The 8 "feeling" variables have lower intercorrelations than the "rectangle" variables, created from built-in artificial relationships. So they share less common variance.

The term "eigenvalue"--which refers to the discriminatory power of the factor--is hard to describe in words. Normally ranging between 0 and k (the number of variables in the analysis), the eigenvalue reflects the total variance among the k variables accounted for by that factor.

Thus, an eigenvalue of 3.47 for the first factor, divided by 8 (variables in this model) = 43.4% of the variance.

 Factor Eigenvalue % of Variance Cumulative % 1 3.47 43.4 43.4 2 1.54 19.3 62.7 3 1.14 14.2 76.9

So for this analysis, three factors accounted for only 77% of the variance among these 8 real "feeling" variables--compared with 98.5% accounted by just two factors for the artificial data.

Attention is typically given only to factors with eigenvalues above 1.0. Factors with lower eigenvalues are regarded as analyzing error variance.

Go to the next page, # 9: Unrotated and rotated factors from the factor analysis