Psychological Methods1999, Vol.4. No. 3.272-299Copyright 1999 by (he American Psychological Association, Inc.1082-989X/99/S3.00Evaluating the Use of Exploratory Factor Analysis inPsychological ResearchLeandre R. FabrigarQueen's UniversityDuane T. WegenerPurdue UniversityRobert C. MacCallumOhio State UniversityErin J. StrahanQueen's UniversityDespite the widespread use of exploratory factor analysis in psychological research,researchers often make questionable decisions when conducting these analyses.This article reviews the major design and analytical decisions that must be madewhen conducting a factor analysis and notes that each of these decisions hasimportant consequences for the obtained results. Recommendations that have beenmade in the methodological literature are discussed. Analyses of 3 existing em-pirical data sets are used to illustrate how questionable decisions in conductingfactor analyses can yield problematic results. The article presents a survey of 2prominent journals that suggests that researchers routinely conduct analyses usingsuch questionable methods. The implications of these practices for psychologicalresearch are discussed, and the reasons for current practices are reviewed.Since its initial development nearly a century ago(Spearman, 1904, 1927), exploratory factor analysis(EFA) has been one of the most widely used statisticalprocedures in psychological research. Despite thislong history and wide application, the use of factoranalysis in psychological research has often beencriticized. Some critics have raised concerns aboutfundamental limitations of factor analysis for contrib-uting to theory development (e.g., Gould, 1981; Hills,1977; Overall, 1964). For instance, Armstrong (1967),in an article entitled "Derivation of theory by meansof factor analysis or Tom Swift and his electric factoranalysis machine," argued that factor analysis hadlimited utility for aiding in the development of theory,because it could not be relied on to provide meaning-ful insights into data.1 He attempted to demonstratethis point by creating artificial data with a knownstructure and then ostensibly showing that EFA failedto accurately represent the structure.2 Other criticshave not challenged the fundamental utility of EFAbut have instead criticized the manner in which it isLeandre R. Fabrigar and Erin J. Strahan, Department ofPsychology, Queen's University, Kingston, Ontario,Canada; Duane T. Wegener, Department of Psychology,Purdue University; Robert C. MacCallum, Department ofPsychology, Ohio State University.Erin J. Strahan is now at Department of Psychology, Uni-versity of Waterloo, Waterloo, Ontario, Canada.This research was supported by a grant from the SocialSciences and Humanities Research Council of Canada andGrant PO1-MH/DA56826 from the National Institutes ofHealth. We thank Stephen G. West for his comments on anearlier version of this article.Correspondence concerning the article should be ad-dressed to Leandre R. Fabrigar, Department of Psychology,Queen's University, Kingston, Ontario, K7L 3N6, Canada.Electronic mail may be sent to [email protected] The name Tom Swift refers to a character in a popularseries of juvenile science fiction novels published in the1960s. In each novel, Tom Swift makes use of futuristicdevices with near-miraculous powers. Armstrong's refer-ence to this character in the title of his article highlightedwhat he regarded as the naive belief by researchers in fun-damental utility of EFA.2 It is important to note that Armstrong (1967) was notthe first person to examine the effectiveness of EFA proce-dures using data sets with a known underlying structure(e.g., see Thurstone, 1947; Cattell & Dickman, 1962; Cattell& Sullivan, 1962; Cattell & Jaspers, 1967). Interestingly, inthese other cases, the authors concluded that appropriateEFA procedures were reasonably effective in revealing theknown underlying structure of the data.272USE OF FACTOR ANALYSIS273sometimes applied (e.g., Comrey, 1978; Ford, Mac-Callum, & Tail, 1986; Gorsuch, 1983; McNemar,1951; Skinner, 1980). In this article, we primarilyaddress the latter issue. That is, we explore the man-ner in which factor analysis is applied in psychologi-cal research and evaluate the soundness of currentpractices. However, we contend that these two issuesare intertwined. The utility of factor analysis fortheory development is dependent on the manner inwhich it is implemented (see Cattell, 1978; Comrey,1978). Furthermore, we suggest that some critics whohave questioned the fundamental value of factoranalysis have not been sufficiently sensitive to thisrelationship.We begin our discussion by reviewing some of themajor methodological decisions that researchers mustmake when conducting a factor analysis. Next, weillustrate with published data sets how poor choiceswhen making these decisions can substantially distortthe results. We then turn our attention to the extent towhich current use of factor analysis reflects soundpractice-. We conclude with discussions of the impli-cations of current factor analytic practices for psycho-logical theory and reasons for the prevalence of cer-tain practices.Methodological Issues in the Implementation ofFactor AnalysisPerhaps more than any other commonly used sta-tistical method, EFA requires a researcher to make anumber of important decisions with respect to how theanalysis is performed (see Finch & West. 1997). Spe-cifically, there are at least five major methodologicalissues that a researcher should consider when con-ducting a factor analysis. First, he or she must decidewhat variables to include in the study and the size andnature of the sample on which the study will be based.Second, a researcher must determine if EFA is themost appropriate form of analysis given the goals ofthe research project. Third, assuming that EFA is ap-propriate, a specific procedure to fit the model to thedata must be selected. Fourth, the researcher mustdecide how many factors should be included in themodel. Finally, it is usually necessary for a researcherto select a method for rotating the initial factor ana-lytic solution to a final solution that can be morereadily interpreted. Each of these decisions can haveimportant consequences for the results obtained (seeArmstrong & Soelberg, 1968; Cattell, 1978; Comrey,1978; Ford et al., 1986; MacCallum, 1983; MacCal-lum, Widaman, Zhang, & Hong, 1999; Velicer &Fava, 1998; Weiss, 1976). To the extent that a re-searcher makes poor