Fr. Lawrence et Gr. Hancock, Conditions affecting integrity of a factor solution under varying degrees of overextraction, EDUC PSYC M, 59(4), 1999, pp. 549-579
Simulated data were employed to test the integrity of orthogonal factor sol
utions when varying sample size, factor pattern/structure coefficient magni
tude, method of extraction (principal component vs. common factor), number
of variables, number of factors, and degree of overextraction. Factor solut
ions' integrity was operationalized by comparing sample factor/component pa
tterns to known population factor patterns, from which measures of nonzero
coefficient bias and variability, and zero coefficient bias and variability
, were determined. Slopes of each of these measures across degrees of overe
xtraction were determined and related to the experimental condition variabl
es. Results showed that nonzero pattern/structure coefficients generally de
grade with increased overextraction, with PCA coefficients exhibiting this
property to a greater degree than CFA coefficients. With smaller sample siz
e and small coefficient magnitude, the degradation was greater; with higher
sample size and larger coefficients, coefficient degradation was usually n
ot enough to yield incorrect factor/component structure interpretations.