Abstract
Studies in factorial ecology have typically used the principal factor procedure coupled with varimax rotation. Since it can be shown that the results one obtains vary according to the factor and rotation models he employs, and since there is no one “best” way of obtaining initial and derived factor solutions, it is proposed that future research in the area adopt an approach involving the simultaneous use of several different computing algorithms for obtaining initial solutions and both orthogonal and oblique rotation procedures to avoid the possibility that the results one obtains are not method-dependent. Ideally, the factor models employed should differ maximally with respect to the principles upon which they are based. If one finds a given factor regardless of the method he uses, only then can he assert with any confidencethat it is not an artifact of his method. Factorial ecologists are often interested in computing “factor scores”, but “true” factor scores are not uniquely computable; they can only be estimated. Since the proposed research strategy involves using either a component or an image model, in which the corresponding scores are exact and uniquely computable, the factor score problem is, in a sense, solved. Next it is suggested that, by using orthogonal solutions, factorial ecologists may be overlooking a very important piece of information—the correlation between the factors. There is some reason to believe that this varies from city to city, and may account for the fact that some factors which emerge in studies of Western cities are sometimes not found in cities elsewhere. Some comments are also made on the value of using census tract data, and on the availability of computer programs for different initial and derived factor solutions.