis has remained the same (Bollen, 1989). Factor analysis may be thought of as a data reduction technique with two major approaches, exploratory (EFA) and confirmatory (CFA).

Exploratory factor analysis commences with no specified à priori model. It is a useful technique for theory building because the number of latent variables is not predetermined, all latent variables typically influence all observed variables, measurement errors are presumed to be uncorrelated, and underidentification of parameters in the model is common.

Confirmatory factor analysis utilizes a model constructed in advance, the number and presumed effect of latent variables has been predetermined, the measurement errors may correlate and parameter identification is a requirement. The goal is to determine whether or not measures created to represent a latent variable really belong together. In substantive areas where little is know, EFA can prove to be valuable approach and can suggest underlying patterns in the data. If, however, hypotheses about plausible model structures exist, then exploratory factor analysis can frustrate attempts to test these ideas.

References

Bollen, K.A. (1989). Structural Equations with Latent Variables. New York: John Wiley & Sons.

Spearman, C. (1904). General intelligence, objectively determined and measured. American Journal of Psychology, 15:201-293.