You are on page 1of 15

European Journal of Operational Research 181 (2007) 453467

www.elsevier.com/locate/ejor

Interfaces with Other Disciplines

Human resource management and performance:


A neural network analysis
Eleni T. Stavrou *, Christakis Charalambous, Stelios Spiliotis
University of Cyprus, Department of Public and Business Administration, 75 Kallipoleos Avenue, Nicosia, Cyprus
Received 24 May 2005; accepted 1 June 2006
Available online 4 August 2006

Abstract
This study utilizes an innovative research methodology (Kohonens Self-Organizing Maps) to explore a subject relatively understudied in Europe. It focuses on the connection between human resource management as a source of competitive advantage and perceived organizational performance in the European Unions private and public sectors. While
practices in these two sectors did not dier signicantly, three diverse but overlapping HRM models did emerge, each
of which involved a dierent set of EU member states. Training & Development practices were strongly related to performance in all three models and Communication practices in two. These results show the usefulness of an innovative technique when applied to research so far conducted through traditional methodologies, and brings to the surface questions
about the universal applicability of the widely accepted relationship between superior HRM and superior business
performance.
2006 Elsevier B.V. All rights reserved.
Keywords: Kohonen self-organizing map (SOM); Articial neural network (ANN); Human resource management; Performance

1. Introduction
The importance of human resource management
(HRM) as a competitive tool and the relationship
between human resource management and organizational performance has been the subject of systematic research (e.g. see Truss, 2001; Huselid,
1995; Delery and Doty, 1996; Becker and Gerhart,
1996). Researchers have identied and examined
certain HRM practices as crucial to developing
organizational competitive advantage (e.g. see
*
Corresponding author. Tel.: +357 22892480; fax: +357
22892460.
E-mail address: eleni1@ucy.ac.cy (E.T. Stavrou).

Flanagan and Deshpande, 1996; Pfeer and Veiga,


1999; Ferris et al., 1999). Researchers also found
connections between HRM and various measures
of organizational performance (e.g. see Truss,
2001; Huselid, 1995; Delery and Doty, 1996; Becker
and Gerhart, 1996).
However, the majority of those studies have been
conducted in the US. The role of HRM elsewhere,
especially within Europe, is understudied (see Guest
et al., 2003; Cunha et al., 2002; Wood, 1999).
Brewster (1995) notes the need for identifying
European models of HRM, pointing out that the
European business environment diers in many
respects from the US. European organizations
are less autonomous than US ones due to the

0377-2217/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejor.2006.06.006

454

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

Key HR
Practices linked
to Competitive
Advantage

GEOGRAPHIC
CONTEXT

HUMAN RESOURCE
MANAGEMENT FUNCTIONS
Human Resource Planning
Staffing
Training & Development
Compensation
Communication & Participation

ORGANIZATIONAL
PERFORMANCE
Service Quality
Productivity
Profitability

Superior-performer
organisations
Lower-performer
organisations

ORGANIZATION
CONTEXT
Private Sector
Wider Public Sector

Fig. 1. Human resource management as a competitive tool: a framework.

restrictions they face at national and organizational


levels from cultural environments, legislation, and
patterns of ownership; and at HRM levels from bargaining patterns, consultative arrangements and
trade union involvement (Brewster, 1995; Guest,
1990).
In addition, previous studies on this subject have
focused on the private sector to the neglect of the
public sector. However, Duey (1988) argues that
human capital can be the greatest source of competitive advantage in a nations eort to serve the public
and to be strategically eective within and beyond its
borders. This may be especially true within the EU
context where member states public sector practices
are inuenced directly by EU Directives as well as by
one another. As Wood (1999) argues, the issue of
environmental t and contextual variables should
be addressed directly in empirical studies.
Finally, existing literature on the subject has been
criticized as to the methods used to explore the link
between human resources and performance, suggesting that the quality of the research base supporting this link is relatively weak (Wood, 1999; Gerhart
et al., 2000a,b; Wright et al., 2001). Much of this
research base is conducted through the use of conventional statistical methods, which Guest (2001)
has described as a traditional research paradigm in
need of much closer scrutiny.

Given the above shortcomings in current


research, we attempt to use visualization techniques
based on Kohonens (1995) self organizing maps
(SOMs), to examine the relationship between
HRM as a competitive tool and organizational performance in both private and public sector organizations in the EU. We hypothesize that superior
performance will be associated with use of key
human resource management practices, and we set
out to explore which sets of practices work best in
EU private and public sectors. Fig. 1 serves as a
summary of the study.
The paper is organized as follows. Firstly, we
address the various parts of Fig. 1, examining the
connection between human resource management
as a source of competitive advantage and organizational performance and the geographic and organizational context in which this linkage may take
place. Subsequently, we describe the methodology
of the study and continue with a presentation of
its results and implications. We conclude with suggestions for future research.
2. HRM and performance within context
Cascio (1992) suggests that todays organizations
must gain competitive advantage through the eective utilization of their human resources. Competi-

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

tive advantage may be dened as the asymmetry or


dierential in any rm attribute or factor that allows
one rm to better serve the customers than others
and hence create better customer value and achieve
superior performance (Ma, 1999, p. 2). This denition points out that competitive advantage and
superior performance are two connected but dierent ideas: successful human resource management
can contribute to superior performance as a source
of competitive advantage by making organizations
more eective.
In recent years, those interested in connecting
human resource management with strategic management often look to typologies of competitive
advantage for the missing link (Boxall, 1996). Based
on these typologies, researchers have proposed a
number of human resource practices that could lead
to competitive advantage, although none has
proved better than the others in relation to organizational competitiveness. As Ferris et al. (1999)
note, extant research is subject to many of the same
limitations and oers little consensus with regard to
precisely which HRM practices should be included.
In addition, questions are raised in the literature as
to whether these practices are universal or context
specic, and if the latter, which practices are appropriate for achieving t in which contexts (Ferris
et al., 1999; Boxall and Steeneveld, 1999).
Regardless of the typology used, HR practices
connected to competitive advantage fall into a number of main categories. We use Stavrou and
Brewsters (2005) key practices groupings of Human
Resource Management Functions: Planning, Staing, Training & Development, Compensation &
Benets, and Communication & Participation.
Looking at categories or groups rather than individual HR practices as sources of competitive advantage may be more requisite since according to
Ma (1999:18), the more compound a competitive
advantage, the more likely it is to have direct implications in the causal chain of performance analysis.
In fact, researchers have positively linked HR practices to organizational performance (Truss, 2001;
Pfeer and Veiga, 1999; Huselid, 1995). Specically,
human resource management has been linked to
increased productivity (Huselid, 1995; Ichniowski
et al., 1997; Fox et al., 1999), good customer service
(Fox et al., 1999), improved eciency (Becker and
Gerhart, 1996), increased rm value, greater protability (Delery and Doty, 1996; Fox et al., 1999) and
overall organizational survival (Welbourne and
Andrews, 1996). Anderson et al. (1997) suggest

455

organizations must pursue superiority in both customer satisfaction and productivity, while Savery
(1998) and Singh et al. (2000) note that increasing
global economy pressures compel organizations to
place ever-greater emphasis on productivity
improvement and Zeithaml et al. (1996) emphasize
the importance of service quality on organizational
performance: all these are sectors where good HR
practices can help give organizations the edge.
The above discussion points to Service Quality,
Productivity and Protability as commonly recognized measures of organizational performance.
Previous research has tended to use them either as
separate measures of organizational performance
(Baltes et al., 1999; Huselid, 1995; MacDue,
1995; Perry-Smith and Blum, 2000), or as benchmarks combined into a composite performance
measure (Sherman et al., 2003). Delaney and Huselid (1996) used two composite perceptual measures
of organizational performance, which included service quality and protability, by asking informants
to assess it relative to that of industry competitors.
In disciplines such as management accounting,
benchmarking measures calculate performance by
combining the three, and have provided superior
results to traditional one-dimensional performance
assessment methods (Sherman et al., 2003). For
the purposes of this study, we have considered as
superior-performers those organizations that perform at the top 10% in all three measures of performance, and labeled those that do not perform at the
top 10% in all three measures of performance as
lower-performers.
2.1. Geographic context
The relevant literature on the subject mainly
involves the US context (Truss, 2001), and the contribution of human resource management towards
organizational eectiveness in Europe has not yet
been the subject of systematic research (Sparrow
and Hiltrop, 1997). However, several researchers
have developed a number of mainly descriptive
international patterns trying to understand that role
around the world. In her study of secondary data
analysis, Filella (1991) found three dierent patterns
of human resource management within the European Union. Her Latin group comprised Portugal,
Spain, southern France, Corsica, Sardinia and Italy,
the UK, the Netherlands, Germany and Switzerland
made up the central European group while the Nordic group included Denmark, Norway and Sweden.

456

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

This separation is similar to that of Hofstedes


(2001) more general cultural classication, which
suggested the presence of Latin, Anglo-Saxon and
North European cultures.
In a dierent descriptive study of secondary data
analysis, Sparrow et al. (1994) developed a worldwide
pattern about the importance of human resource
management practices for organizations. They
formed ve clusters of countries, which were analyzed
across fteen areas of human resource management.
The rst cluster included countries of the AngloSaxon business culture (the UK, Australia, Canada
and the USA), while the second cluster consisted
solely of France and the third of Korea. The fourth
cluster included Brazil, Mexico and Argentina, and
the fth consisted only of Japan. Moss-Kanters
worldwide survey of HRM-relevant issues (quoted
in Sparrow and Hiltrop, 1997) reaches similar conclusions and identies similar groupings, with the main
exception of clustering Germany in with France.
These dierent models and geographic clusters of
human resource management practice may reect
the dierent foci of the researchers, as well as a variety of cross-border political, economic, social and
cultural considerations which create convergence
or divergence among various business structures
practices. Sparrow and Hiltrop (1997) claim that,
while a clear model of European human resource
management does not exist yet, many large and
some medium-sized European successful organizations already display some distinct pan-European
forms of human resource management principles.
2.2. Organizational context
Besides the geographical context, Brewster et al.
(1993) report that the relative strength of the private
and public sectors across Europe represents an
important context. Countries like Italy have particularly large public sectors that are institutionally
separate from private sectors, leading to a social
focus in public sector personnel management, while
in some other European countries the separation
may not be so distinct (Filella, 1991). In either situation, the traditional human resource management
role of record keeping and dealing with functional
personnel issues in a public organization is being
questioned due to the changes that both private
and public organizations have been facing
(McHugh et al., 1999; Klingner and Lynn, 1997).
As a result, researchers and administrators have
debated the appropriateness of transferring eective

managerial processes from the private to the public


sector, based on the question of whether the two
sectors are fundamentally dissimilar or not. But a
critical evaluation of 34 empirical studies on the
subject found little evidence of fundamental dierence, and suggested that public sector managers
can derive useful lessons from private sector management provided they understand the determinants
of private sector performance clearly (Boyne, 2002).
Along similar lines, Braddon and Foster (1996) suggest a strategy through which private sector management processes are used towards a more
eective and ecient public sector, a transition
which can be facilitated through the appropriate
human resource management functions and their
contribution towards organizational eectiveness
(Brown et al., 2000).
3. The study
As noted above, the majority of extant research
relating to the role of HRM as a competitive tool
is rooted in the US private sector; nevertheless, it
provides a foundation for exploring the European
private and public sector context as well. Our general aim is to use SOMs to help understand the relationship between human resource management and
organizational performance among organizations in
the private and wider public sectors of the EU. We
propose the conceptual framework shown in Fig. 1
to explore the following research questions:
1. Do superior-performer organizations form a specic pattern that distinguishes them from lowerperformer ones on the basis of the HR functions
identied in Fig. 1? If so:
(a) Is this pattern the same throughout the
European Union or does it dier?
(b) Is this pattern similar or dierent in the private and the wider public sectors?
2. Which HR functions contribute most to the differentiation between superior and lower performing organizations?

3.1. Measures
The independent variables Key HR Practices,
grouped into ve Human Resource Management
Functions, are operationalized into 80 questions
measured on a binomial scale as to whether the specic practice is used or not (yes or no).

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

The Organization Context variable is also binomial (private versus wider public sector). Specically, we separate organizations into those
for-prot private sector businesses and those organizations and services that belong, all or in part to
the government.
The Geographic Context involves the countries
comprising the European Union prior to May
2004. Luxemburg was not included as it did not
have sucient organizations, and organizations in
the former East and West Germany were studied
separately as we wanted to take into account the
socio-economic dierences of the two former states.
Finally, the dependent variables Organizational
Performance measures are service quality, productivity and protability. These three measures
involve the perceptions of study participants as to
the performance of their organization in comparison with that of competitors leading to their placing
as superior (top 10% in all three measures) organizations and lower-performers (not top 10% in all
three measures). The use of a perceptual performance measure, even though not optimal, is acceptable and consistent with prior research (e.g. Smith
and Barclay, 1997; Perry-Smith and Blum, 2000;
Jap, 2001). Pearce et al. (1987) found that senior
managers perceptions of performance were consistent with nancial and other measures. The choice
of 10% as the minimum for excellence relates this
study to other disciplines which use a similar denition within-100% evaluation scales (Hax and Wilde,
1999; Rank and Hirschl, 2001; Kerr and Beaujot,
2002; Blum and Clegg, 2003).
3.2. Sample
The present study draws on data generated by the
Cranet Network questionnaire (see Brewster et al.,
2004). The samples in each country have been
selected from lists provided by the national federations, such as chambers of commerce or national statistical services. Researchers ensure that all sectors of
the economy are represented so that samples in the
Cranet database are demonstrative of the population of organizations in each country. The overall
response rate averaged 17%. For the present study,
the nal sample for the EU member states was
4759, of which 3559 were private sector businesses
and 1200 were public sector (governmental or
semi-governmental) organizations. The unit of analysis was the organization and the respondent was the
highest-ranking ocer from the corporate HR team.

457

Although data for this study come from a single


source, the Cranet research teams data collection
methods, include certain steps to minimize the eects
of single-method bias, as suggested by Podsako
et al. (2003). Respondents were guaranteed anonymity to increase the accuracy of the responses to the
questionnaire. Criterion measures were placed in different sections of the questionnaire from predictor
variables which were posed at the end of the questionnaire after a set of demographic variables.
Finally, criterion variables were posed in a set of different response formats from predictor variables.
In addition to these safeguards, Huselid and
Becker (2000) explain that the validity of singlesource measures depends on the size of organizations in the sample, the expertise of the source
responding to the questions and the clarity of items
comprising the survey. The Cranet survey meets
these requirements: the average number of employees in the organizations of our study was 1546; the
respondents were members of the corporate HR
team; and the international Cranet team took great
care in the methods and procedures used to make
the questionnaire specic and clear, leaving little
room for ambiguity.
3.3. Procedure
This study explores the HR-performance relationship within the EU through Kohonens Self
Organizing Map (SOM). While new to the eld of
human resource management, this methodological
approach has been used successfully across dierent
disciplines and lines of research (Moreno et al.,
2006; Deichmann et al., 2003; Veiga et al., 2000;
Smith, 1999; Mazanec, 1995).
Self-Organizing Maps (SOMs) belong to the
broad category of unsupervised neural networks
and the basic idea behind them is competitive learning (Kohonen, 1995). This procedure is superior to
other methods. Thus it provides a visual representation of the relationships that exist in the original
data, while avoiding creating articial clusters
(Deichmann et al., 2003; Mazanec, 1995).
SOMs have additional advantages which make
them appropriate in the current study. First, they
are able to outperform the traditional data reduction
and clustering techniques, in both speed and quality
of solution (Smith, 1999). Second, they have the
capacity to operate on very large samples and need
no a priori assumptions about the distribution of
the sample (Deboeck and Kohonen, 1998). Using

458

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

SOMs helps overcome structuring task problems


associated with nding the appropriate underlying
distribution and the functional form of the underlying data. (Such problems are often encountered, for
example, when using cluster analysis.) Finally, SOMs
are useful in visually examining the relationship
between input data and identifying important patterns and clusters. In the current research, it was
the most appropriate tool to examine the relationship
between superior- and lower-performing organizations, public and private ones, and the HR-performance relationship in dierent EU countries.
The basic SOM consists of M neurons located on
a low- (usually two-) dimensional grid. The twodimensional map grid of neurons (neuron space)
can be visualized to reect spatial properties of the
original data (Kohonen, 1995). The SOM algorithm
is iterative. Each neuron i has a d-dimensional
(d represents the number of variables) prototype
vector mi = [mi1mi2    mid] which represents a sample unit in the original data set. Prototype vectors of
the SOM are trained in such a way that they represent the original data set in a substantially smaller
size and they can form a map where each prototype
vector is a unit (neuron) of that map.1
At each such training step, one sample unit x
from the original data set is chosen randomly and a
similarity measure is calculated between it and all
the prototype vectors of the map. The simplest
way is to nd the neuron whose prototype vector
is closest to the sample unit in terms of Euclidean
distance. This neuron is called the Best Matching
Unit (BMU) for that sample unit. Next, the SOM
is trained by updating the prototype vectors, moving the prototype vectors of the BMU and its topological neighbors closer to the sample unit in the
original data set. The update rule for the prototype
vector of neuron i is
mi t 1 mi t at  hbi t  x  mi t
where t denotes discrete time (t = 0, 1, 2, . . .), a(t)
is the learning rate (a suitable monotonically
decreasing function of t, 0 < a(t) < 1), and hbi(t) is
a neighborhood Kernel in the two-dimensional
map grid of neurons centered on the BMU. In par2
2
ticular, a(t) = a(0) * (1  t/T) and hbi t ed bi =2rt
(Gaussian neighborhood function). T is the number
of training steps, a(0) is the initial learning rate, r(t)
1
SOMs and other visual display maps were implemented using
SOM-toolbox software, which can be implemented in the Matlab
program.

is the neighborhood radius at time t, and dbi is the


distance between neuron b(BMU) and neuron i on
the map grid.
During training, the SOM behaves like a net that
folds onto the cloud formed by training data, taking up its exact shape. Because of neighborhood
relationships, prototype vectors of neighboring neurons are pulled to the same direction, and thus prototype vectors of neighboring neurons resemble
each other. The most ecient tools in the visualization of the cluster structure of SOM are distance
matrices, the most widely used of which is the
U-matrix. The U-matrix holds the Euclidean distance in the original data set between the prototype
vectors of a neuron with its neighbors in the neuron
space, as well as the average Euclidean distance in
the original data set of each map neuron to its
neighbors (Ultsch and Siemon, 1990). [For a demonstration of the methodology followed, please
refer to the Appendix.]
4. Results
First we explored the data to uncover clustering
patterns among EU organizations by training an
SOM on the whole sample of the 4759 organizations.2 The SOM consisted of 360 neurons on a 24
by 15 map grid, with hexagonal lattice and
Gaussian neighborhood function. The resulting
U-matrix visualization, which is created on the
whole sample, is shown in Fig. 2, and acts as the
fundamental graphic display for identifying clusters
of organizations based only on the way they use the
sets of HR practices. High values are represented by
darker shades and low values by lighter ones. High
values on the U-matrix indicate cluster borders and
uniform areas of low values indicate clusters themselves. From this gure, we identied four regions
labeled 1, 2, 3 and 4 which represent dierent
combinations of intensity of the ve HR functions
used by EU organizations.
Given the results in Fig. 2, we visualized the position of superior and lower-performer organizations
on the map for each EU member state separately.3

2
The whole data sample was used for the purpose of training
Kohonens SOM. However, only top-performers and lower
performers (as dened in Section 3.1) were used for subsequent
analysis on the maps and other visualization.
3
These visualizations were too numerous to be depicted in the
present study. However they are available on request from the
lead author.

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

459

Fig. 3. Three dierent HR Clusters among EU superior-performer organizations. The bottom of the map, which corresponds
to region 3 of Fig. 2, is covered by all models and includes the
majority of superior-performer organizations, while the top of the
map, which corresponds to region 1 of Fig. 2, includes the fewest
superior-performer organizations.

Fig. 2. The whole-sample SOM and the U-matrix visualization.

This allowed us to visualize the relationship of key


HR practice use to superior/lower performance on
the map for each EU member state separately, as
well as their relationships with organizational
groupings shown in Fig. 2. Fig. 3ac are summaries which represent the emerging common geographical clusters of this relationship. From these
visualizations, superior-performers among EU
member states formed three clusters (Clusters A, B
and C, see Fig. 3) which represent similar combinations of use of the ve HR functions. These encompass parts of regions 2, 3 and 4 in Fig. 2. Region 3 of
Fig. 2 is covered by all models and includes the
majority of superior-performer organizations, while
region 1 includes the fewest superior-performer
organizations.
Cluster A, located in the bottom-left of the map
(Fig. 3a), includes superior-performers in Sweden
and Finland; Cluster B, in the bottom half of the
map (Fig. 3b) includes superior-performers in the
UK and the former W. Germany; while Cluster C,
in the bottom-right of the map (Fig. 3c) includes
superior-performers in Greece, Portugal, Spain,
Italy, France, Ireland and Belgium. Denmark, Austria, The Netherlands and the former East Germany
(GDR) did not appear in any clusters.
Next, we proceeded to explore whether these
clusters were similar for both private and public sector EU organizations. These results are concentrated and shown in Fig. 4, which shows the

position of organizations on the U-matrix by cluster


(also demonstrating a graphical conrmation of the
three clusters), type (private versus public) and performance (superior-performers versus lower-performers). According to Fig. 4, with the exception
of public sector superior-performer organizations
in Cluster B (Fig. 4g), superior-performer organizations in the public and private sectors follow the
basic shape relating to their respective geographic
Cluster as depicted in Fig. 3. Furthermore, even
though no specic pattern is found for public superior-performers of Cluster B, none is located in what
corresponds to region 1 of Fig. 2.
Contrasting the three clusters of public organizations in Fig. 4, Cluster C organizations provide the
best results both in terms of concentration of superior-performers and dierentiation between superior
and lower performers. More specically, 80% of
superior-performer organizations are located in
Cluster C whereas only 52% of lower performers
are located in the same area. The next best results
are obtained from Cluster A organizations (65%
superior-performers and 53% of lower-performers),
while Cluster B locates 60% of superior-performers
and 50% of lower-performers.
Comparing the three clusters of private organizations of Fig. 4, organizations in Cluster B provide
the clearest dierentiation between superior and
lower-performer organizations, with 76% of superior-performers compared to 52% of lower-performers. Cluster C located 84% of superior-performers
and 72% of lower-performers, while Cluster A
located 71% of superior-performers and 60% of
lower-performers. The country specic analysis
revealed that the largest dierence between superior

460

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

Fig. 4. Visualization of the position of organizations by cluster, type and performance. Organizations are represented in the gures as
dots. The size of the dots is analogous to the number of organizations located in a specic area of the map. (a), (c), (e), (g), (i) and (k) show
the position of top-performer private and public organizations; (b), (d), (f), (h), (j) and (l) show the position of lower-performer
organizations.

and lower-performer organizations for a given country corresponds to the former W. Germany. Fig. 4(1)
illustrates the specic visualization of the W. German organizations position on the U-matrix, with
(a) showing superior-performing private organizations, and (b) showing private lower-performers.
Fig. 4(1)(a) shows that 74% of superior-performer organizations in the former W. Germany
are located in Cluster B (i.e. at the bottom of the
Fig. 2 map, mainly in region 3), but only 34% of
lower-performers are located in the same area. The
majority of lower-performers are located in region
1 of Fig. 2, where only a very small percentage of
superior-performers are located in that area. These
results lead us to conclude that organizations of
Cluster B in the former W. Germany that follow
the HR practices used in this study as competitive
tools, have 74% probability of being superior-performers as against 34% of being lower-performers.
Overall, the concentration of private and public
superior-performer organizations in their respective

clusters is high, especially for private organizations


of Clusters B and C and for public organizations
of Cluster C. The results for the public organizations compared to the private ones were less pronounced, both in terms of concentration of
superior-performers in their corresponding cluster
and of dierentiation between superior and lowerperformers. The fact that lower-performers were
found in all three clusters, reminds us that HR practices are by no means the only factors that dene
superior/inferior organizational performance.
The above visualizations lead us to the following
conclusions regarding our rst research question:
(a) three dierent HRM models of competitive
advantage exist which correspond to the three specied clusters of EU countries; (b) HRM models are
similar in the private and public sectors. However,
the probability of becoming superior organization
following one of the three dierent HRM models
diers. Those organizations classied in Cluster B
(UK and W. Germany) appear to have a greater

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

461

Fig. 4(1). Visualization of West German private sector organizations on the U-matrix (a) the position of superior-performer
private organizations on the U-matrix; (b) the position of lowerperformer private organizations on the U-matrix.

chance (compared to organizations in the other two


clusters) of becoming superior performers by following their clusters HRM model. Also private sector organizations appear to have a greater chance
than public sector bodies of becoming superior performers by following their clusters HRM model.
The clustering pattern of SOM and the three clusters created, dierentiating superior from lower-performer organizations, are strongly aected by certain
HR functions. Therefore, to answer the second
research question, we explored which of the ve categories of HR functions contribute most to the creation of the cluster structure of the SOMs (see Fig. 5).
According to this gure, Training & Development and Communication & Participation are the
most important functions among superior EU organizations, while Planning is less important. When
the above gure is examined in light of the three
clusters, it becomes clear that Training & Development is a very important function in contributing
to clusters B and C and the only signicant function
for Cluster C. The Communication & Participation
function is most important for Cluster A and of secondary importance to Cluster B. Planning, though
less signicant, seems to contribute most to Clusters
B and A.
5. Discussion, implications and conclusions
The contribution of this study is twofold: it
shows the utility of Kohonens SOM to HRM
research and it advances the extant knowledge and
research regarding the HR-performance relationship. First, we have demonstrated how Kohonen

Fig. 5. (a) The Fig. 2 U-matrix visualization using all the


prototype vectors. (b)(f) visualize the U-matrix using prototypes
vectors calculations of the components corresponding to individual key HR functions. The contribution of a variable or a set
of variables to the SOM cluster structure is found by comparing
(a) U-matrix and the U-matrices of (b)(e).

maps can aid in understanding the overall pattern


of human resource practice in the EU and creating
new possibilities of exploration within the eld of
human resource management. The concept of
SOM is one of the most elegant examples of unsupervised learning, where an articial neural network
attempts to extract stable features or prototypes

462

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

from a database without an outside teacher


(Kohonen, 1995). This procedure is superior to
other methods, not in only reducing multi-dimensional data through clustering, but also projecting
them non-linearly in a two-dimensional map
(Vesanto, 1999). Furthermore, it detects clusters
existing in the original data while avoiding creating
articial ones, thus providing a true representation
of the original datas characteristics. SOMs may
be viewed as a combination of principal components and cluster analyses with both procedures
inuencing each other in the algorithm (Deichmann
et al., 2003). More specically, SOMs provide an
intuitively useful method of visualizing a dataset
that is otherwise too amorphous and complex to
conceptualize.
Perhaps the most outstanding feature of this
method for this study is the nature of clustering of
the specic countries on the map, clearly reecting
spatial patterns in the vectors of non-spatial variables. Following the work of Kaski et al. (1998),
no geographical data were entered into the model,
but remarkably, the countries organized themselves
in a manner that mimicked their relative cultural
and geographical locations, with geographical neighbors and culturally similar pairs being closely located
in the U-matrix, revealing very similar HR practices.
Through this analysis we were also able to gain
insights also on the positions of individual countries
in clusters that did not match our geographic expectations. Thus, geographically, Ireland should fall in
a cluster with the UK, but a closer examination of
Irish culture reveals closer relations with Cluster C
rather than Cluster B (Hofstede, 2001; Ignjatovic
and Sveltic, 2003). Similarly Denmark would
appear to fall culturally and geographically into
Cluster A, yet organizations in Denmark appeared
to form no clustering. Finally, while superior W.
German organizations fall into the same cluster
with those of the UK, they dierentiate themselves
as a model to follow within the EU in utilizing
HR as a competitive tool. First they have a much
higher percentage of superior organizations using
these key HR practices as functions of competitive
advantage, and second, these superior organizations
are better dierentiated from the lower-performers.
Second, in relation to furthering HR knowledge,
this study shows that geographic context does
appear to make a dierence to the HR-performance
relationship, but that organizational context does
not: at the organizational level, these results clarify
that EU public sector organizations can utilize HR

as a competitive tool in a manner not inherently different from their private counterparts. In turn,
human resource management techniques may be
exported successfully from one sector to another
within the context of superior performance. For
the EU, which depends heavily on national public
sectors and strives to become the most competitive
and dynamic knowledge-based economy in the
world, capable of sustainable economic growth with
more and better jobs and greater social cohesion,
these results are very encouraging.
At the EU level, three clear yet overlapping clusters are formed on the basis of the HR-performance
relationship. Specically, the set of Training &
Development is of primary importance among superior organizations in the clusters. Overall, this is consistent with existing research suggesting that training
and development add value to an organization by
maximizing productivity; enriching employee skills;
and helping it confront external pressures more
eectively (Peteraf, 1993). In addition, superior
organizations in Clusters A and B emphasize Communication & Participation. This does not hold true,
however, for Cluster C, perhaps because in organizations in these countries hierarchies may prevail over
atter structures, and thus place less emphasis on the
potential contribution of communication as a competitive tool. As Pfeer and Veiga (1999) note, sharing information is a very important component of
high performance work systems and consequently
of achieving competitive advantage.
Finally, in contrast to existing research ndings,
Planning appeared not to be a very strong consideration among superior EU organizations of any cluster. Maybe HRM does not yet hold a clearly
strategic role among EU organizations, leaving
HR somewhat disconnected from organizational
strategy and performance (Truss, 2001). Or, maybe
the key here is responsiveness to planning, as the
eects of planning on performance will be evident
(and measurable) later than when it actually takes
place (Brewster et al., 2004). In addition to Planning, Compensation & Benets and Stang practices
appear to have weak relationships with superior
performance in EU organizations. It is quite possible that the former are too contingent on the socioeconomic and regulatory structures in Europe to
have any signicant impact on performance, while
the latter is contingent on the generally high unemployment in Europe (Brewster et al., 2004).
Based on the above, further research should be
conducted in order to investigate specic parts of

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

the maps and develop greater understanding of the


specic dierences among clusters and among countries within clusters. Also, SOMs can be utilized to
examine how organizations have changed their
HR practices over the years and their eect in organizational performance. Finally, while SOM clusters
of HR practices in the EU closely mimic their cultural and geographic proximity, some exceptions
were noted in this study. This raises the possibility
that other types of HR-related distances such as
political, administrative and industrial eects
may inuence country classications. Therefore,
future research should include measures in addition
to HR practices per se as explanatory variables in
the SOM algorithm.
Such research may facilitate methodological
advancements and conceptual as well as practical
breakthroughs in the study of HRM. Beyond the
research sphere, the hope must be that increased
understanding will lead to a better utilization of
HRM as a tool for increasing competitiveness in
the private sector and service quality as well as productivity in the public sector, as well as improving

463

the working conditions of work-forces of all grades,


both in the EU and around the globe.
Acknowledgements
This research was supported in part by a University of Cyprus grant on HRM and Competitiveness.
We would like to express our appreciation to our
colleagues at the EU member countries part of Cranet who gave us permission to use their data. We
would also like to thank the two anonymous reviewers and Roman Slowinski the editor of the European Journal of Operational Research as well as
our copy editor Jonathon Morgan, for their invaluable contributions towards improving the quality
and rigor of this manuscript.
Appendix. Technical demonstration of the SOM
procedure
We would like to demonstrate the SOM procedure with an example. In the example, the number
of sample units (organizations) is 300 and the num-

Fig. 6. The application of SOM algorithm and its visualizations on a sample of 300 organizations by the use of 25 neurons. (a) The
original data set of 300 organizations, (b) two-dimensional map grid neuron space, and (c) the training procedure.

464

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

ber of variables is two. This data set forms three


natural clusters (see Fig. 6a). For demonstration
purposes, we chose a two-dimensional example,
where the clusters in the original data are visible
without visualization though SOM. When the original data set has more than two dimensions, especially multi-dimensional data sets, it is impossible
to see the clusters without further analysis. Using
the U-matrix, multi-dimensional information is
reduced visually into two dimensions, making the
data easy to assess.
When visualizing these data through SOM, a
two-dimensional hexagonal map grid with 25 neu-

rons (M = 25) is created (see Fig. 6b). Each neuron


on this map has a two-dimensional prototype vector
in the original data set. At each training step, one
sample unit x from the original data set is chosen
randomly and its Best Matching Unit (BMU) is
found. Next, the BMU and its topological neighbors are moved closer to the sample unit in the original data set through the procedure demonstrated
in Fig. 6(c). The position of prototype vectors (black
dots) of the neurons after training the SOM is
shown in Fig. 7(a). One can easily see that prototype
vectors have been concentrated in the area of the
three clusters that existed in the original data.

Fig. 7. The output of the SOM algorithm and the clustering of data via the U-matrix. (a) The original data set with the prototype vectors
(after training), (b) the two-dimensional map grid neuron space, and (c) the U-matrix, showing the clustering structure of the data set.

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

465

Fig. 8. The U-matrix and the contribution of a variable to the cluster structure of SOM (a) the U-matrix, showing the clustering structure
of the data set, (b) the U-matrix by using in calculations variable 1, and (c) the U-matrix by using in calculations variable 2.

In addition, and most importantly, neurons


whose prototype vectors belong to the same cluster
in Fig. 7(a) form a cluster in the neuron space in
Fig. 7(b). A way of identifying the three clusters in
the neuron space (and hence in the original data
set) is by using the U-matrix. The interpretation of
the U-matrix is done as follows: High values on
the U-matrix indicate cluster borders and uniform
areas of low values indicate clusters themselves.
High values are represented by darker shades and
low values by lighter ones. The U-matrix visualization is shown in Fig. 7(c). This map grid consists
of a 9 by 9 grid of hexagonal cells. The shade of
these cells shows the distance as demonstrated by
the scale on the right side denoting the value of each
shade. In the specic example, the U-matrix shows
three clusters in the data. In the U-matrix, the shade
of the numbered hexagons represents the average
distance between the prototype vector of the corresponding neuron and the prototype vectors of the

neighboring neurons. To illustrate, the shade of


the hexagon denoted with the number 7 in
Fig. 7(c) represents the average distance between
the prototype vector of neuron 7 in Fig. 7(b) and
the prototype vectors of neighboring neurons,
namely 6, 2, 8, 13, 12, and 11. Finally, the shade
of the hexagons that are not numbered represents
the distance between the prototype vectors of the
two adjacent neurons on the U-matrix. For example, in Fig. 7(c), the dark shade of the hexagon
between hexagons denoted with numbers 11 and
16 shows that the distance of prototype vectors
between neurons 11 and 16 is large. In turn, 11
and 16 belong to two dierent clusters.
An interesting question then is where on the map
a specic data sample is located. The simplest
answer is to nd the BMU of the data sample. From
Fig. 7(a), we selected a sample unit from Cluster 1
and named it Company A. As we can see from
Fig. 7(a), Neuron 6 is the closest to Company A.

466

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467

Next, the position of Company A is visualized in


Fig. 7(c) and again is located in the area of cluster 1.
Finally, the contribution of a variable to the cluster structure of data is also very important. To
achieve this, the U-matrix of each variable should
be visualized and compared to the U-matrix of all
variables together, as shown in Fig. 8. By comparing
Fig. 8(a) and (b), it can be concluded that variable 1
contributes to the creation of the three clusters. By
comparing Fig. 8(a) and (c), it can be concluded
that variable 2 contributes to the creation of cluster
2.
References
Anderson, E.W., Fornell, C., Rust, R.T., 1997. Customer
satisfaction, productivity, and protability: Dierences
between goods and services. Marketing Science 16 (2), 129
145.
Baltes, B.B., Briggs, E.T., Hu, W.J., Wright, A.J., Neuman,
A.G., 1999. Flexible and compressed workweek schedules: A
meta-analysis of their eects on work related criteria. Journal
of Applied Psychology 84, 496514.
Becker, B., Gerhart, B., 1996. The impact of human resource
management on organizational performance: Progress and
prospects. Academy of Management Journal 39 (4), 779802.
Blum, E., Clegg, R., 2003. Percent plans: Admissions of failure.
The Chronicle of Higher Education 49 (28), B10B11.
Boyne, G.A., 2002. Public and private management: Whats the
dierence? Journal of Management Studies 39 (1), 97122.
Boxall, P., 1996. The strategic HRM debate and the resourcebased view of the rm. Human Resource Management
Journal 2, 6079.
Boxall, P., Steeneveld, M., 1999. Human resource strategy and
competitive advantage: A longitudinal study of engineering
consultancies. Journal of Management Studies 36, 443463.
Braddon, D., Foster, D., 1996. Privatization: Social Science
Themes and Perspectives. Dartmouth Publishing, Aldershot.
Brewster, C., 1995. Towards a European model of human
resource management. Journal of International Business
Studies 26 (1), 121.
Brewster, C., Hegewisch, A., Lockhart, J.T., Holden, L., 1993.
The European Human Resource Management Guide. Academic Press, London.
Brewster, C., Mayrhofer, W., Morley, M., 2004. Trends in
Human Resource Management in Europe: Convergence or
Divergence. Butterworth, Heinemann, UK.
Brown, K., Ryan, N., Parker, R., 2000. New modes of service
delivery in the public sector: Commercializing government
services. The International Journal of Public Sector Management 13 (3), 206221.
Cascio, W.C., 1992. Managing Human Resources: Productivity,
Quality of Work Life, Prots. McGraw-Hill, New York.
Cunha, R.C., Cunha, M.P., Morgado, A., Brewster, C., 2002.
Market impacts on strategy, HRM practices and organizational performance: Toward a European model. Management
Research 1 (1), 7991.
Deboeck, G., Kohonen, T., 1998. Visual Explorations in Finance
with Self-Organizing Maps. Springer, London.

Deichmann, J., Eshghi, A., Haughton, D., Sayek, S., Teebagy,


H., Topi, H., 2003. Geography matters: Kohonen classication of determinants of foreign direct investment in transition
economies. Journal of Business Strategies 20 (1), 2325.
Delaney, T.J., Huselid, A.M., 1996. The impact of human
resource management practices on perceptions of organizational performance. Academy of Management Journal 39,
949969.
Delery, J.E., Doty, D.H., 1996. Modes of theorizing in strategic
human resource management: Tests of universalistic, contingency, and congurational performance predictions. Academy of Management Journal 39 (4), 802835.
Duey, J., 1988. Competitiveness and human resources. California Management Review 30 (3), 92101.
Ferris, G.R., Hochwarter, W.A., Buckley, M.N., Harrell-Cook,
D., Frink, D., 1999. Human resources management: Some
new direction. Journal of Management 25 (3), 385418.
Filella, J., 1991. Is there a Latin model in the management of
human resources? Personnel Review 23 (20), 1423.
Flanagan, D.J., Deshpande, S.P., 1996. Top managements
perceptions of changes in HRM practices after union elections
in small rms: Implications for building competitive advantage. Journal of Small Business Management 34 (4), 2335.
Fox, D., Byrne, V., Rouault, F., 1999. Performance improvement: What to keep in mind. Training & Development 53 (8),
3849.
Gerhart, B., Wright, P.M., McMahan, G.C., 2000a. Measurement error in research on the human resources and rm
performance relationship: Further evidence and analysis.
Personnel Psychology 53 (4), 855872.
Gerhart, B., Wright, P.M., McMahan, G.C., Snell, S.A., 2000b.
Measurement error in research on human resources and
performance: How much error is there and how does it
inuence eect size estimates? Personnel Psychology 53 (4),
803834.
Guest, D.E., 1990. Human resource management and the
American dream. Journal of Management Studies 27 (4),
377397.
Guest, D.E., 2001. Human resource management: When research
confronts theory. International Journal of Human Resource
Management 12 (7), 10921106.
Guest, D.E., Michie, J., Conway, N., Sheehan, M., 2003. Human
resource management and corporate performance in the UK.
British Journal of Industrial Relations 41 (2), 291314.
Hax, A.C., Wilde II, D.L., 1999. The delta model: Adaptive
management for a changing world. Sloan Management
Review 40 (2), 11.
Hofstede, G., 2001. Cultures Consequences: Comparing Values,
Behaviors, Institutions and Organizations across Nations.
Sage Publications, London.
Huselid, M.A., 1995. The impact of human resource management
practices on turnover, productivity, and corporate nancial
performance. Academy of Management Journal 38 (3),
635672.
Huselid, M.A., Becker, B.E., 2000. Comment on measurement
error in research on human resources and rm performance:
How much error is there and how does it inuence eect size
estimates? Personnel Psychology 53 (4), 835854.
Ichniowski, C., Shaw, K., Prennushi, G., 1997. The eects of
human resource management practices on productivity: A
study of steel nishing lines. American Economic Review 87
(3), 291312.

E.T. Stavrou et al. / European Journal of Operational Research 181 (2007) 453467
Ignjatovic, M., Sveltic, I., 2003. European HRM clusters. ESB
Review 17, 2539.
Jap, S.D., 2001. Pie Sharing in complex collaboration contexts.
Journal of Marketing Research 38 (1), 86100.
Kaski, S., Nikkila, J., Kohonen, T., 1998. Method for interpreting a self-organizing map in data analysis. In: Verleysen, M.,
Proceedings of ESANN98, 6th European Symposium on
Articial Neural Networks, D-facto, Brussels, pp. 185190.
Kerr, D., Beaujot, R., 2002. Family relations, low income and
child outcomes: A comparison of Canadian children in
intact-, step-, and lone-parent families. International Journal
of Comparative Sociology 43 (2), 134153.
Klingner, D.E., Lynn, D.B., 1997. Beyond civil service: The
changing force face of public personnel management. Public
Personnel Management 26 (2), 157174.
Kohonen, T., 1995. Self-organizing Maps. Springer, Berlin.
Ma, H., 1999. Creation and preemption for competitive advantage. Management Decision 37 (3), 259267.
MacDue, P.J., 1995. Human resource bundles and manufacturing performance: Organizational logic and exible production systems in the world auto industry. Industrial and
Labor Relations Review 48, 197221.
Mazanec, J.A., 1995. Positioning analysis with self-organizing
maps: An exploratory study on luxury hotels. Cornell Hotel &
Restaurant Administration Quarterly 36 (6), 8096.
McHugh, M., OBrien, G., Ramondt, I., 1999. Organizational
metamorphosis led by front line sta. Employee Relations 21
(6), 556576.
Moreno, D., Marco, P., Olmeda, I., 2006. Self-organizing maps
could improve the classication of Spanish mutual funds.
European Journal of Operational Research 174 (2), 1039
1054.
Pearce, J.A., Robbins, D.K., Robinson, R.B., 1987. The impact
of grand strategy and planning formality on nancial performance. Strategic Management Journal 8, 2534.
Perry-Smith, J.E., Blum, T.C., 2000. Work-family human
resource bundles and perceived organizational performance.
Academy of Management Journal 43, 11071117.
Peteraf, M.A., 1993. The cornerstones of competitive advantage:
A resource-based view. Strategic Management Journal 14 (3),
179192.
Pfeer, J., Veiga, J.F., 1999. Putting people rst for organizational success. The Academy of Management Executive 13
(2), 3738.
Podsako, P.M., Mackenzie, S.B., Lee, J.Y., Podsako, N.P.,
2003. Common method biases in behavioural research: A
critical review of the literature and recommended remedies.
Journal of Applied Psychology 88, 879903.
Rank, R.M., Hirschl, A.T., 2001. Rags or riches? Estimating the
probabilities of property and auence across the adult
American life span. Social Science Quarterly 82 (4), 651670.

467

Savery, L.K., 1998. Management and productivity increases.


Journal of Management Development 17 (1), 6874.
Sherman, D., Young, D., Collingwood, H., 2003. Prots You
Can Trust: Spotting and Surviving Accounting Landmines.
Prentice Hall, New York.
Singh, H., Motwanis, J., Komar, A., 2000. A review and analysis
of the state-of-the-art research on productivity measurement.
Industrial Management & Data Systems 100 (5), 234241.
Smith, K.A., 1999. Neural networks for combinatorial optimization: A review of more than a decade of research.
INFORMS Journal on Computing 11 (1), 1534.
Smith, J.B., Barclay, D.W., 1997. The eects of organizational
dierences and trust on the eectiveness of selling partner
relationships. Journal of Marketing 61 (1), 321.
Sparrow, P.R., Hiltrop, J.M., 1997. Redening the eld of
European human resource management: A battle between
national mindsets and forces of business transition? Human
Resource Management 36 (2), 201219.
Sparrow, P.R., Schuler, R.S., Jackson, S.E., 1994. Convergence
and divergence: Human resource practices and policies for
competitive advantage worldwide. International Journal of
Human Resource Management 5 (2), 267299.
Stavrou, E., Brewster, C., 2005. The congurational approach to
linking strategic human resource management bundles with
business performance: Myth or reality? Management Review
16 (2), 186201.
Truss, C., 2001. Complexities and controversies in linking HRM
with organizational outcomes. Journal of Management Studies 38 (8), 11221149.
Ultsch, A., Siemon, H.P., 1990. Kohonens self organizing
feature maps for explanatory data analysis. In: Proceedings
of INNC90, International Neural Networks Conference.
Kluwer, Dordrecht, pp. 305308.
Veiga, F.J., Lubatkin, M., Calori, R., Very, P., Tung, Y.A., 2000.
Using neural network analysis to uncover the trace eects of
national culture. Journal of International Business Studies 31
(2), 223245.
Vesanto, J., 1999. SOM-based data visualization methods.
Intelligence Data Analysis 3 (2), 111126.
Welbourne, T.M., Andrews, A.O., 1996. Predicting performance
of initial public oerings: Should human resource management be in the equation?. Academy of Management Journal
39 (4) 8911101.
Wood, S., 1999. Human resource management and performance.
International Journal of Management Reviews 1 (4), 367413.
Wright, P.M., Dunford, B.B., Snell, S.A., 2001. Human resources
and the resource based view of the rm. Journal of Management 27 (6), 701722.
Zeithaml, V., Berry, L.L., Parasuraman, A., 1996. The behavioral
consequences of service quality. Journal of Marketing 60 (2),
3147.

You might also like