issue that further increases the inherent complexities of measuring internalWt. The
basic assumptions of SHRM are based on multilevel theory (Kozlowski and Klein
2000 ; OstroVand Bowen 2000 ). The SHRM paradigm holds that HRM systems
aVect the characteristics (e.g. attitudes, KSAs, motivation, and empowerment) of
organizational actors (individual level) and groups of actors (group level). These
characteristics, in turn, aVect workforce productivity and organizational perform-
ance (plant and/or organizational level). Inherent in this perspective are cross-level
and multilevel models (Kozlowski and Klein 2000 ). This further complicates the
design and measurement of HRM systems. Even if HRM managers use valid
selection tools to select the best possible applicants, there is virtually no research
that speciWcally establishes a link between selection techniques at the individual
level and group or organizational performance (Ployhart 2004 ). Similar observa-
tions can be made for virtually every other HRM activity area. There is only very
limited research that uses data from multiple levels of analysis to investigate these
complex relationships.
For researchers, besides the obvious data collection problems, there are theor-
etical and methodological problems and challenges of this multilevel perspective
(Kozlowski and Klein 2000 ). Besides some well-known measurement problems
(e.g. Gerhart in press; Gerhart et al. 2000 ), simple abstractions from lower-level to
higher-level relationships and vice versa are likely to be inappropriate. Combining
the issues of levels of abstraction and analysis, we can make the predictions that the
HRM philosophy may aVect an organization, that the HRM policies have the
potential to aVect large parts of the workforce (all employees who are covered by
the speciWc policies), and that HRM practices and processes may aVect speciWc
work groups, those that the practices and processes target. This has important
implications for SHRM research. Collecting data on HRM practices that cover the
entire workforce may be inappropriate since individual practices are unlikely to
target an entire workforce (Delery and Shaw 2001 ; Kepes and Delery in press; Lepak
and Snell 1999 ). HRM measures must match the level on which outcome measures
are collected. Measuring HRM practices (level of abstraction) and organizational
performance (level of analysis), for example, may denote a misalignment and the
empirical results could be a methodological artifact rather than a true relationship
(Kozlowski and Klein 2000 ).
There is a critical need in this research area to match levels of abstraction and
levels of analysis. Theory should guide the appropriate determination of the
diVerent levels. Only then can the eVects of internally consistent or inconsistent
HRM systems on individual, group, or organizational eVectiveness be understood.
Recently, OstroVand Bowen ( 2000 ) proposed a multilevel framework that links
HRM systems toWrm eVectiveness (see also Bowen and OstroV 2004). Their model
illustrates the need to take a multilevel perspective when investigating HRM
systems and theWt between the diVerent elements within it.
hrm systems and the problem of internal fit 395