BLUCK ET AL.
principal components analysis (PCA) because PCA originally was developed to produce a strong first, general (principal) factor. Since we expected three factors (not a general factor), we used EFA, which allows the variance to be better distributed across multiple factors. We also used EFA because PCA attempts to explain all of the vari- ance in a set of items (including unique and error variance), rather than only the reliable, common, shared variance among a set of items (Gorsuch, 1983).
Factors were extracted using a common factors model (principal axis) with a Promax (oblique) rotation in SPSS Version 11.0. We used an oblique rotation rather than the commonly used Varimax or- thogonal rotation procedure for two reasons. First, we had no reason to believe that the factors would be orthogonal. Previous discussions about the three theoretical functions of AM suggest that they may be intercorrelated (Bluck, 2003; Conway, 2003). In fact, it has been ar- gued that since most social science data involves correlated con- structs, oblique rotation procedures are preferable (Nesselroade, personal communication). Our inspection of the simple correlation matrix confirmed that, indeed, the items are intercorrelated. The sec- ond reason for using an oblique rotation procedure, particularly Promax, is that it produces a more simple structure than orthogonal or other oblique rotation methods, thereby increasing the ease of interpretation (Hendrickson & White, 1964).
In the initial EFA, Kaiser’s rule (Kaiser, 1960) of extracting factors with Eigen values > 1 suggested seven factors. Examination of the in- flection point on the scree plot suggested that six factors should be extracted (Cattell, 1966). These two solutions accounted for 63.12% and 59.25% of the variance, respectively. In addition, based on sam- ple size criteria delineated by Cliff and Hamburger (1967), a factor loading of .40 was used to identify meaningful factor loadings. Upon examining the factor pattern matrix of these solutions, not all factors were meaningful—two factors had only one or two item loadings greater than .40. Thus, a combination of Eigen value, scree inspec- tion, and loading criteria suggested either a four- or five-factor solution.
Next, because theory suggested the presence of three latent con- structs underlying responses to the TALE, we reran the EFA and forced three factors. The 3–factor solution, however, accounted for less than half the variance (44.15%). This solution was not ideal in that eight items did not load on any factor. Moreover, while the sec-