Third, there is also recognition among selection researchers that multilevel
theorizing and research must be motivated by levels of conceptualization and
construct operationalization that may involve more than simply aggregating indi-
vidual-level data. We turn to this speciWc issue in a later section.
- 2 Reconceptualization of the Performance Domain
The meaning of job performance has changed throughout the decades, but it has
changed radically in recent years. Austin and Villanova ( 1992 ) chronicled the early
history of job performance concepts, operational deWnitions, and measures from
1917 – 92 and noted that a major limitation of the research conducted prior to 1990
was thecriterion problem. Job performance was often treated too narrowly (i.e.
deWcient) and sometimes inappropriately when applied to a particular context (i.e.
contaminated), with a large focus on the technical aspects of a person’s production
or service delivery. However, a conXuence of events beginning in the late 1980 s,
including drastic changes to the nature of work in organizations and societal
pressures toward fair evaluations of individual eVectiveness, fostered the growth
of new performance theories that have added structure to an expanding criterion
domain.
Largely consistent with emerging theories of job performance (e.g. Borman et al.
1983 ; Brief and Motowidlo 1986 ; Organ 1988 ), Campbell ( 1990 ; Campbell et al. 1996 )
proposed a comprehensive performance model consisting of eight work behavior
categories, with (job-speciWc and non-job-speciWc) task proWciency and commu-
nication task proWciency central to all jobs. He strongly emphasized deWning
performance in terms of behavior, but limited the domain to behaviors that are
relevant to organizational goals. Empirical work from Project A (see Campbell et al.
1990 ; McCloy et al. 1994 ; McHenry et al. 1990 ) supported aspects of the Campbell
model, as well as related models like Borman and Motowidlo’s ( 1993 ) task and
contextual performance notions (e.g. Borman et al. 1995 ; Van Scotter et al. 2000 ).
At the same time, the dimensions proposed by diVerent models have received
varying degrees of support (see reviews by Coleman and Borman 2000 ; Motowidlo
2003 ; Viswesvaran and Ones 2000 ). Viswesvaran et al. ( 2005 ) recently argued that a
single, construct-level factor accounts for 60. 3 percent of the variance in perform-
ance ratings across the 303 studies in their meta-analysis, but also acknowledge that
certain assumptions about true score and error underlie their interpretations.
Additionally, the majority of the studies in the Viswesvaran meta-analysis were
not guided by the recent conceptual deWnitions of performance and often did not
include measures of the dimensions suggested by Campbell and others.
Part of the impetus for research on non-task performance behaviors, particularly
‘citizenship behaviors’ (LePine et al. 2002 ; Organ 1997 ; Rotundo and Sackett 2002 ),
stems from the adverse impact created against racial minorities by cognitive ability
selection decision-making 305