These same authors then argue that three criteria need to be used to establish
cause. First, cause requires that the eVect be present when the cause is present and
be absent when the cause is absent. Second, the proposed cause must exist in time
prior to the proposed outcome and, third, all other variables that might cause the
outcome are controlled for (Wright et al. 2005 : 411 ). These are tough criteria and,
when used as far as possible by Guest et al. ( 2003 ) and by themselves, no unam-
biguous HR-performance eVect is uncovered (but by the same token one cannot
‘suggest that HR practicesdo nothave a positive impact on performance’) (Wright
et al. 2005 : 433 , their emphasis).
26.3 The Problem of Performance
.........................................................................................................................................................................................
Interestingly, Guest et al.’s study used both ‘objective’ (i.e. published) performance
data and their respondents’ subjective evaluation of their Wrms’ performance
relative to their competitors. Using this subjective test, HRM was associated with
‘high comparative performance’ ( 2003 : 311 ). It is premature to write oVsuch
subjective evaluations and rely exclusively on objective performance or proWt
measures. In multisector studies, such as Guest et al. ( 2003 ), it is necessary to
control for sectoral variances in proWtability. Some sectors generate much higher
returns than others. Informed respondents may have a better idea of sectoral
conditions than revealed by simple published returns. It is also extraordinarily
diYcult to meet the third criterion for causal studies, that of controlling for all
other potential causal inXuences. As Hitt et al. note in their study in professional
serviceWrms, ‘Wrm performance is a function of many variables both inside and
outside aWrm... Thus, for one set of variables (human capital in their study) to
explain 3. 6 % of the variance inWrm performance may be signiWcant’ ( 2001 : 25 ).
Published corporate accounts are problematic since they cover diVerent business
units and perhaps countries where diVerent HR systems are likely to be in place in
part because of diVerent regulatory or institutional regimes (Paauwe 2004 ).
There are two much more fundamental problems with the use ofWnancial
performance data. First, it is far removed from HRM inXuence—too distal.
Second, it takes for granted thatWrms seek to structure their HR systems to
maximizeWnancial outcomes, often in the short term seen in shareholder value.
The problem with publishedWnancial measures is that no convincing explanation
can be provided as to why, or indeed how, HR practices have an inXuence. As such,
the research is remarkably uninformative from a practitioner point of view beyond
the knowledge that the ‘more’ HR practices the better (the so-called ‘Huselid
curve’). This has led many researchers to use, or at least call for, the adoption of
536 john purcell and nicholas kinnie