Several times in the previous chapter we mentioned the possibility of retaining only the first few PCs-those accounting for a major percentage of the variation in the original system of variables-as a more economical description of that system. The PCs having low associated eigenvalues are likely to be describing error variance, anyway, or to be representing influences (causal factors?) that affect only one or a very few of the variables in the system. However, the basic PCA model leaves no room for error variance or for nonshared (specific) variance. Models that do provide explicitly for a separation of shared and unique variance lead to the statistical techniques known collectively as factor analysis (FA).