Journal of Applied Psychology1993, Vol. 78, No. 2, 218-225Copyright 1993 by the American Psychological Association, Inc.002I-9010/93/$3.00Nature and Consequences of Halo Error: A Critical AnalysisKevin R. Murphy, Robert A. Jako, and Rebecca L. AnhaltThe definition of halo error that dominated researchers' thinking for most of this century impliedthat (a) halo error was common; (b) it was a rater error, with true and illusory components; (c) it ledto inflated correlations among rating dimensions and was due to the influence of a general evalua-tion on specific judgments; and (d) it had negative consequences and should be avoided or removed.We review research showing that all of the major elements of this conception of halo are eitherwrong or problematic. Because of unresolved confounds of true and illusory halo and the oftenunclear consequences of halo errors, we suggest a moratorium on the use of halo indices as depen-dent measures in applied research. We suggest specific directions for future research on halo thattake into account the context in which judgments are formed and ratings are obtained and thatmore clearly distinguish between actual halo errors and the apparent halo effect.When an individual is rated on multiple performance di-mensions or attributes, the rater's overall impression or evalua-tion is thought to strongly influence ratings of specific attri-butes (Cooper, 1981 b), a phenomenon that is referred to as haloerror (Thorndike, 1920). Discussions of halo error are mostfrequently encountered in the context of evaluative judgment(e.g., in interviews and performance appraisals), but similar phe-nomena have been noted in research on illusory correlation(Chapman & Chapman, 1969), implicit personality theory (Lay& Jackson, 1969), and interpersonal judgments (Nisbett & Wil-son, 1977).Research on halo errors in rating can be traced back to theearly part of this century (Thorndike, 1920; Wells, 1907). Al-though there are a number of different conceptual and opera-tional definitions of halo (Balzer & Sulsky, 1992; Saal, Downey,& Lahey, 1980), throughout most of the history of research onhalo error, there has been some consensus regarding the natureand consequences of halo error. First, halo error is thought tobe common (Bernardin & Beatty, 1984; Blum & Naylor, 1968;Cascio, 1991; Feldman, 1986; Jacobs & Kozlowski, 1985). Inthis vein, Cooper (198 Ib) referred to the halo error as "ubiqui-tous." Second, halo is typically treated as a rater error. That is,halo error is assumed to reflect the rater's inability (Cooper,198 la, 198 Ib; Feldman, 1986; Lance & Woehr, 1986; Murphy &Jako, 1989; Nathan & Lord, 1983; Saal et al., 1980; Vance,Winne, & Wright, 1983) or unwillingness (Banks & Murphy,1985) to discriminate among different attributes of the individ-uals rated.Kevin R. Murphy, Robert A. Jako, and Rebecca L. Anhalt, Depart-ment of Psychology, Colorado State University.This article was written, in part, during sabbatical visits by Kevin R.Murphy to the University of California, Berkeley, and to the NavyPersonnel Research and Development Center. The generous support ofboth institutions is appreciated.We are also grateful to George Thornton, Bill Balzer, and two anony-mous reviewers for their many comments and suggestions.Correspondence concerning this article should be addressed to Ke-vin R. Murphy, Department of Psychology, Colorado State University,Fort Collins, Colorado 80523.Third, observed halo (often operationalized in terms of theaverage intercorrelation among ratings) is thought to reflecttwo distinct influences, true halo and illusory halo; the labelsvalid halo and invalid halo are also sometimes used. (Bartlett,1983; Bingham, 1939; Cooper, 1981b; Lance & Woehr, 1986;Murphy, 1982; Pulakos, Schmitt, & Ostroff, 1986). That is, ob-served halo reflects both the real overlap among the dimen-sions being rated (true halo) and a host of irrelevant factors,including deficiencies in measurement, observation and mem-ory errors on the part of the rater, incorrect theories of covaria-tion, and so on, that taken together constitute illusory halo.This formulation is analogous to the statement in classical mea-surement theory that observed scores equal true scores pluserrors in measurement (Lord & Novick, 1968). As in classicmeasurement theory, the goal of much of the existing researchon halo has been to understand and eliminate the error compo-nent, which in this case is referred to as illusory halo.Fourth, halo error leads to inflated correlations among theattributes rated (Bernardin & Beatty, 1984; Cascio, 1991;Cooper, 1981b; Lance & Woehr, 1986; McCormick & Ilgen,1985; Nathan & Lord, 1983; Pulakos et al., 1986); that is, theobserved correlations among ratings are usually assumed to belarger than the corresponding true intercorrelations.Fifth, the most probable cause of halo is that the rater's over-all evaluation of the ratee influences his or her ratings of spe-cific attributes (Bernardin & Beatty, 1984; Cooper, 198 Ib; Feld-man, 1986; Fisicaro & Lance, 1990; Landy, 1989; Muchinsky,1987; Murphy, 1982). Thorndike (1920), who coined the termhalo error, defined it as "suffusing ratings of special featureswith a halo belonging to the individual as a whole" (p. 25).Cognitively oriented research has examined a number of alter-native ways in which global evaluations or evaluative schemamight affect halo errors (e.g., Feldman, 1986; Nathan & Lord,1983) but, with few exceptions (e.g., Lance & Woehr, 1986), theassumption has been that the direction of the effect is top-down, in other words, that general evaluations affect specificratings.Sixth, halo error is thought to have a negative impact on thequality of ratings (Cooper, 198 Ib; Landy, Vance, Barnes-Farrell,218NATURE AND CONSEQUENCES OF HALO219& Steele, 1980; Kinicki, Bannister, Horn, & DeNisi, 1985). Al-though it is generally accepted that halo errors do not necessar-ily indicate inaccuracy in rating (Cooper, 1981b; Murphy &Balzer, 1989; Nathan & Tippins, 1990; Vance, Winne, &Wright, 1983), halo errors are nevertheless thought to reducethe usefulness of ratings for a variety of purposes, especiallyfeedback (Saal & Knight, 1988).Seventh, it is often assumed that removing halo error is bothplausible and worthwhile (Bartlett, 1983; Bernardin & Beatty,1984; Cooper, 198 Ib; Holzbach, 1978; Kenny & Berman, 1980;Landy et al, 1980; Myers, 1965). A variety of approaches toremoving halo, including statistical control, training, increas-ing