A Meta-Analysis of
Country Differences in the High Performance Work System-
Business Performance
Relationship:
The Roles of
National Culture and Managerial Discretion
High
performance work practices (HPWPs) are human resource (HR) practices designed
to increase business performance by enhancing employee ability, motivation, and
opportunity to contribute (AMO). HPWPs include, for example, selectivity in
staffing, investments in training, pay for performance, and employee
participation in decisions
(Becker & Huselid, 1998; Boxall & Purcell, 2003; Combs, Liu, Hall,
& Ketchen, 2006; Delery & Shaw, 2001). A high performance work system
(HPWS) includes combinations or bundles of multiple HPWPs that are hypothesized
to create mutually reinforcing, synergistic effects (Appelbaum, Bailey, Berg, & Kalleberg, 2000; Barney & Wright, 1998; Chadwick, 2010;
Huselid, 1995; Ichniowski, Shaw, & Prennushi, 1997; Lado
& Wilson, 1994;
MacDuffie, 1995; Ramsay, Scholarios, & Harley, 2000; Wright,
Dunford, & Snell, 2001; Wright & Snell, 1998).
Research has documented a positive and meaningful relationship between the use
of HPWSs and business (establishment, firm) performance (Arthur, 1994;
Becker & Huselid, 1998; Datta, Guthrie, & Wright, 2005;
Huselid, 1995; Jiang, Lepak, Hu, & Baer, 2012; MacDuffie, 1995; Way, 2002; Youndt, Snell, Dean, & Lepak, 1996). For
example, the Combs et al. (2006) meta-analysis reported a corrected correlation
of .28 between HPWSs and business performance and showed that a one standard
deviation increase in HPWSs was thus associated with a 126% increase in
profitability. Further, Combs et al. (2006) also reported that most (roughly
70%) of the between-study variance in the effect size was not explained by
statistical artifacts, suggesting there may be important substantive contextual
moderators of the HPWS-business performance relationship.
The
degree to which a body of theory and empirical evidence developed in one
contextual setting can be generalized to others is a key question in management
and applied
psychology broadly (e.g., Bamberger, 2008; Johns, 2006) and in strategic
HR specifically
(e.g., Baird & Meshoulam, 1988; Dyer & Reeves,
1995; Lengnick-Hall & Lengnick-Hall, 1988; Miles & Snow, 1984; Schuler & Jackson, 1987; Wright &
Sherman, 1999). In a global economy, an important question is whether HR
practices that work in one country work the same in other countries. Our study
provides the first systematic assessment of this question, based on 156
(including 108 non-United States) effect sizes from 35,767
firms/establishments.
Moderating Effects:
National Culture and Managerial Discretion
Overview
That HR strategies,
such as HPWSs, may not necessarily work the same everywhere, but instead depend
on their fit with a country’s national culture and other country (e.g.,
institutional and/or legal) characteristics, is a core theoretical idea in the
HR, strategy, and international management literatures (e.g., Adler, 2008; Aycan et al., 2000; Bae, Chen, Wan, Lawler, & Walumbwa, 2003; Brewster, 1995, 2007; Briscoe, Schuler, &
Claus, 2009; Dowling, Festing,
& Engle, 2008; Evans, Pucik, &
Barsoux, 2002; Ferner, 1999; Kostova, 1999; Kostova & Roth, 2002; Stahl
& Björkman, 2006; Stavrou, Brewster, & Charalambous, 2010). Hofstede (1993) states that there are “cultural
constraints in management theories” (p. 81), meaning that “not only practices
but also the validity of theories may stop at national borders” (p. 82). In a
similar vein, Brewster (1995) argues that what works in the United States may
not work in Europe because of differences in institutional (especially greater
legal/regulatory) pressures in Europe. As such, he called for a “European model
of human resource management” (p. 1), and “exploring of the appropriateness of
U.S. notions of HRM for adoption and institutionalization into European
practice” (Lazarova, Morley, & Tyson, 2008, p. 1997) became a major
research area. A similar debate about the transferability of HR strategies is
taking place in the context of China, the world’s second largest economy, but
not one of its most productive.[1]
Tsui (2009) challenged scholars to consider which of two strategies would be
more effective for management theory and practice in China: (1) applying
existing North American based theory and evidence to the Chinese context or (2)
developing new theories and corresponding streams of empirical research
tailored specifically to the Chinese context.
To
address the question of whether the HPWS-business performance relationship is
moderated by country differences in national culture and other institutional
factors, we rely primarily on national culture perspectives (e.g., Adler, 2008; Boyacigiller & Adler, 1991;
Hofstede, 1980, 1983, 1993, 2001; House, Hanges, Javidan, Dorfman, & Gupta, 2004;
Johns, 2006; Newman & Nollen, 1996). National culture perspectives
emphasize that national culture differences constrain or even dictate whether
management practices such as HPWSs will be effective in different countries.
Our conceptual framework additionally includes two aspects of managerial
discretion (e.g., Crossland & Hambrick, 2011) as key country-level
moderator variables: cultural tightness-looseness and institutional flexibility
(defined primarily in terms of legal/regulatory constraints, with a focus on
labor markets and labor unions).
National Culture
National
culture has been defined as the “collective programming of the mind which
distinguishes the members of one human group from another” (Hofstede, 1980, p.
25). National culture creates institutional/normative pressures for
organizations to conform their management practices and cultures to a national
model. Taras, Kirkman, and Steel (2010, p.
405) describe Hofstede’s (1980, 2001) work as “groundbreaking” and
state that it has “inspired thousands of empirical studies” and that such
studies are growing “exponentially”. Yet, they note that “conspicuously absent
from … reviews is an attempt to provide a more quantitative examination of the
empirical research” on national culture effects (p. 405). Taras et al.’s (2010)
meta-analysis of 598 studies examined correlations between Hofstede’s culture
scores and individual workplace attitudes and behaviors. However, they did not
look at management practices or at unit or firm level outcomes such as
performance. They also did not examine whether national culture moderates the
effectiveness of HPWSs specifically or management practices in general, an
important question that remains unanswered (Kirkman, Lowe, & Gibson, 2006).
As
noted, Hofstede (1993) argues that national culture is an important moderator.
He states that there are “cultural constraints in management theories” (p. 81),
meaning that “not only practices but also the validity of theories may stop at
national borders” (p. 82). In response to the classic standardization versus
localization debate in the international management literature (Bartlett & Ghoshal, 1989; Hannon, Huang,
& Jaw, 1995; Lillrank,
1995; Liu, 2004; Lu & Björkman, 1997; Perlmutter, 1969; Pudelko
& Harzing, 2007; Rosenzweig & Nohria, 1994),
Hofstede’s (2001) answer is clear: “For best results a multinational’s
management practices should fit the local culture” (p. 441).
This
fit logic is central to the national culture literature. Like Hofstede, Newman
and Nollen (1996) emphasize that “national culture is a central organizing
principle of employees’ understanding of work, their approach to it, and how
they expect to be treated” (p. 755) and that fit occurs when management
practices are “consistent” or “congruent” with employee expectations. In
contrast, “when management practices are inconsistent with these deeply held
[cultural] values, employees are likely to feel dissatisfied, distracted,
uncomfortable, and uncommitted [and] as a result, they may be less able or
willing to perform well” (Newman & Nollen, 1996, p. 755).
This
use of the term fit is consistent with its use in the strategy literature,
where it is also typically defined in terms of how “consistent” (using a priori
theoretical logic) one or more components (e.g., strategy and environment) are
(Nadler & Tushman, 1980, p. 45, cited in Ansari, Fiss, & Zajac, 2010,
p. 73). By far the most typical way to assess an a priori hypothesized fit in
the strategy literature (Boyd, Haynes, Hitt, Bergh, & Ketchen, 2012) is by
estimating a statistical interaction between a strategy variable and a
moderator variable such as the environment in predicting a criterion variable
such as performance (Venkatraman, 1989). In our paper, the strategy is HPWS,
the criterion is business performance, and the moderator is national culture.
To
examine the influence of national culture as a moderator of the HPWS-business
performance relationship, we followed advice to focus on a small number of
“well-chosen dimensions” (Zaheer, Schomaker, & Nachum, 2012, p. 33) that,
based on theory (e.g., Aycan,
2005; Early & Erez, 1997; Fischer &
Smith, 2003; Gomez-Mejia & Welbourne, 1991;
Newman & Nollen, 1996; Peretz & Fried, 2012; Peretz &
Rosenblatt, 2011; Schuler & Rogovsky, 1998), seem most relevant in
understanding how HPWSs might be differentially effective in different
countries.
As
described below, national culture perspectives lead to the specific a priori
prediction that the best fitting national culture for the success of an HPWS is
one low on power distance, low on collectivism (i.e., high individualism), and
high on performance orientation.[2]
That is not to say that such a national culture is always a better fit with
each and every aspect of an HPWS. However, based on accumulated theory and
evidence, this specific national culture is expected a priori, on average, to
better fit an HPWS and this stronger fit is, in turn, expected to result in
higher HPWS effectiveness.
Power
distance is “the degree to which members of a … society expect and agree that
power should be shared unequally” (House et al., 2004, p. 537 and Table 17.2, p.
536). In high power distance cultures, not only are power, authority, and
information unequally distributed, this hierarchy is also rigid and
institutionalized (Hofstede, 2001; Peretz & Fried, 2012) and advancement to
higher levels of the hierarchy is based on factors other than performance such
as seniority, social class, family, or political connections (Early & Erez,
1997; Sturman, Shao, & Katz, 2012). Similarly, “reward allocation is based
on criteria other than performance, such as seniority or being on good terms
with management” (Aycan, 2005,
p. 1106) and “the performance-reward contingency is low” (Aycan, 2005,
p. 1106). In addition, relatively few people have access to resources and
information in high power distance cultures. As a result, hierarchy is
reinforced and perpetuated (based on factors other than performance), making it
difficult to involve a broad range of employees to participate in decisions, as
well as hampering the use of performance in making promotion and pay decisions.
The lack of emphasis on using and rewarding ability and skill in staffing
decisions can translate into a workforce lacking in human capital (Ployhart
& Moliterno, 2011) and performance. Thus, overall, a high power distance
culture does not seem to be a setting that would enhance the effectiveness of
HPWSs, which are designed to increase performance by enhancing workforce AMO.
In
contrast, in low power distance cultures, ability, skills, and performance are
expected to play a larger role in staffing, employee participation, and
compensation. The de-
is part of the GLOBE
framework and, most important, its conceptual relevance seems clear given our
focus on performance as a dependent variable.
emphasis on hierarchy and status in low power distance countries favors
a high level of discretion and self-management that enables and encourages
taking full advantage of employees’ input and participation to improve
performance. Low power distance provides an environment that supports
investment in the whole workforce’s training and development (Aycan, 2005; Coget, 2011; Peretz &
Rosenblatt, 2011). Ability and performance, rather than seniority or social
connections, play a more important role in selecting, developing, and
motivating employees. Similarly, motivation-enhancing HPWS components such as
pay for performance are likely to be more effective in low power distance
countries (Aycan, 2005; Newman & Nollen, 1996).
Consequently, the lower a country’s power distance, the stronger will be the
HPWS-business performance relationship.
Collectivism
reflects “the degree to which individuals express pride, loyalty, and
interdependence in their families” (House et al., 2004, p. 463). The greater
the in-group collectivism, the closer are the ties among group members and the
stronger is people’s concern for others. Although that concern for and focus on
the collective has the potential for some positive influences on HPWS
effectiveness (e.g., for using teams and other participation in decision
practices, especially when combined with lower power distance, Earley & Erez,
1997, Table 2.1, p. 27; Kirkman & Shapiro, 1997), there are also potential
drawbacks. For example, in collectivist cultures, there is a norm of protecting
reputations and avoiding social anxiety, which may cause unease with practices
(performance appraisal/feedback, pay for performance) that recognize and
differentiate employees on the basis of individual performance (Peretz &
Fried, 2012). Selection and training/development decisions, like in high power
distance societies, may give less weight to ability and performance and more
weight to personal connections, in-group status, and social obligation
(House et al., 2004),
which can lower performance.
In
cultures low on collectivism (i.e., high on individualism), there is a stronger
focus on “rationality” (House et al., 2004, p. 453), which translates into
greater weight given in decisions to individual differences in ability, skills,
and performance. Thus, HPWS components that aim to enhance employees’ ability
such as training and development and selection are likely to be more effective
in countries low on in-group collectivism (i.e., high on individualism) because
there is greater acceptance of allocating opportunities based on individual
ability and performance, rather than on the basis of group affiliation or
status (Hofstede, 2001, Exhibit 5.3, pp. 226-227). Similarly, this emphasis on
ability and performance as legitimate criteria also means a better fit with
motivation-oriented HPWPs such as pay for performance (which use the
equity/performance allocation rule rather than the equality or need rules that
are more common in collectivist cultures) and thus greater effectiveness in low
collectivism cultures (Early & Erez, 1997; Fischer & Smith, 2003;
Gomez-Mejia & Welbourne, 1991; Schuler & Rogovsky, 1998; Zhou & Martocchio, 2001). Low
collectivism’s influence on the successful use of opportunity to contribute
practices such as employee participation are less clear. On the positive side,
low collectivism would be helpful because of the strong focus it places on
individual responsibility (Newman & Nollen, 1996). On the other hand, high
collectivism suggests that particular types of participation such as teams
could work more effectively under high collectivism. An important caveat,
however, is that teams, at least self-directed teams, must be able to carry out
tasks formerly carried out by managers, including selection, work
assignments/scheduling, and monitoring of individual performance. As noted
above, making distinctions between individuals may be more difficult in
collectivist cultures. On balance then, we expect that the lower the
collectivism in a country, the stronger the HPWS-business performance link.
High
performance orientation reflects “the extent to which a society is perceived to
encourage and reward performance improvement” (House et al., 2004, p. 246)
versus a focus on loyalty, harmony, and belongingness in low performance
orientation cultures. Further, high performance orientation countries
“emphasize results, set high performance targets, value taking initiative, and
prefer explicit and direct communications” (House et al., 2004, p. 276).
In contrast, in low performance orientation cultures, people “value social and
family relations, loyalty, tradition, and seniority, and use subtle and
indirect language” (House et al.. 2004, p. 276). Given the focus on results and
performance in strongly performance-oriented cultures, HPWSs, with their focus
on building performance through higher AMO, should be a strong fit and work
especially well in such cultures (Aycan, 2005; Bae & Lawler, 2000; Coget, 2011; Peretz
& Rosenblatt, 2011; Recht & Wilderom, 1998).
Specifically, the use of pay for performance and making staffing decisions
based on ability, skills, and performance should be a good fit. Also, the focus
on direct communication in performance orientation cultures should facilitate
the use and effectiveness of opportunity to contribute practices such as
information sharing, teams, and other forms of employee participation in
decisions. Based on the above discussion of national culture-based logic, when
a country’s national culture is consistent with the cultural values of an HPWS,
stronger business performance is expected: Hypothesis
1: The HPWS-business performance relationship will be more strongly
positive in countries characterized by low power distance (H1a), low
collectivism
(H1b), and high
performance orientation (H1c).[3]
Managerial Discretion
Although
the potential value of having management practices fit with/conform to national
culture is the primary emphasis in mainstream national culture perspectives, at
least some authors taking a national culture perspective (Gelfand, Nishii,
& Raver, 2006) recognize that conformance is a choice, and that more
conformance is not always better.
Indeed, Gelfand et al. (2006, Figure 1) identify “conformity vs.
deviance” with respect to national culture as a key choice facing
organizations. Likewise, Oliver (1991; see also Oliver, 1997) notes that
institutional theory, which provides the broadest theoretical logic for the importance
of institutional pressures and constraints such as national culture, has been
called on to remedy “its lack of attention to the role of organizational
self-interests and active agency” (p. 145) and further, to better recognize
that the existence of institutional pressures toward conformity does not
require the assumption that firms are “invariably passive and conforming to all
institutional conditions” (p. 146). More recently, Heugens and Lander (2009)
describe as a “central debate” in institutional theory literature the question:
“Does conformity to institutional norms enhance or diminish organizational
performance?” (p. 61). Thus, it is important to recognize that automatic
conformance of HR practices to national culture norms is neither always inevitable
nor necessarily always the most effective strategy.
The
preceding discussion makes clear that managerial discretion must be
incorporated into any conceptual treatment of whether a close fit between
management practices and national culture is the best path to strong business
performance. Managerial discretion plays a dominant role in strategic
perspectives (Barney, 1991; Newbert, 2007; Wernerfelt, 1984)
and related strategic HR perspectives (Barney & Wright, 1998; Dyer & Reeves,
1995; Lado & Wilson, 2004; Wright et al., 2001). In this vein, Cappelli and
Crocker-Hefter (1996) observe that “there are examples in virtually every
industry of highly successful firms that have very distinct management
practices” (p. 7).
As
we have seen, the role of managerial discretion is also recognized in at least
some strands of institutional (e.g., Oliver, 1991, 1997) and national culture
(Gelfand et al., 2006) perspectives. However, managerial discretion remains
largely absent from mainstream national culture perspectives. As we have also
seen, Hofstede (2001) takes a strong view that national culture constrains the
effectiveness of management practices (and thus managerial discretion).
Similarly, the GLOBE study (House et al., 2004) states that the “the major
force shaping organizational practices is rooted in societal-level systems”
(Brodbeck, Hanges, Dickson, Gupta, & Dorfman, 2004, p. 665) and that
“organizations mirror societies from which they originate” (Javidan et al., 2004, p. 726).
Note
that even if one accepts that organizations are compelled to conform, there is
the complication that organizations have multiple constituents and their norms
may conflict (Oliver, 1991). In the case of a multinational enterprise, for
example, an organization would need to decide which national culture to conform
to if cultural practice norms relating to HR differ in the host country and the
home country (Kostova, Roth, & Dacin, 2008). Likewise, there may be one set
of cultural/managerial practices within an industry in the home country, but a
different set when looking at the full set of global competitors in that
industry. This
“multiplicity of constituent demands” (Oliver, 1991) means that
conforming to one norm (e.g., a global industry norm) may require it to be out
of conformance with other norms (e.g., national culture norms). As such,
managerial discretion must be exercised and “isomorphism
[with the norms of the local country] is not a necessary condition for
legitimacy or survival”
(Kostova et al., 2008, p. 999).
Indeed,
conforming to the (local) national culture norm could actually endanger
survival if a new and different model is proving more successful for competing
globally (e.g., in a particular industry). For example, Bae and Lawler (2000,
p. 504) argued that, in Korea, “the pressures of globalization” made it
necessary for firms to begin implementing HPWSs and that “cultural changes”
necessarily followed that made workers and employers “more open” to such
systems, despite the fact that Korea’s national culture was conventionally seen
as having “important cultural traits that” would “undermine” the effectiveness
of HPWSs. In other words, fit with the existing national culture in this case
was not conducive to being competitive globally and, at some point, managerial
discretion was required that looked past traditional fit with national culture
(see related work by Gamble, 1993; Katz & Darbishire,
2000; Pudelko & Harzing, 2007; Quintanilla & Ferner, 2003.)
It
seems likely that organizations, as a general rule, seek to achieve “strategic
balance” (Deephouse, 1999) between conformance/fit and
nonconformance/differentiation (Oliver, 1991, 1997). It also seems likely that
organizations must balance achieving fit with the need to avoid a situation
where inertia is so strong that it undermines the flexibility needed to achieve
(dynamic) fit when the environment and/or competitive landscape changes (Teece,
Pisano, & Shuen, 1997; Wright & Snell, 1998; Zajac, Kraatz, & Bresser, 2000).
Such balance suggests the need to better incorporate the role of managerial
discretion in our thinking. We focus on two specific aspects of managerial
discretion as potential country-level moderators of the HPWS-business
performance relationship: cultural tightness-looseness and institutional
flexibility.
National culture tightness-looseness.
Although most national culture perspectives and national culture research have
focused on mean differences in institutional/cultural factors (see our
hypothesis 1), work by Gelfand et al. (2006) demonstrates that variance in
national culture, the tightness-looseness of national culture, may also be
relevant. Cultural tightness-looseness reflects “the strength of social norms,
or how clear and pervasive norms are within societies, and the strength of
sanctioning, or how much tolerance there is for deviance from norms within
societies” (Gelfand et al., 2006, p. 1226).
Our
hypothesis 1, based on national culture perspectives, states that the
HPWSbusiness performance relationship will be more positive in countries having
a national culture consistent with an HPWS, and less positive in others. Here,
we additionally make the case that hypothesis 1 is more likely to be supported
in countries with tighter national cultures because in tight cultures, norms
are stronger, tolerance for deviation from norms is less, and sanctioning for
deviation from norms is stronger. In other words, managerial discretion is low.
Our logic is similar to that of Taras et al. (2010), who found a significant
moderating effect of cultural tightness-looseness such that national culture was
more strongly predictive of individual workplace attitudes and behaviors in
countries having tighter cultures, consistent with the idea that tighter
cultures allow less room for individuals to deviate from societal norms. To the
degree a similar process operates for firms, we would anticipate that tighter
cultures not only provide less managerial discretion in choosing to use an HPWS
if it does not fit the (mean) national culture, but also undermine the positive
effects of an HPWS on business performance. Thus, the fit between an HPWS and
national culture is more important in tight cultures.
Hypothesis 2: Cultural
tightness-looseness will moderate the HPWS-business performance relationship,
such that the relationship will be more positive in countries characterized by
low power distance (H2a), low collectivism (H2b), and high performance
orientation (H2c), when the national cultures are tighter.
Institutional flexibility. To complete
the conceptual framework, it is necessary to incorporate stronger forms of
institutional control than national culture. Oliver (1991) defines
institutional control as “the means by which pressures are imposed on
organizations” (p. 168). In contrast to national culture, the legal coercion
form of institutional control is “legal or government mandates … imposed by
means of authority rather than through pressures for voluntary compliance”
(Oliver, 1991, p. 168). Oliver’s (1991) theory predicts that the typical
organizational response to legal coercion is to acquiesce. According to Kostova
(Kostova, 1999; Kostova et al., 2008), national culture captures the cognitive
(shared schemas, frames, inferential sets, etc. that people use to process
information) and normative aspects (values and norms) of how countries’
institutional environments differ, but it excludes the regulatory aspect (laws
and rules) of country differences. Consistent with Oliver (1991), Kostova
(Kostova, 1999; Kostova et al., 2008)
argues that, unlike in the case of national culture, firms have little
discretion when it comes to complying with regulatory aspects of institutional
pressure.
The
importance of country differences in the regulatory environment is also
emphasized in the comparative HR and industrial relations literatures (Björkman, Fey, & Park, 2007; Brewster, 1995, 1999, 2007; Freeman, Kruse,
& Blasi, 2008; Paauwe & Boselie, 2003; Siegel & Larson, 2009;
Sparrow & Brewster, 2006; Tregaskis & Brewster, 2006). Brewster (1999,
p. 224), for example, highlights the following argument from Pieper (1990,
p. 8): “The major difference between HRM in the U.S. and Western Europe
is the degree to which [HRM] is influenced and determined by [country
differences in] regulations.”
We
use the term institutional flexibility to capture country differences in
regulatory and related constraints (e.g., centralized collective bargaining,
inflexible labor markets) on employer discretion and flexibility. We define
institutional flexibility as employer flexibility in hiring and firing
practices (i.e., not impeded by laws/regulations), wage flexibility (i.e., not
determined by regulation or centralized bargaining), and few legislative
restrictions overall.
Unlike
national culture norms, which may vary in how they constrain HPWS effectiveness
based on how tightly these norms are adhered to, institutional (in)flexibility,
with its strong regulatory/coercive control dimension, seems likely to
constrain HPWS effectiveness much more systematically. If certain forms of pay
for performance (e.g., stock options) are simply not legal or if rates of pay
are set at the country or industry level, managerial discretion may not exist.
Similarly, if a form of employee participation in decisions (e.g.,
codetermination in Germany or the Netherlands) is mandated by law, managerial
discretion is limited. For employers wishing to implement certain HPWS
practices even where managerial discretion is constrained, it may be necessary
to invest substantial resources (time and money) to challenge legal
restrictions and/or gain regulatory approval (and success is not assured).
Thus, as Oliver (1991) notes, in the face of legal coercion or enforcement, the
most likely strategic response is to acquiesce. Consequently, the heterogeneity
among firms in HPWS use and effectiveness is less (more) likely where
institutional flexibility is low (high).
Hypothesis 3: The higher the level of
institutional flexibility in a country, the more positive the HPWS-business
performance relationship.
Method
Sample
We
searched for published and unpublished studies through March 2013 to include in
our meta-analysis. We searched for articles with titles and abstracts in
ABI/INFORM,
Business Source Premier/Complete, Google Scholar, JSTOR, ProQuest
Dissertations and
Theses, ProQuest Dissertation Abstracts International, PsycINFO,
PsycARTICLES, and
ScienceDirect databases using the terms “human resource management”, “human resource”,
“personnel”, “high performance work system”, “high performance”, “high
involvement”, or
“high commitment” in combination with “performance”, “turnover”, or
“productivity” in
English, as well as “Personalmanagement” or “HR-Praktiken” in
combination with “Unternehmenserfolg” in German. The results from these
searches were electronically screened for whether the terms “method” or
“sample” and either “correlation” or “effect” appeared anywhere in the text of
the article. We additionally searched for relevant in press articles across all
journals from our literature search in which any research on the HPWSbusiness
performance relationship had been published. We also searched for relevant
articles in all available conference programs of the Academy of Management and
the Society for Industrial and Organizational Psychology. In addition, we sent
requests for unpublished manuscripts, working papers, conference papers,
dissertations, and in press papers via the listservs of the following Academy
of Management divisions: Business Policy and Strategy, Entrepreneurship, Health
Care Management, Human Resources, International Management, Public and
Nonprofit, and Technology and Innovation Management. We contacted all authors
who had previously published on the HPWS-business performance relationship
(based on the results of our literature search) to ask for any unpublished work
that was relevant. We crosschecked these conference submissions, dissertations
and working papers with papers already published and/or already included in our
sample, and e-mailed the authors in cases where a full version of the article
was not publicly available. Finally, we cross-checked our search results
against previous meta-analyses and reviews related to our research topic (Boselie, Dietz, & Boon, 2005; Combs et al., 2006;
Gmür & Schwerdt, 2005; Jiang et al., 2012;
Paauwe, 2004; Stock-Homburg, Herrmann, & Bieling, 2009; Subramony,
2009; Wright, Gardner, Moynihan, & Allen, 2005) and with Huselid’s (2003)
summary of empirical studies from 1995 to 2003 that he provides on his
homepage.
To
be included in our meta-analysis, studies had to, first, measure and report the
use of an HPWS or HR system. Studies focusing solely on individual HPWPs or HR
practices were not included, unless the authors also examined the impact of an
additive or multiplicative index of such practices. Also, studies examining the
impact of broader strategic HR constructs (e.g., strategic HR effectiveness,
HR-strategy fit) or other related constructs (e.g., role of HR department,
downsizing practices) that did not represent the use of a system, bundle, or
configuration of HR practices were excluded. Second, studies had to measure
both HPWSs and operational or financial performance, and report effect size
estimates at the firm- or establishment-level. Studies conducting analyses
based on employee- or group-level measures of HPWSs and business performance
were excluded, unless they were aggregated to the firm- or establishment-level
prior to producing the effect size estimate. Studies not examining financial or
operational business performance outcomes (e.g., job attitudes, occupational
safety, and organizational citizenship behaviors) were also excluded even if
the outcomes had been aggregated to the firm- or establishment-level prior
COUNTRY DIFFERENCES 19
to conducting analyses. Acceptable measures of business performance
were either a) objective (e.g., market-to-book ratio, sales turnover), with data
sourced from company archives or reported by executives and/or employees; or b)
subjective ratings of performance (e.g., perceived organizational performance,
perceived labor productivity). Measures of financial performance across studies
in our sample included accounting returns, growth, and market returns, while
measures of operational performance included productivity, quality, and
retention. Certain studies included multidimensional measures of performance
that were an aggregate of multiple financial and/or operational performance
metrics. Third, studies had to report bivariate correlations for the
HPWS-business performance relationship or an effect size statistic that could
be converted to a bivariate correlation. Finally, studies based on samples of
firms/establishments across multiple countries were excluded because this would
require our coding of country characteristics (e.g., national culture) to vary
within study rather than between studies.
Applying
these criteria resulted in a final sample of 156 studies representing 35,767
firms and establishments across 29 countries. Appendix A [not part of this
pre-publication version] lists our sample of studies and information coded
regarding each study. Online Supplement 1 provides detailed references for each
study. Appendix B lists the countries in our sample along with country-level
characteristics coded. Online Supplement 2 lists studies excluded from our
meta-analytic sample and reasons for exclusion. In the description of our
results, tables, appendices, and online supplements, N is the number of firms and establishments, K is the number of studies, and Ncountries
is the number of countries.
Effect Sizes and Variables
Effect size. For each study in our
sample, we coded the bivariate correlation coefficient estimate of the
HPWS-business performance relationship, sample size (i.e., number of firms
and/or establishments), as well as estimates of measurement error (e.g.,
inter-item reliability–alpha, inter-rater reliability–intraclass
correlation) in both predictor and criterion variables. This information was
coded by two of the authors for a sub-sample of 20 studies. Because the mean
inter-rater agreement was 94%, and all disagreements were subsequently resolved
through discussion, the task of coding the remaining studies was split between
two of the authors.
We
contacted the authors of the original studies for any information that was not
reported in the studies. When multiple correlation coefficient estimates were
reported within a single study, for example between an HPWS and either measures
of alternate business performance constructs or multiple measurements of the
same business performance construct across time, a linear composite correlation
(Hunter & Schmidt, 2004) was calculated using
information reported in correlation matrices. If reliability estimates and
sample sizes corresponding to each correlation coefficient within a study
differed, we coded the arithmetic average of the multiple reliability estimates
and the minimum of the multiple sample sizes. Multiple studies that had been
published using the same dataset were treated as a single study by using an
arithmetic average of the correlation coefficient estimates reported across
these studies. In studies that used multiple types of HR systems composed of
different bundles of practices, the bivariate correlations of each system with
business performance were aggregated to create a single effect size estimate
for the study, provided the items used in measuring the component practices in
each system resembled the content of HPWSs.
Moderators. For each study, we
identified the country from which its sample of firms and/or establishments
originated. Across countries represented in our sample, we coded information
pertaining to the country-level characteristics of interest: (1) national
culture, (2) cultural tightness-looseness, and (3) institutional flexibility.
National
culture scores on power distance, collectivism (specifically, in-group
collectivism), and performance orientation were obtained from the GLOBE study
(House et al., 2004). We used the practices (“as is”) scales which range from 1
(low) to 7 (high). We chose to focus on practices (“as is”) rather than values
(“should be”) because the national culture-related part of our theoretical
framework focuses primarily on the normative/cognitive (national culture) and
coercive (regulatory) institutional pressures organizations actually face when
deciding whether to adopt management practices (in this case, HPWSs).
Institutional and strategic perspectives also focus on the decision by
organizations to conform or not conform to practices and strategies used by
other organizations in their institutional fields and/or competitor groups. In
addition, an examination of House et al.’s (2004) tables 16.9 (p. 475) and 17.6
(p. 543) reveals that the GLOBE practice scores are more strongly correlated
with corresponding Hofstede (1980, 2001) national culture scores than the GLOBE
value scores. For example, collectivism scores from Hofstede correlate .81 with
the GLOBE in-group collectivism practices scale, but only .20 with the GLOBE
in-group collectivism values scale.
Similarly, the
GLOBE power distance practices and values scales have correlations of .57 and
.03, respectively, with Hofstede’s power distance scale. Given that Hofstede’s
work forms the primary theoretical basis for national culture perspectives, we
feel it is important that our measures of national culture correspond not only
conceptually, but also empirically with his work. We chose to use the national
culture scores from GLOBE because they are based on more recent data than
Hofstede’s scores (publication date of 2004 versus 1980), more closely
corresponding to the time period during which most of the studies in our
meta-analysis were conducted.
We
obtained scores for cultural tightness-looseness from Gelfand et al. (2011, p.
1103). In our sample, scores range from 1.60 (Ukraine) to 11.80 (Malaysia),
with higher scores corresponding to greater tightness of cultures (i.e.,
greater strength, clarity, and pervasiveness of cultural norms, and stricter
sanctioning of deviation from norms).
Institutional flexibility was measured using three indicators from the
Global Competitiveness Report 2010-2011 (World Economic Forum, 2010): 1) lack
of burdensome government regulation (“How burdensome is it for businesses in
your country to comply with government administrative requirements? [1 = extremely burdensome; 7 = not burdensome at all]”); 2) flexibility
of wage determination (“How are wages generally set in your country? [1 = by a centralized bargaining process; 7 =
up to each individual company]”); and
3) hiring and firing practices (“How would you characterize the hiring and
firing of workers in your country? [1 = impeded
by regulations; 7 = flexibly
determined by employers]”). We constructed an overall institutional
flexibility scale by summing the z
scores of each dimension (α = .85). Scores on this index range between -3.82
(Italy) and 6.20 (Singapore), with higher scores indicating higher
institutional flexibility.
Control variables. We included three
industry categories (manufacturing, service, or mixed), level of analysis (firm
versus establishment), and the composition of the HPWS as controls. For each
HPWS, we measured the percentage of practices that fell into six areas:
compensation, employee relations, performance management, training and
development, promotion, and recruitment and selection. The mean inter-rater
agreement for the information on the content of HPWSs coded by two of the
authors across a subsample of 10 studies was 100%.
Analyses
We began our analyses by re-examining the HPWS-business performance
relationship using the Hunter and Schmidt (2004) meta-analysis framework.
Bivariate correlation coefficient estimates were aggregated across studies to
generate the samplesize weighted overall average effect size (̅). Corrections were made for
measurement error artifacts to obtain the weighted mean true score correlation
(ρ) estimates. As internal consistency (coefficient alpha) for measures of the
predictor (K = 108) and criterion (K = 64) variables were not reported by
all studies, the artifact distribution method was used to correct for the
attenuating effects of unreliability (due to items). The average internal
consistency was .81 for HPWS measures and .82 for business performance measures.
Estimates of inter-rater reliability in HPWS measures were even less commonly
reported among studies (K= 13), and
the average intraclass correlation was .36. Therefore, except in the case of
the overall HPWS-business performance effect size estimate over all
countries/regions (see Table 2), we did not have sufficient information to make
corrections for rater measurement error.
To
detect substantive moderators, we first employed the “75 percent rule” which
suggests that moderators may be present whenever artifacts fail to explain 75%
or more of the observed (uncorrected) variance in correlation coefficient
estimates across studies (Hunter & Schmidt, 2004). Due
to the tendency towards type-I error that is characteristic of this decision
rule (Sagie & Koslowsky, 1993), 80%
credibility intervals constructed around the true score correlation (ρ), and
tests of homogeneity provided an alternative means of obtaining information
regarding the true population variance of HPWS-business performance
correlations. Additionally, we constructed 95% confidence intervals using the
mean uncorrected correlation (̅)
and its standard error to gain a sense of the precision of our metaanalytic
estimates (Whitener, 1990).
We
use hierarchical linear modeling (HLM) to examine moderating effects of the
hypothesized country-level characteristics because the studies are nested
within countries. Due to its implicit hierarchical structure and potential for
random effects through heterogeneity in the mean corrected correlation,
meta-analysis has been recognized as a special case of multilevel analysis (Hox, 2010; Hox & De Leeuw, 2003; Raudenbush & Bryk, 2002). We
used HLM v.7.01 to construct three-level “variance-known” models (Raudenbush & Bryk, 2002, Chapter 7) with
firms/establishments at level-1, studies at level2, and countries and
between-country characteristics (e.g., national culture, institutional
flexibility ) at level-3.[4]
For each study in our sample, the level-1 sampling variance (σ²) of effect
estimates is given by σ² = 1 / (Nj – 3), where Nj is the study sample size (i.e., number
of firms and establishments) of the jth
sample or study (Hox & De Leeuw, 2003, p. 93; Raudenbush & Bryk, 2002, p. 219). We entered
the inverse of the sampling variance as nonnormalized weights at level-1, and
constrained the level-1 variance (σ²)
to be equal to 1 (see Hox, 2010; Hox & De Leeuw, 2003; Raudenbush, Bryk,
Cheong, Congdon, & du Toit, 2004 for a discussion of v-known analysis using
HLM). Also, as it may be incorrect to assume that our sampling distribution of
correlation effect estimates (r) is
normal (Hox & De Leeuw, 2003; Raudenbush & Bryk, 2002),
effect sizes were transformed using Fisher’s z. Full maximum likelihood estimation was used and all main effect
and interaction variables were grand-mean centered (Kreft & De Leeuw,
1998).
As
an example, the equations corresponding to the model examining the moderating
effect of power distance on the HPWS-business performance relationship are
listed below, where i is firms and
establishments (level-1 unit), j is
studies (level-2 unit), k is
countries (level-3 unit), zijk
is the transformed Fisher’s correlation, σ²ijk
is the level-1 sampling variance (which is externally specified and constrained
to be equal to 1), U0jk is
the level-2 variance, V00k
is the level-3 variance, β0jk is the level-1 intercept, δ00k
is the level-2 intercept, β1 to β9 are regression
coefficients of level-2 control variables, γ000 is the level-3
intercept, and γ1 is the regression coefficient of the level-3
predictor variable (i.e., power distance). The statistical and practical
significance of γ1 tests our hypotheses on the extent to which
level-3 (i.e., country-level) variables (e.g., national culture, institutional
flexibility) moderate the magnitude of HPWS-business performance correlations (zijk).
Level-1 model: zijk = β0jk + σ²ijk
Level-2 model: β0jk
= δ00k + β1 Plant-level (dummy)0jk + β2
Percent
Compensation0jk
+ β3 Percent Employee Relations0jk + β4
Percent Performance
Management0jk
+ β5 Percent Training & Development0jk + β6
Percent
Promotion0jk +
β7 Percent Recruitment & Selection0jk + β8
Manufacturing
(dummy)0jk +
β9 Service (dummy)0jk + U0jk
Level-3 model: δ00k
= γ000 + γ1 Power Distance00k + V00k
Results
Table 1 reports means, standard deviations, and correlations. The mean
unweighted observed HPWS-business performance correlation was .28. Table 2 shows
that the sample size weighted mean observed HPWS-business performance
correlation (̅) was .22. The
sample size weighted mean correlation corrected for random measurement error
due to items (ρ) was .28.[5]
In the full sample, artifacts (i.e., sampling error and measurement error due
to items) accounted for 14.7% (i.e., less than 75%) of the observed variance in
raw (uncorrected) correlations, suggesting the likely presence of substantive
moderators. Further corroborative evidence of heterogeneity in effect estimates
was found in the form of a statistically significant test of homogeneity (χ2
= 1189.93, p < .05) and what might
be described as a relatively wide 80% credibility interval (.03 : .53). Table
2, which shows the 19 countries with K
≥ 2, indicates that
̅ and
ρ are positive for all 19 countries. The smallest ̅ is .05 (Canada and the Netherlands), while the largest
are .57 for Greece and .59 for Singapore. Only the confidence intervals for
Canada, the Netherlands, and Ghana include zero. Table 2 also shows results by
region (as defined in House et al., 2004). The
highest ̅, .51, is for the
Middle Eastern region, but that is based on only K = 3, N = 300. The next
highest ̅, .33, is for the
Confucian region (based on K = 40, N = 8,016). The lowest ̅, .06, is for the Germanic region
(based on K = 5, N = 592).
In our dataset, studies are nested within countries and we are using
country characteristics to explain between-study variance. The intraclass
correlation for country (ICCcountry)
is .322, indicating that 32.2% of the between-study variance in effect sizes
occurs between countries and 67.8% occurs within countries. (Note: ICCregion = .188.) Consistent
with standard HLM logic, we now examine the degree to which specific country
characteristics are able to explain the variation in effect size magnitude. In
doing so, we include controls[6]
for industry, level of analysis, and HPWS content. The control variable model
results are provided in Appendix C.
According to hypothesis 1, which is based on national culture
perspectives, countries with low power distance, low collectivism, and high
performance orientation will have a more strongly positive HPWS-business
performance relationship. As Table
3 shows, performance orientation is not statistically significantly
related (B = -.149, p =
.24) to the HPWS-business performance effect size. In contrast, the
coefficients for collectivism (B =
.125, p < .05) and power distance
(B = .229, p < .05) are statistically significant and (contrary to
prediction) positive. As such, hypotheses 1a, 1b, and 1c, based on national
culture perspectives, are rejected.[7]
Given
the mean unweighted HPWS-business performance correlation of .28 (see Table 1), the
effect size (r) is expected to be .35
in countries one standard deviation above the mean power distance and .21 in
countries one standard deviation below the mean power distance.[8][9]
Similarly, the effect size (r) is
expected to be .36 in countries one standard deviation above the mean in-group
collectivism and .19 in countries one standard deviation below the mean
in-group collectivism.Table 3 also shows that country
differences in power distance explain 34.7% of the between-country variation in
effect sizes and 15.3% of the total between-study variation in effect sizes.
Country differences in collectivism explain 49.2% of the between-country
variation in effect sizes and 21.8% of the total between-study variation in
effect sizes.
Hypothesis
2 stated that the HPWS-business performance effect size would be more strongly
positive in low power distance (H2a), low collectivism (H2b), and high
performance orientation (H2c) countries with tight national cultures. Table 4
shows that the cultural tightness-looseness * performance orientation
coefficient is statistically significant and positive (B = .105, p < .05) and
the cultural tightness-looseness * power distance coefficient is statistically
significant and negative (B = -.084, p < .05). Although the cultural tightnesslooseness
* in-group collectivism coefficient was not statistically significant, its sign
was still negative. These findings are consistent with hypotheses 2a and 2c
since countries that are culturally tight and either score lower on power
distance or higher on performance orientation have more strongly positive
HPWS-business performance effect sizes.
As illustrated in
Figures 1 and 2, the national culture-based logic of hypothesis 1 appears to
hold in high tightness national cultures, as indicated by the negative
relationship between the HPWS-business performance effect size and power
distance and the positive relationship between the HPWS-business performance
effect size and performance
orientation. In contrast, the opposite
appears to hold in low tightness (i.e., looser) national cultures as indicated
by the positive relationship between the HPWS-business performance effect size
and power distance and the negative relationship between the HPWS-business
performance effect size and performance orientation. Thus, hypothesis 2 is
supported. However, in all cases, the HPWS-business performance effect size
remains positive.
Hypothesis 3 stated that
the HPWS-business performance relationship would be more strongly positive when
institutional flexibility was high. Table 5, however, shows that neither the
separate dimensions (lack of burdensome government regulation, flexibility in
hiring and firing, flexibility in wage determination) of institutional
flexibility nor a scale combining these dimensions was statistically
significantly related to the HPWS-business performance effect size. Thus,
hypothesis 3 is not supported.
Discussion
Contextual
differences must be considered in management theories (e.g., Bamberger, 2008; Johns, 2006). In a global economy, the
potential importance of country-level contextual differences is especially
relevant. Indeed, many scholars (e.g., Black & Porter,
1991; Boyacigiller & Adler, 1991; Brewster, 1995, 2007; Gerhart, 2009; Hofstede,
1983, 1993, 2001; Newman & Nollen, 1996; Tsui,
2006, 2009) have asked to what degree management (including HR) theory and
practice generalizes across countries and to what degree new theories and
research evidence must be developed for different countries. Here, we addressed
one part of this broader question.
Specifically,
we asked: Are HPWSs more effective when they fit the national culture and the
institutional environment of a country? According to national culture
perspectives, HPWSs should work better (i.e., the HPWS-business performance
effect size should be more positive) in countries where they fit the national
culture, which are hypothesized to be countries lower on power distance, lower
on collectivism, and higher on performance orientation. Our findings, however,
suggest that, contrary to national culture-based logic, HPWSs do not, on
average, work better when they fit the national culture. Likewise, we found no
evidence that the HPWS-business performance effect size was more positive in
countries where HPWSs were hypothesized to better fit the institutional
environment (i.e., countries having higher institutional flexibility). Thus,
our main results suggest that national culture and institutional flexibility do
not, on average, constrain the effectiveness of HPWSs.
In
fact, although we found moderating effects of national culture, these effects
were mostly the opposite of those hypothesized a priori using standard national
culture-based logic. In other words, the HPWS-business performance effect size
was, contrary to national culture-based expectations, more strongly positive in
countries high on power distance and high on collectivism. The effect size was
also more strongly positive in low performance orientation countries, again
contrary to conventional national culture-based logic, but the difference was
not statistically significant.[10]
Finally,
our results suggest an important caveat: National culture-based logic (our
hypothesis 1) receives empirical support in countries where managerial
discretion is low, as indicated by the existence of a tight national culture.
In tight cultures, the HPWS-business performance relationship was somewhat more
positive in national cultures lower on power distance and/or higher on
performance orientation. However, it is important to note that even in tight
national cultures and in national cultures high on power distance and low on
performance orientation, the HPWS-business performance effect size is positive.
Indeed, even in tight national cultures, greater fit between national culture
and HPWS resulted in effect sizes that were only slightly more positive than
those observed when fit was expected a priori to be lower.
What
explains the fact that some organizations choose to use HPWSs in countries even
where the national culture of such countries is theorized to be a poor fit and
further, that they succeed in using HPWSs under such conditions? First, as
noted earlier, organizations may face a “multiplicity of constituent demands”
(Oliver, 1991, Table 3, p. 160) or norms. It is increasingly rare that
organizations compete only with other domestic organizations and only in their
domestic market. Therefore, it is possible that the stronger positive
HPWS-business performance relationship we have observed for firms that use
HPWSs in countries where there is, according to national culture perspectives,
a lack of fit with the national culture, is partly a result of such
organizations conforming to a different, more global standard of management
practice. For example, to be competitive in their industry on a global basis,
they may decide to use a management strategy (e.g., an HPWS) that does not,
according to national culture perspectives, fit the (mean) national culture.
They may decide they must make it work, despite challenges, to be able
to compete globally. Similarly, Kostova et al. (2008) observe that “isomorphism
[with the norms of the local country] is not a necessary condition for
legitimacy or survival” (p. 999). One might take this logic one step further
and suggest that isomorphism with the norms of the local country may, in some
cases, undermine both legitimacy and survival in the global market.
Second,
it is possible that organizations adapt HPWSs to some degree to reduce friction
with (local) national culture norms. It seems likely that most management
practices will be tailored by firms, whether in different countries or in the
same country. Speaking of management practices broadly, Ansari et al. (2010) contend
that “management practices often cannot be adopted by user organizations as
‘off the shelf’ solutions” but rather that “diffusing practices are likely to
evolve during the implementation process, requiring custom adaptation,
domestication, and reconfiguration to make them meaningful within specific
organizational contexts” (pp. 67-68). Similarly, Pfeffer (1994), focusing on HR
practices specifically, argued that although there was a set of best practices
for managing people that apply to virtually all firms, “obviously, how one
would implement these practices will vary significantly, based on a given
organization’s strategy and its particular technology and environment”
(Pfeffer, 1994, pp. 64-65). Pfeffer referred to this as “the contingent nature
of the implementation of these practices” (p. 65), something he argued that
“everyone would agree is necessary” (p. 65). Becker and Gerhart (1996) distinguished
between (from most general to most specific) HR principles, HR policies, and HR
practices. For example, a principle could be that employee performance is
valued, but this principle could be operationalized with different policies and
practices in different organizations.
Third,
in a very similar vein, the international management literature, similar to
Pfeffer’s (1994) concept of contingent implementation and Ansari et al.’s (2010) concept
of customization, recognizes “hybridization” as “the insertion of a business
system into a new society or context, and the processes of adaptation and
learning involved” (Tolliday, Boyer, Charron, & Jürgens, 1998, pp. 3-4).
For example, in the case of multinational enterprises, some may seek to achieve
“convergence to a worldwide best practices model” (Pudelko & Harzing, 2007,
p. 535) in the area of HPWSs, but at the same time, this “convergence” may
accommodate unique local characteristics. For example, worker participation in
decision making is mandated and institutionalized through co-determination in
all but the smallest Dutch firms. Thus, to improve opportunity to contribute in
a Dutch firm, the focus would need to be elsewhere (e.g., the use of
self-directed work teams). Alternatively, it is possible that in this case the
HPWS would focus more on the motivation and ability aspects. Thus, it is
possible that the HPWSs examined in the studies that are part of our
meta-analysis may have represented hybrids where “the ‘essence’ of [an HPWS is transferred],
but in a reinterpreted and reinvented form that better fits the different
institutional and cultural context” (Kühlmann, 2012, pp. 95-96). Kühlmann
(2012) frequently observed such hybrids in his study of the transfer of HR
practices within MNCs from Germany to China. In the international context,
Evans et al. (2002) argue that in considering how much management practices
need to conform to local norms, ‘‘the real question concerns what to respect,
what to ignore, and what to reinvent when adapting work practices to another
[country] environment” (p. 163).
Fourth,
the above discussion, combined with some of our findings may point to the need
to give greater attention to managerial discretion and the ability that
organizations have to differentiate themselves from competitors even in the
face of national culture and institutional constraints. The intraclass
correlation for country indicated that 32% of between-study variance in effect
sizes was between countries. That means that the majority of variance in effect
sizes, 68%, was not explained by country. Thus, the effectiveness of an HPWS
does not appear to be dictated by the country (or by specific country
characteristics such as national culture or institutional flexibility).
Further, the effect of country that we observe does not operate in the manner
expected under conventional national culture perspectives. Specifically, the
HPWS-business performance effect size was always positive, never negative and,
except in tighter national cultures, the HPWS-business performance effect size
was actually more strongly positive in national cultures where HPWSs were
expected to fit less well (according to standard national culture-based logic
and a priori hypotheses). Resource-based view (RBV) perspectives (Barney, 2001;
Barney & Wright, 1998; Lado & Wilson, 1994; Wernerfelt, 1984;
Wright et al., 2001) may help explain our findings, given that RBV perspectives
recognize that sustained competitive advantage (i.e., outperforming industry
competitors over time) can theoretically be achieved by following a strategy
that creates value in a way that is rare and difficult to imitate. RBV
perspectives may help explain why some firms choose to (successfully) implement
an HPWS strategy in countries that have a national culture and/or institutional
environment that is not seen as consistent with an HPWS or conducive to the use
of an HPWS. Future research may wish to assess whether an HPWS has a higher
payoff in countries where use of an HPWS is rare.
Fifth,
our study suggests that an organization’s ability to successfully implement an
HPWS in a country where fit with (mean) national culture is weak may depend, to
some degree, on how tight or loose that national culture is. Specifically, as
noted, we found that such a strategy was more successful in countries having
looser national cultures, which provide more tolerance for deviation from
cultural norms and likewise sanction such deviations less strongly. In
contrast, in tight national cultures, we found that HPWSs were most effective
when they fit the (mean) national culture. Importantly, under all combinations
of (in)consistency with national culture and cultural tightness-looseness, the
mean effect size for the HPWS-business performance relationship was positive.
Although
we also expected institutional flexibility, especially with its strong
regulatory/coercive component (as we defined and measured it), to influence the
HPWSbusiness performance relationship, we did not find such evidence. One
possibility is that coercive/legal institutional pressures are so strong and
pervasive that HPWSs are either simply not feasible in some countries or are
strictly required in other countries and thus, we do not observe their
relationship with performance in those countries. As a consequence, it would be
difficult to observe any moderating effect of institutional flexibility.
Alternatively, it may be that institutional forces are either not as strong as
sometimes argued or that our measures of institutional flexibility were not
adequate for capturing the hypothesized effects. Perhaps there are particular
aspects of the regulatory environment that vary by country and that influence
the adoption and success of at least some components of HPWSs. Another
possibility is that even in countries where regulatory constraints are high and
some aspects of HPWSs may be difficult or even, in effect, not possible to use,
organizations that are strongly committed to the use of HPWSs are able to use
enough core HPWS practices to obtain positive effects.
Limitations and Future Directions
Our
study has several limitations. First, our comparison of effect sizes across
countries relies on non-representative samples, meaning that there may be file
drawer/publication bias issues. We constructed a funnel plot (Hox, 2010; Hunter & Schmidt, 2004) of sample size (N) against effect size (̅).
The funnel plot was symmetric (excluding one outlying large sample study
reporting a negative effect estimate) and therefore not suggestive of a
disproportionate exclusion of small-sample studies reporting small effect
estimates. In addition, the fail-safe/file-drawer test of effect sizes (Hunter & Schmidt, 2004,
p. 501) indicated that the number of missing or excluded studies
averaging null results (i.e., no statistically significant HPWS-business
performance relationship) required to decrease our mean observed effect size (̅ =
.22) to a small effect size of .10 (Cohen, 1988) would be 187. Given that this
number is larger than our current meta-analytic sample size, it is unlikely
that publication bias had a significant effect on our results.
We
also checked the potential for publication bias by constructing funnel plots
and conducting fail-safe/file-drawer tests on subsamples of studies from the
United States, China, and Spain, the three countries from which the most number
of studies originated. The funnel plots of effect estimates from each country
failed to provide clear and convincing evidence of publication bias. Further,
in each case the number of excluded studies reporting an effect size of zero
required to reduce the mean observed r
to a small effect size is 1 to 2.5 times as many studies (United States = 62,
China = 40, Spain = 13) as currently included in each subsample. These results
make it unlikely that the population of studies on the HPWSbusiness performance
relationship in each country may have been disproportionately sampled across
countries.
Second,
it is possible that differences in effect sizes across countries reflect
differences in the definition and measurement of the HPWS used. As noted above,
HPWSs may be implemented differently in different studies. Whether within or
across countries, this has been an issue at least since being highlighted in
1996 by Becker and Gerhart. In 2013,
17 years later, Posthuma, Campion, Masimova, and Campion stated that
“High Performance Work Systems are designed to enhance organizational
performance …Yet there is very little consensus about the structure of these
systems and the practices therein” (p. 1184). Thus, what is in an HPWS and how
that differs across studies is an ongoing issue and one that does not seem
specific to multi-country studies such as ours. Nevertheless, we did address
this issue empirically in three ways. First, we found that our results on
moderation of the HPWS-business performance relationship were invariant to the
inclusion of controls for six areas of HPWS content. This analysis is included
in Appendix C. Second, a detailed comparison found that for 55 of 68 HPWS
practices, China and the United States, which provided the most effect sizes
for our study, showed significant similarity in use. This analysis is presented
in Online Supplement 3.
Third, we then
took the six countries providing the most effect sizes (United States,
China, Spain, United Kingdom, South Korea, Canada) and examined the
percentage of HPWS practices that fell into the six HPWS content areas. These
percentages were similar, again pointing to similarity in HPWS composition
across countries in our sample (see Online Supplement 4). We next correlated
the columns in the HPWS practice areas (6 rows) x countries (6 columns) matrix.
The mean correlation between the countries was r = .82. The mean correlation of each non-U.S. country with the
U.S. was r = .89.[11]
We interpret these findings as indicating substantial similarity in the
composition of HPWSs across countries.
These analyses do not rule out the possibility (noted earlier) that
within the six HPWS areas, organizations in different countries use different
practices and/or customize/adapt the same practices. However, we neither know
of any evidence that documents such differences nor of any evidence that this
is more likely in comparing organizations across countries versus within
countries. We already know, for example, that the composition of HPWSs differs
significantly across studies done even in a single country such as the United
States. Future research on this question would be very helpful.
As
a third potential limitation, our findings may be affected by selection bias.
It is possible that we are more likely to observe successful/surviving cases of
HPWSs. To the degree that occurs, the effect size estimates we have reported,
at least for some countries, could be biased upward. Thus, future research that
models the HPWS adoption/survival process in different countries would be
useful. Our results indicate that some portion of organizations is able to
successfully negotiate or overcome any country-level constraints and challenges
to realize the positive payoff to HPWSs observed in our study. However, work on
adoption might shed more light on the risk of following an HPWS strategy in
different countries.
Fourth,
although we investigated the potential moderating effects of national culture
and institutional flexibility, it is possible that other country differences
may be relevant. Future studies may, for example, wish to examine other
institutional factors such as a country’s educational system (Sparrow &
Brewster, 2006), within-country differences in sector (e.g., state-owned versus
privately-owned enterprises in China), economic factors such as varieties of
capitalism (Hall & Soskice, 2001; Whitley, 1999), or how common HPWS use is
in the country or industry (Bloom & Van Reenen, 2007).
However, we know, based on the current study and previous studies, that
country-level factors are often significantly correlated, thus placing a limit
on how much additional country-level factors will contribute to our
understanding. For example, Hall and Gingerich (2004)’s coordination index,
which distinguishes liberal versus coordinated markets in the varieties of
capitalism approach, correlates r = -
.60 and r = .83, respectively, with
the national culture variables of hierarchy and harmony (Schwartz, 2007).
Summary
Our
study is the first to systematically examine, using the full body of available
evidence, the degree to which the effectiveness of HPWSs is similar or
different across countries and to what degree country-level moderators
(national culture, cultural tightnesslooseness, institutional flexibility) may
help explain any such differences. Our most important finding was that the mean
effect size for the HPWS-business performance relationship was positive overall
(corrected r = .28) and in every
country (and statistically significant in all but three countries). Contrary to
standard national culture-based logic, the HPWS-business performance effect
size was actually more strongly positive in countries where the a priori
hypothesized fit between an HPWS and the national culture was expected to be
lowest, except in tight cultures. But, even here, the HPWS-business performance
effect was always positive. As such, our findings challenge the conventional
view of national culture perspectives that management practices must be fully
tailored to the national culture and that managerial discretion is dominated by
national culture (and/or home country institutional constraints).
[1] The most recent World Bank
(2013) figures on productivity (gross domestic product per employed person,
adjusted for purchasing power) are $14,196 for China, compared to an average of
$51,965 for the other largest economies (United States, Japan, Germany,
France).
[2] In addition to their
central role in conceptual treatments of national culture’s role in
organizations and in influencing the effectiveness of HR practices such as
HPWSs, collectivism (individualism) and power distance are part of both
Hofstede’s (1980, 2001) original framework and the more recent GLOBE framework
(House et al., 2004).
They have also received the most empirical attention of any national culture
dimensions
(Taras et al., 2010). Performance
orientation, per se, is not part of Hofstede’s original framework. However, it
[3] Note that our fit hypothesis specifies, a
priori, that a specific form of statistical interaction(s) must be supported to
infer support for national culture-based logic.
[4] Hox (2010, p. 206) notes
that “the random-effects model for meta-analysis assumes that study outcomes
vary across studies, not only because of random sampling effects, but also
because there are real differences across studies.” As such, under
random-effects meta-analysis, “we have a hierarchical data set, with subjects
within studies at the first level, and studies at the second level” (Hox, 2010,
p. 206). When conducting a metaanalysis, ordinarily not the raw data are used
but rather “the summary statistics that are the available data for the
meta-analysis” (Hox, 2010, p. 207). Therefore, instead of representing a
parameter to be estimated, the within-study variance is known.
[5] When additionally
corrected for attenuation by random measurement error due to raters (Gerhart,
Wright, McMahan, & Snell, 2000) in the predictor variable (i.e., HPWS)
based on an average ICC(1) of .36
reported in a subsample of studies (K
= 13), the mean true score correlation was .46.
[6] For comparison, results
excluding control variables are presented in Online Supplement 5 and Online
Supplement
8.
[7] For interested readers,
Online Supplement 7 additionally provides the results on the relationships
between the six other national culture dimensions from the GLOBE study (House et al., 2004) and HPWS-business
performance effect sizes. Descriptive statistics and correlations including
these other national culture dimensions are presented in Online Supplement 6.
[8] A change in .229 in Fisher’s z transformed r translates to a change of .23 in r. Given the standard deviation of power distance of .30 reported
in Table 1, the change in r for a one
standard deviation change in power distance can be calculated by multiplying
.30 by .23, which is .069.
[9] A change in .125 in Fisher’s z transformed r translates to a change of .12 in r. Given the standard deviation of in-group collectivism of .71
reported in Table 1, the change in r
for a one standard deviation change in
ingroup collectivism can be calculated by multiplying .71 by .12, which is
.0852.
[10] Given that the mean
absolute correlation between the three national culture dimensions is .45, we
also examined their combined moderating effect. We reverse-scored power
distance and collectivism, converted each of the three national culture scores
to z-scores, and then summed the z-scores to create a scale (α = .70)
where higher scores indicate a national culture, based on national culture
perspectives, that would better fit an HPWS. Consistent with hypothesis 1,
countries with higher scores on this scale should have more strongly positive
HPWS-business performance effect sizes. However, the results of this analysis
confirmed our prior findings on the individual national culture dimensions with
HPWS-business performance effect sizes found to be statistically significantly
weaker (i.e., more negative or less positive) (B = -.044, p < .05) in
countries having higher scores on the scale. A change of -.044 in Fisher z transformed r translates to a change of -.044 in r. The standard deviation for the national culture scale is 1.98.
Thus, the change in r for a one
standard deviation change in the national culture scale is .08712 (-.044*1.98).
Given the mean unweighted HPWSbusiness performance correlation of .28 (see
Table 1), this -.044 coefficient indicates that the expected HPWS-business
performance effect size r would be
.37 in countries one standard deviation below the mean on the national culture
scale versus .19 in countries one standard deviation above the mean on the
national culture scale. Again, these results do not support hypothesis 1.
[11] Given the small number of
observations (countries), we computed correspondingly the shrunken r using the formula given by Cohen and
Cohen (1983, p. 106). These were 77 and .86, respectively.
No comments :
Post a Comment
Note: only a member of this blog may post a comment.