One standard deviation increase in HR system practices (relative to the mean) designed to
enhance workforce ability, motivation and opportunity to contribute was associated with
roughly 20 % betterWrmWnancial performance. Consider that thisWnding means thatWrms
one standard deviation above the mean are 120 % of mean performance, while those one
standard deviation below the mean are at 80 % of mean performance, making for a 120 / 80 ¼
50 % advantage of beingþ 1 standard deviation versus 1 standard deviation. This is a large
difference.
Gerhart ( 2005 : 177 ) goes on to rather spoil the story by suggesting that the
‘eVect size is so large as to perhaps not be credible.’ Large-scale databases of multiple
numbers ofWrms or branches of large organizations have been able to establish repre-
sentative data showing the nature of this relationship. ‘Sophisticated’ HRM, often
referred to as High-Performance Work Systems (HPWSs) or High-Commitment
Management (HCM), has been associated with better performance.^1 What, of
course, is not at all clear is the direction of the relationship. Cross-sectional research
can only reveal associations, not causality, and it is equally plausible that excellent
Wrms will both be able to aVord sophisticated HR systems and wish to invest in them.
The methodological downside to this type of research, which some say is fatally
Xawed (Wall and Wood 2005 ), can be summarized in three ways. First, respondents
may have incomplete knowledge, for example of how many employees are covered
by a particular practice, especially if the respondent is located at the corporate
oYce of aWrm with numerous business units. There is also the diYculty of
assuming that HR practices are translated into actual practices, as discussed later.
If multiple respondents are used to try to overcome this problem there is surpris-
ingly little consensus between them (Gerhart et al. 2000 ). The overall reliability of
HR practice measures is ‘frighteningly low’ (Wright and Gardner 2003 : 316 ). There
is also an attribution problem where the HR respondent in a successfulWrm may
assume that a practice exists or how else could theWrm be successful? Second, one
cannot rely on the same person to estimate HR practices and performance. This
‘common method variance’ is equally frightening. Third, no account is taken of the
lag eVect. How long does it take for an HR practice to impact on performance?
Measuring HRM and performance in the same time period cannot possibly show
that HRM drives performance (or the reverse that good performance drives better
HRM). Wright et al. ( 2005 : 412 ) are even harsher. They note that ‘by far the
most prevalent design [in the sixty-six studies they reviewed] is what we call
‘‘post-predictive’’ because it measures HR practices after the performance period
resulting in actually predictingpastperformance’ (their emphasis). Thus, overall,
‘the literature on the HRM-performance relationship has (a) universally reported
signiWcant relationships between HRM and performance, (b) almost exclusively
used designs that do not logically allow one to draw causal conclusions, and (c) very
seldom actually tested for a reverse causal order’ (Wright et al. 2005 : 416 ).
(^1) We discuss the limitations of this approach, especially the importance of contextual factors (Guthrie
2001 ; Arthur 1994 ; Datta et al. 2005 ; Capelli and Neumark 2001 ;Way 2002 ), in more detail below.
hrm and business performance 535