Journal of Operations Management: Lilian M. de Menezes, Stephen Wood, Garry Gelade

Journal of Operations Management 28 (2010) 455–471
Contents lists available at ScienceDirect
Journal of Operations Management

journal homepage: www.elsevier.com/locate/jom
The integration of human resource and operation management practices and its
link with performance: A longitudinal latent class study
Lilian M. de Menezes a,*, Stephen Wood b,1, Garry Gelade c,2
a
Cass Business School, City University London, London, UK
b
School of Management, University of Leicester, UK
c
Business Analytic Ltd., London, UK
A R T I C L E I N F O A B S T R A C T
Article history: This paper reviews the literature on the association between lean production and performance. From this,
Received 17 June 2008 propositions on the integration and evolution of operation and human resource management practices
Received in revised form 20 January 2010 associated with the lean production concept are developed. Using 24 years of data on the use of seven
Accepted 25 January 2010
core OM and HRM practices in British manufacturing firms, the potential link between integration in the
Available online 2 February 2010
use of practices and productivity is tested. In each year, three latent clusters are identified via ordered
restricted latent class models; the cluster that consistently makes a more integrated use of practices
Keywords:
outperforms the others. Furthermore, the longitudinal nature of the data permits modeling the growth
Operation and human resource
management practices
curves of each practice in the sample, recognizing any similarity in growth and investigating whether or
Lean production not an early integration in adoption of practices is associated with higher final productivity. The results
Performance show that pioneers are more productive, thus suggesting that the head start in integrating core OM and
Latent classes, latent growth and multi-level HRM practices associated with the lean production concept has paid off.
models ß 2010 Elsevier B.V. All rights reserved.
1. Introduction use of operation (OM) and human resource management (HRM)

practices.
Markets are increasingly more competitive and global, so quality The quality management literature, which can be traced back to
standards and product or service efficiency is crucial to the survival 1949 when the Union of Japanese Scientists and Engineers formed
of any organization. In this context, the lean production concept has a committee to improve Japanese productivity (Powell, 1995), also
been increasingly advocated beyond manufacturing (e.g. in services: offers competing definitions for total quality management (TQM).
Abdi et al., 2006; Abernathy et al., 2000; Swank, 2003; in the public Some authors argue that it is a managerial philosophy (e.g.
sector and healthcare: Kollberg et al., 2007). The spread and interest Martinez-Lorente et al., 1998), whose principles imply an
that it has attracted are possibly due to claims of cost and waste integration of operations and people management (Wickens,
reduction, superior performance and competitive advantage that 1987). Yet Deming (2000) focused on an organizational system
have been consistently made by academics and practitioners alike that promotes cooperation and learning for facilitating process
(e.g. Chandler and McEvoy, 2000; Dennis, 2002; Hart and Schle- management, resulting in the constant improvement of processes,
singer, 1991; MacDuffie, 1995; Powell, 1995; Womack and Jones, products and services, as well as employee fulfillment and
1994; Womack et al., 1990; Wood et al., 2004; Youndt et al., 1996). customer satisfaction. This system, then, becomes critical to the
Despite controversies about the precise meaning of lean production survival of organizations (Anderson et al., 1994). Juran (1993)
(Li et al., 2005; Shah and Ward, 2007), it has commonly been taken to emphasized not only team and project work, participation and
involve techniques concerned with production, work organization, recognition, but also highlighted the need to achieve the best
quality management, logistics, supply chain, customer satisfaction, return on investment. According to him, the primary responsibility
efficient delivery and continuous improvement methods. In other of management is to minimize the total cost of quality and to
words, the adoption of lean production implies integration in the decide when quality management efforts should be stopped.
Deming also argued that the improvement of the work process
reduces cost because of less re-work, fewer mistakes or delays and
less waste in manpower and materials, but (in contrast to Juran)
* Corresponding author. Tel.: +44 2070408359; fax: +44 2070408328. prescribed constant improvements.
E-mail address: [email protected] (L.M. de Menezes).
1
In this context, important questions for managers and scholars
Institute of Work Psychology, University of Sheffield, Sheffield, UK.
2
Garry Gelade was a research fellow at Cass Business School, City University are: How do different managerial practices impact on perfor-
London, 2007–2009. mance? Which practices have greater impact on performance
0272-6963/$ – see front matter ß 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jom.2010.01.002
456 L.M. de Menezes et al. / Journal of Operations Management 28 (2010) 455–471
outcomes? Does the integration of operations management and its elements and performance. If practices co-exist, the longitudi-
human resource management practices pay off? Are expectations nal nature of our data also permits an assessment of the overall
of long-lasting gains fulfilled? For several decades the link with evolution of practices and the association with performance. Our
performance has mainly been investigated by addressing the first longitudinal study thus investigates whether: (1) the integrated
two questions. However, not only do results and measures vary use of core OM and HRM lean practices is associated with higher
within studies but the focus has generally either been on individual performance; (2) there are performance gains from being an early
practices or sets of practices that are presumed to result in superior adopter of such an integrated approach.
performance. For example, a common view is that HRM practices – In the following sections, we review the empirical evidence on
for instance those that empower or involve the employee – support the link with performance as well as the literature on the evolution
specific OM practices; jointly they have a higher, possibly of management practices, which set the background to the
nonlinear, effect on performance (Kufidu and Vouzas, 1998; propositions that are tested. We then describe the data, the
Lawler et al., 1995; MacDuffie, 1995). Longitudinal studies remain methodology that we use and report our results. Finally, we assess
rare and tend to rely either on subjective assessments of practice our contribution and conclude.
implementation (Easton and Jarrell, 1998; Fullerton et al., 2003) or
on relatively short periods (e.g. Callen et al., 2000; Capelli and 2. Background
Neumark, 2001).
In this paper, we concentrate on the integration of HRM and OM Empirical studies of the link between a lean production
practices and its potential impact on performance, by reviewing philosophy and performance have concentrated on assessing
the empirical literature and developing a set of propositions, which whether one or more of its elements are key, or whether
we then test. In our analysis, we assume that there is a philosophy combinations of these lead to higher performance. In general,
or culture underlying the integration of OM and HRM practices, cross-sections were analyzed; definitions of constructs as well as
whose principles are not only concerned with eliminating waste sample sizes vary considerably. Consequently, it is not only
and adding value for customers but also entail the pursuit of difficult to compare studies, but, as observed by Shah and Ward
continuous improvements. The latter cannot be achieved without (2003: 130), analyses of the performance effects that focus on
worker participation, such as quality at source inspections that multiple components of lean production remain rare.
imply responsibility for checking one’s own work. Learning and In the HRM literature, diverse sets of practices have been
cooperation thus facilitates process management, and integration assumed to be part of a high performance work system (Becker and
should continue to pay over time; those organizations that adopt Gerhart, 1996; Huselid, 1995) and as such are linked with
this managerial philosophy should outperform those that do not. performance. Fundamental practices like empowerment or work
We use secondary data of 448 British firms (Birdi et al., 2008) on enrichment that we would associate with the high involvement
seven management practices that are the most extensively concept (Bailey, 1993) are often omitted in empirical analyses.
employed in British manufacturing (Wood et al., 2004), and Results on the link with performance also reflect the diversity in
company records for the period 1980–2003. Of the seven measurement constructs and dependent variables (Wall and
management practices, the OM practices are total quality Wood, 2005). The empirical evidence ranges from positive
management, just-in-time procedures (JIT), integrated comput- association with all practices (Arthur, 1994), to lack of association
er-based technology (ICT, also known as advanced manufacturing and even a negative correlation (Capelli and Neumark’s 2001 panel
technology), and supply-chain partnering. The first three are study showed no association, but one practice was negatively
central to the lean manufacturing system and the last adds the related to both productivity and its change).
customer–supplier relationships that are necessary for the delivery Overall, longitudinal studies are rare and the potential impact of
of desired standards at a minimum cost. HRM practices included integrating practices has been generally neglected, though it is
are learning culture (or extensive training), empowerment and often argued that practices should be adopted in an integrated
teamwork. The first two have consistently been associated with fashion. We now review the empirical evidence from studies on
quality management, as shown by the criteria of the 1992 Baldrige quality management and lean production, which justify our
Award (George, 1992) and can be easily traced back to two of assessment of the current state of the literature and the need
Deming’s 14 points (Walton, 1986). Empowerment has been for studies like ours.
described as critical to successful JIT initiation and implementation
(Koufteros and Vonderembse, 1998), whereas teamwork is core 2.1. The empirical evidence on the link between lean practices and
both to high involvement (Bailey, 1993; de Menezes and Wood, performance
2006) and quality managements (Wickens, 1987). Together, we
would expect that these HRM practices will allow employees to Initial reviews of the evidence reported a lack of association
enhance their knowledge of the company, work more flexibly, and between practices and performance. Powell (1995) offered several
take more responsibility in quality management and decision- anecdotal examples where the costs of implementing a TQM
making. strategy outweighed its benefits. Reed et al. (1996) concluded that
By focusing on the philosophy that underlies the co-existence any observed benefits were simply contingent on matching
(correlation in the use) of different types of practices, we also environmental conditions to firm orientation. Samson and
address the measurement problem that has been identified in the Terziovski (1999) argued that the lack of evidence was due to
academic literature (e.g. Shah and Ward, 2007). This is done not in small samples and measures that did not reflect the various
terms of what should be included in the measure, but in terms of dimensions of quality management. They examined the relation-
how it should be constructed. If the lean production concept ship between TQM and quality, operational and business perfor-
implies a managerial philosophy, culture or way of thinking, as mance in Australian and New Zealand manufacturing (n = 1200),
advocated in the literature (Womack and Jones, 1994), the co- using multi-item measures that were thoroughly constructed and
existence of OM and HRM practices that are traditionally validated. A positive link with performance was then observed, but
associated with this concept should reflect this philosophy. not all categories of TQM were strong predictors of performance. It
Moreover, if this philosophy is geared towards improving appeared that gains from quality management might be due to the
performance, the direct link between the philosophy and perfor- interaction of practices, thus leading to subsequent research where
mance should be investigated rather than the association between synergistic effects were found between distinct elements of lean
L.M. de Menezes et al. / Journal of Operations Management 28 (2010) 455–471 457
(JIT, TQM, HRM) on multi-item measures of manufacturing and To sum up, the literature supports synergies between the
employee performance (Challis et al., 2002). different practices and we would expect OM and HRM practices to
Indeed, the search for promising combinations of practices has coexist, at least in the long run. It is reasonable, then, to expect
been a dominant subject of empirical research (e.g. Bou and significant correlation between the different elements of lean
Beltran, 2005; Cua et al., 2001; Dow et al., 1999; Flynn et al., 1995; production, and we propose:
MacDuffie, 1995; Narasimhan et al., 2004; Shah and Ward, 2003).
Flynn et al. (1995) claimed that TQM and JIT mutually support each Proposition 1. OM and HRM practices that are associated with the
other: TQM reduces process variance, which can simplify the lean production concept are integrated.
manufacturing process and closely match production and custom-
er demand, while JIT affects quality management by reducing lot Yet in the literature addressing the potential link with
sizes and thus decreasing potential waste and re-work. Besides the performance, the analysis of the correlation structure has been
combination of TQM and JIT, Flynn et al. argued that other practices limited to separate elements of lean. Potential synergies between
(a combination of OM and HRM that they named infrastructure) two or three elements tend to be analyzed, thus implying that
should interact with either of them. Their sample of US elements may be perceived as being independent. As an example,
manufacturing plants (n = 42) within three industries (machinery, we refer to Bou and Beltran (2005: 71), who claimed that ‘‘the
transportation components and electronics) showed strong influence of TQM on organizational results will be greater when its
synergistic effects on quality, but these were especially from implementation is supported by a high-commitment strategy’’,
combinations that also included HRM practices. The potential gain and therefore HRM and TQM should have a synergistic effect on
from the interactions with HRM practices was also observed by performance. They tested this hypothesis via a structural equation
MacDuffie (1995) in a seminal study (n = 62) that was part of a model, where TQM and high commitment management (HRM
wider investigation on the human resource side of lean production, practices) are two independent latent variables that positively
using data from the MIT Future of the Auto Industry project. interact on financial performance (Bou and Beltran, 2005: 81,
According to two performance criteria – productivity (hours taken Fig. 2). They quoted Dwyer (2002: 536) by stating that ‘‘a model
to build a vehicle adjusted by relevant factors, e.g. size) and quality integrating quality and people [practices] could have an impact on
(defects/vehicle) – MacDuffie’s results showed significant syner- the bottom line of the organization’’, but did not test for this
gies between OM and HRM systems. integration by assessing whether their two latent variables were
Positive interactions between supply-chain and total quality correlated and loaded on a single dimension. Moreover, by
managements have also attracted considerable interest (e.g. Carter assuming that the correlation between TQM and the HRM-
and Narasimhan, 1996; Carter et al., 2000; Mehra and Inman, centered commitment variable is negligible, they contradicted
2004). The predominant view is that TQM expands supply-chain their observation that the correlation could be significant, as an
management from cooperative relationships aimed at minimizing ‘‘analysis of the foundations of high-commitment management
costs or trade-offs, to a focus on achieving cumulative competitive shows clear parallelism with the quality climate proposed by TQM
capabilities that target the production of the highest value product experts, since both approaches [i.e. TQM and high-commitment
for customers. A minimum integration of OM practices is required management] emphasize employee participation, commitment
to achieve this effectiveness, and thus one would expect a and trust within the organization’’ (Bou and Beltran, 2005: 75). In
significant correlation between the uses of TQM, JIT and supply- short, despite acknowledging their expectation of integration
chain practices. Flynn and Flynn (2005) tested for synergies between OM and HRM, the authors followed the general trend in
between the perceptions of how extensively TQM and supply- the literature and limited the investigation to synergistic effects.
chain management were employed on a sample of manufacturing Proposition 1 remains to be tested.
plants from five industrialized countries (n = 164). Their TQM
measure included JIT delivery and HRM practices and is thus 2.1.1. The potential link between the integration of lean practices and
broader than most studies. Overall, TQM supported supply-chain performance
management, though not all combinations of practices were The first stage in examining a link between the integrated use of
positively associated with all measures of performance (quality, practices and performance is an assessment of integration. Few
volume flexibility, manufacturing cycle time, on-time delivery). authors have actually modeled the correlation in the adoption of
With a few exceptions (e.g. Cua et al., 2001; Challis et al., 2002), different types of practices when developing their construct.
most studies have concentrated on just two elements of lean, Among those that did, not all assumed that underlying the
though in some cases HRM practices might have been included correlation between practices (or elements) was a management
within TQM. Shah and Ward’s (2003) study is one of the most philosophy. For example, Callen et al. (2003) investigated the risk–
comprehensive, since it focused on multiple elements in a large profitability trade-off of JIT manufacturing (lean) on a sample of
sample of US manufacturing plants (n = 1757) from several 100 plants. Seventeen practices were considered, of which only
industries. Plants were classified into four bundles of related two can be broadly classified as HRM (quality circles and flexibility
processes and procedures: JIT, TQM, HRM and TPM (Total of worker’s skills). Principal component analysis was used to
Preventive Maintenance). After controlling for industry, union develop an index of JIT manufacturing, which is therefore a
presence, size and age of plant, the adoption of these bundles was formative rather than a reflective construct (latent variable). By
positively associated with operational performance, measured as choosing this statistical method, as described by Diamantopoulos
the five-year change in six items (scrap and re-work costs, and Siguaw (2006), the authors assumed that the use of practices
manufacturing cycle time, first pass yield, labor productivity, unit causes the construct being measured (JIT manufacturing) and not
manufacturing cost and customer lead time). Shah and Ward then that the coexistence of practices reflects an underlying managerial
concluded that ‘‘a separate and identifiable incremental effect can approach. According to each plant’s score, the sample was then
be attributed to the four major lean practices areas’’ (2003: 145– split: JIT manufacturing (lean), non-JIT and the first group of plants
146), and that these findings provide ‘‘unambiguous evidence that were found to be more profitable, even after adjusting for the
the synergistic effects of all lean practices are associated with operational risks relative to non-JIT manufacturing. Had similar
better manufacturing performance’’. Thus, all elements should results been found based on a construct derived from factor
coexist, though they did not formally test for the four-way analysis, we would have evidence of a link between integration and
interaction (synergy) nor modeled a secondary factor (integration). performance.
The integration of TQM and HRM practices and its association

with performance have been explicitly investigated by de Menezes
and Wood (2006) in a representative sample of British workplaces
(n = 2191). According to their results, TQM and HRM practices
loaded on a single discrete factor, but JIT was a separate element.
Their integrated HRM–TQM measure was associated with subjec-
tive measures of workplace performance (labor productivity and
its change).
To sum up, the above studies suggest that the integration
between OM and HRM practices as prescribed by the lean
production concept may be associated with performance and
we propose:
Proposition 2. The integration of OM and HRM practices associated

with the lean production concept is linked to performance.
Fig. 1. Akaike criterion–latent class models.
Most studies did not set out to investigate this link and only one
study was longitudinal. We need evidence, especially from
Fullerton et al. (2003) examined the link between JIT (lean) longitudinal studies because they enable the investigation of the
manufacturing systems and financial performance (profitability: associations over time, of changes in the dependent variable
return on assets, return on sales and cash flow margins) in 253 US (performance) as well as the independent variables (e.g. practice
manufacturing firms. They argued that the conflicting empirical use, integrated management approach or lean production) and any
evidence (see Kinney and Wempe, 2002, who extended the work of effect of these changes in the relationships.
Balakrishnan et al. (1996), but in contrast to them found a negative We now consider two longitudinal studies, namely Patterson
link) suggested that the degree of implementation affects the link et al. (2004) and Birdi et al. (2008), because they were based on the
and thus any measure should consider the extent of implementa- Institute of Work Psychology (University of Sheffield) Modern
tion. Hence, based on the extent to which 10 practices were Management Practices Survey Series, from which we draw our
implemented, they found three factors: JIT manufacturing, which data. Patterson et al. (2004) examined the individual and
underlies the adoption of OM and HRM practices, JIT quality and JIT interactive effects of three OM practices (TQM, JIT and advanced
unique, both of which have underlying pairs of OM practices. manufacturing technology—AMT) and two HRM practices (job
Cross-sectional and longitudinal links with financial performance enrichment and skill enhancement) on labor productivity and
were examined, with the respondents’ self-assessments as JIT profit. They found neither support for synergies nor an association
adopters being the basis of the longitudinal investigation. Only JIT between JIT or TQM and performance; only the HRM practices
manufacturing, which is the factor that integrates HRM and OM, were associated with increases in both productivity and profits.
paid off and its level of implementation and maintenance led to Birdi et al. (2008) extended the analysis to 22 years and seven
sustainable rewards. practices: empowerment, extensive training (learning culture) and
Kaynak (2003) developed a structural equation model of teamwork, which are central to strategic HRM (Becker and Huselid,
potential effects of HRM and TQM on perceived financial, market 2006); TQM, JIT and AMT, which are the components of integrated
and operating performances relative to the competition, which manufacturing (Dean and Snell, 1991; Snell and Dean, 1992); and
was tested on a sample of US manufacturing and service firms supply-chain partnership that complements the others. They
(n = 42). In contrast to most studies, the model (Kaynak, 2003: 414, examined the association with productivity, but the focus
Fig. 1) shows some integration between OM and HRM–for instance, remained on separate practices; only empowerment and extensive
management leadership is a latent variable that loads on HRM and training were positively associated with productivity, representing
quality management practices. Results supported all paths in the gains of at least 6%. Their conclusions on the link between
model, and the author concluded that a range of TQM practices had individual practices corroborated the earlier study (Patterson et al.,
positive effects on firm performance and thus ‘‘practitioners as 2004): OM practices on their own and the combination of the four
well as researchers cannot simply pick and choose a few were not linked to performance. Yet there were synergies:
techniques to implement’’ (Kaynak, 2003: 427). teamwork interacted with most practices; four of the six two-
Nahm et al. (2003, 2004) examined the relationships between way interactions between OM practices were positively linked to
organizational structure (number of layers in the hierarchy, level productivity, and supply-chain partnering appeared to enhance
of horizontal integration, locus of decision-making, nature of the effect of all other OM practices. Moreover, they observed a high
formalization and level of communication), culture, composite proportion of positive two-way interactions between HRM and OM
measures of manufacturing practices (re-engineering setups, practices, which they concluded to be a support for the lean
cellular manufacturing, quality initiatives, preventive mainte- production concept ‘‘that the effectiveness of operational practices
nance and pull production) and performance (sales, growth, depends on human resource ones’’ (Birdi et al., 2008: 493).
return on investment, market share gain and overall competition). We will revisit this second dataset and further explore its
In their structural equation models, OM practices load on to a longitudinal nature. In contrast to previous work, we will focus on
latent variable that is shown to mediate the link between the integration and the association in the evolution of practices,
performance and both organizational structure (Nahm et al., based on which we may identify different stages of adoption and
2003) and organizational culture (Nahm et al., 2004). Overall, they any link with performance, as we discuss below.
demonstrated that managerial beliefs, locus of decision-making
and level of communication are associated with what they call 2.2. The evolution of lean management practices and performance
time-based manufacturing practices (OM latent variable), directly
linked to performance. Although these elements may be To understand the evolution of OM and HRM practices, we refer
expressed by HRM practices, in their framework the authors to Powell (1995: 20–21) and his use of the diffusion of innovation
have not explicitly examined the association between OM and literature (Rogers, 1983). Diffusion of innovation refers to the spread
HRM practices. of concepts, technical information and actual practices within a
social system. The spread denotes a flow from a source to an adopter, others may suffer great resistance; therefore, investment can be
typically by communication or influence (for example, the adoption perceived as too risky. Innovators would then be those that have
of lean practices by Japanese subsidiaries in the UK). Diffusion low transaction costs, are willing to take the risk, or have the
trajectories as a function of time (growth curves) are such that at the capacity to invest in continuous learning. Early adopters would
beginning adoption takes longer and growth is barely observed, then then benefit from the sustainable learning and continuous
there is a period of growth, and finally there is a plateau. improvement that are commonly associated with the lean
This pattern is observed and modeled in many areas, as production concept. Hence we propose:
described in Wejnert’s (2002) review of diffusion processes. In
management, the spreads of specific practices have rarely been Proposition 4. There is an association between stages of evolution
investigated. Bryson et al. (2007) and Freitas (2008) use British and performance.
data from the Workplace Employment Relations Surveys and focus The longitudinal nature of the data that we analyze enables us
on workplace characteristics that predict the diffusion of HRM to test our four propositions and thus explore whether the
practices. They argue that practices are governed by considerations integration of core lean HRM and OM practices in British
of costs and benefits that are similar to those involved in any manufacturing may be linked to sustainable productivity.
innovation and thus can be examined by diffusion models. The first
study measured the diffusion or intensity of HRM through an
additive score, while the second focused on specific practices and 3. The empirical study
their implementation. Neither modeled the growth curve nor
focused on performance, for both aimed to identify factors that 3.1. The data
may predict diffusion. They found that industrial sector, size and
age of the establishment, set-up date and number of organizational We use company-level data from the Institute of Work
affiliations are associated with diffusion. Such results are consis- Psychology (University of Sheffield) Modern Management Prac-
tent with the general view that innovations propagate more tices Survey Series. The dataset, as described by Birdi et al. (2008:
rapidly among those that share similar visions, values and 479–80), was constructed by matching the evidence from three
approaches. Consequently, firms can be classified based on their surveys on the use of management practices to yearly audited
stages of adopting a managerial philosophy and we propose: records of company data that include a measure of performance.
The surveys established the use of practices as well as when they
Proposition 3. There is a pattern in the evolution of lean practices, so were implemented, and were based on interviews with either the
firms can be classified according to their adoption stages. company’s managing director or a senior representative responsi-
ble for HRM or production.
Few studies have examined the link between the stages of During these interviews, a definition of working practices was
adopting a managerial philosophy and performance. Among this given, which was formulated from literature reviews, focus groups
literature, Easton and Jarrell (1998) developed an event history and pilot studies. Respondents were asked whether and how
study on the impact of TQM between 1981 and 1991. Implemen- extensively the practice was used across the company and, if so,
tation was judged according to the Baldrige Award criteria, in the year it was introduced. The descriptions of the practices that
which case an advanced implementation would combine OM and were common to all surveys in the series, as relayed to the
HRM practices (George, 1992). Each firm’s performance (account- interviewees, are listed in Table 1 of Birdi et al. (2008: 480) and are:
ing and stock returns) was compared with a control benchmark
that captured what performance would have been without TQM. Empowerment: Passing considerable responsibility for opera-
Independently of the performance measure, a positive link with tional management to individuals or teams (rather than keeping
TQM was established and the observed improvement in perfor- all decision-making at the managerial level).
mance was larger three to five years after TQM implementation, Integrated computer-based technology: Linking together com-
which led the authors to support the view of a longer term impact. puterized equipment to enable enhanced integration (such as
Yet expectations of sustainable effects might have been too CADCAM, computer-integrated manufacturing and flexible
optimistic. For example, Hendriks and Singhal (2001) found no manufacturing systems).
significant difference in performance between earlier and later Just-in-time production: Making products or providing services
‘‘effective’’ implementers (award winners) of TQM, and argued that in direct response to internal or external customer demands
the positive effects of TQM could be diluted by other factors within (rather than building in advance to maintain stock levels).
the organization. Moreover, Fullerton et al. (2003: 400) observed Learning culture, which Birdi et al. (2008) relabeled ‘‘extensive
that greater degrees of implementation of JIT quality practices training’’: Providing a range of development opportunities for all
decreased profitability, but JIT manufacturing that integrated OM employees (rather than training people occasionally to meet
and HRM showed sustainable profits. All in all, we lack coherent specific needs).
evidence on any link between the evolution of management Supply-chain partnering: Developing strategic alliances and
practices and firm performance, possibly because of sampling long-term relationships with suppliers and customers (rather
variations, different constructs and measures, but also due to than negotiating on a short-term basis).
distinct perspectives on long-term effects. Team-based work: Placing employees into teams with their own
We may argue that expectations of long-term effects could responsibilities and giving them the freedom to allocate work
discourage diffusion, because only firms that have higher among team members (rather than having everyone work as
investment capacity would be willing to adopt an innovation individuals).
and wait for medium- or long-range returns. In which case, a Total quality management: Seeking continuous change to
higher performance would be associated with early adoption. improve quality and making all staff responsible for the quality
Furthermore, the diffusion process and its link with performance of their work. (Such practices include Kaizen and continuous
can be highly influenced by learning effects, which may create improvement.)
asymmetries among firms and affect the spread of adoption. In fact,
when practices are viewed in bundles, learning can be subject to In 1996, these practices were identified as the most popular in
considerable variance: some features are easily transferable, the lean management practitioner literature and were estimated to
be implemented in at least half of the UK’s manufacturing Our models, which are described in the next section, aim to fit
companies (Waterson et al., 1999), where their combined used all possible combinations of practice use in the data; if there were
was pioneered in the early 1990s by Japanese transplants and firms no missing observations, we could observe 128 (27) types of
directly competing with or supplying them (Wood, 1996). responses. Hence, we need as much information on practice usage
Together with the HRM practices, they are also included within as possible (see de Menezes, 1999, for the effect of sample sizes)
elements that define total quality management as a philosophy and started with all 684 companies. However, the exclusion of
(Dean and Bowen, 1994). All seven practices appear in most missing values practice-by-practice resulted in inconsistent
accounts of lean production (e.g. Hopp and Spearman, 2004; models over the years, so that we use the data from 448
Parker, 2003; Shah and Ward, 2003; Womack et al., 1990) and have companies that have complete information on all seven practices
been linked to descriptions of the Toyota Production System (e.g. in every year of our study period (1980–2003). A total of 157
Spear and Bowen, 1999). They represent two (supplier-related and companies had complete information on all seven practices as
internally related) of the three underlying constructs identified by well as productivity; the latter is unavailable for the first two years
Shah and Ward (2007) in their comprehensive investigation on in the dataset, so when addressing the association with perfor-
what constitutes lean production. Individually, TQM, JIT and ICT mance we are limited to this subset of 157 companies and the
were core to integrated manufacturing technology centers (Snell period from 1982 to 2003 (22 years). Yet this sample size is
and Dean, 1992, 1994) or the Toyota Production System (Ohno, comparatively larger than other studies of the link with
1988) and were considered in studies of JIT manufacturing (e.g. performance in the manufacturing industry that rely on data
Fullerton et al., 2003; Koufteros and Vonderembse, 1998). The data, selected from far larger populations (e.g. Flynn et al., 1995;
however, does not cover the involvement of customers or TPM Kaynak, 2003; Powell, 1995). Our sample differs from Birdi et al.
(Total Preventive Maintenance–Cua et al., 2001; Shah and Ward, (2008), since they excluded the first two years because these
2003, 2007), which was only included in the 1996 survey where it lacked information on productivity thus leading to a total of 308
was rarely used. Other HRM practices in the survey series firms, but, depending on the practice or combination analyzed,
(appraisal, selection procedures and reward schemes) were absent their sample sizes vary from 122 to 284 firms.
in most years and their link with employee involvement remains
questionable (Beer et al., 1984; de Menezes and Wood, 2006). 3.2. Analysis procedure
Birdi et al. (2008) checked the consistency between the
reported years of introductions and the stated practice use in a Table 1 summarizes the steps, goals and methods that are
sub-sample of over 200 companies for which they had additional involved in testing our four propositions, which are detailed below.
data. Very few inconsistencies in the reported year of introduction
of practices were found between surveys, in which case they opted
3.2.1. Assessing integration of practice-use and its link with
for the most recent information. Their yearly data was then
performance
constructed, by combining the practice usage information from
To test Propositions 1 and 2, we need to examine the association
the three surveys with the information on when the practice had
in practice use in all years studied and then analyze the link with
been introduced. The usage of each practice in each year of the
performance. Our analysis differs from most of the literature, since
study was then recorded as a binary indicator (1 if a firm was using
lean production has often been treated solely as a set of
a practice that year, otherwise 0). Only a small number of firms (1%
management practices (e.g. Birdi et al., 2008; Li et al., 2005; Shah
of the total) had stopped using a practice within the period
and Ward, 2007), which on their own or combined may be
studied. Company-related yearly data (productivity, size of
associated with performance. In which case, if xi is a binary variable
workforce, value of assets) were extracted from the annual
that indicates the use of practice i (i = 1 to n), the link with
audited accounts held at Companies House in London, where
performance can be assessed by the following model:
checks were done and any anomalies were corrected prior to the
creation of the dataset. Per formance ¼ b0 þ b1 x1 þ þ bn xn þ bnþ1 C 1 þ þ bnþk C K
Productivity, the performance variable in the dataset, is our
dependent variable, and was measured as the logarithm of value- þe
added derived profit and loss (operating profit, labor costs and where the Cks (k = 1 to K) are a set of control variables. In addition,
depreciation), adjusted for GDP inflation so that the measure is if two or more practices are assumed to have a synergistic effect on
standardized to 2001 prices. The focus on productivity as an index performance, interactions or product-terms are added to the
of performance is justified on grounds that ‘‘it should more directly model. For example, when considering two practices, the model
reflect the impact of implementing management practices’’ (Birdi becomes:
et al., 2008: 481); it represents a direct link between human capital
and organizational performance (Datta et al., 2005) and is the key Per formance ¼ b0 þ b1 x1 þ b2 x2 þ b3 x1 x2 þ bnþ1 C 1
indicator of workforce performance (Delery and Shaw, 2001). þ þ bnþk C K þ e
Value-added is an important indicator of a company’s output
because it evaluates a company’s efficiency relative to its input . The estimated coefficients of the individual practices and their
(Birdi et al., 2008; Kato and Morishima, 2002). interaction term are then expected to be positive and to
When testing Proposition 2 and thus considering the (yearly) demonstrate that the effect of the combination is greater than
association between the integrated use of practices and produc- that of individual practices.
tivity, the models are controlled for: size of workforce (log number By contrast, when we view lean production as a philosophy, in
of employees), value of assets (log fixed assets adjusted for line with Proposition 1, practices co-exist due to this philosophy
inflation), and time (accounting year). Finally, when examining the (latent variable) and can be measured by the common factor
association between stages of adoption and performance, we add underlying the correlation of individual practices or elements.
initial productivity (value-added) to our controls, and measure Integration implies that a common factor, which can be modeled
time by the number of years from the starting accounting year. as the latent variable that measures the philosophy, can be
These controls are consistent with the Cobb-Douglas production estimated. Moreover, when testing Proposition 2, it is the link
function (Birdi et al., 2008: 481), as well as previous analysis of the between this factor and performance that should be examined, for
diffusion of management practices (Bryson et al., 2007). example:
Table 1
Methodological summary.
Steps in testing propositions Goal Method
Proposition 1 Establish evidence of integration and develop yearly measures

For each year of practice use data:
1. Test for association in practice use Examine evidence for integration of OM and HRM practices Chi-square tests of independence
2. Test for a common factor (latent variable Estimate the underlying construct (common factor) to be analyzed:
that measures the integrated approach) Continuous Latent trait model (Bartholomew
et al., 2002: 175–206)a
Discrete (clusters of firms that similarly integrate practices) Latent class analysis (McCutcheon, 1987)b
Proposition 2 Establish link between integration and performance

Test for a link between extent of Continuous
integration and productivity
Estimate equation 1 if the common factor is continuous; Linear two-level model (Goldstein, 2003)
Discrete Level 1 = firm,

Assess significance and identify the sign of the estimated Level 2 = time
coefficients for the common factor (cluster membership). Dependent variable = productivity
In our case, a significant positive coefficient for the high

usage cluster (or a significant negative coefficient for the
low-usage cluster) indicates a positive link between
combined practice use (integration) and productivity.
Proposition 3 Establish evidence of evolution and classify firms

according to their different stages
1. Estimate the evolution of each practice Identify the evolutionary stages of each practice Latent class growth analysis
(Andruff et al., 2009)c
2. Estimate the underlying evolution of all Identify stages of evolution of an integrated approach Latent class analysis
practices, using results from the previous
stage as input
Proposition 4 Establish a link between the identified stages and performance Linear model (GLM or OLS)
Test for a link between stages of adoption Estimate the linear model and assess the significance Independent variable = evolution (trajectory
and productivity of the adoption stages class)
Dependent variable = productivity
a
If the extent of practice use were available rather than binary indicators of practice use, standard factor analysis would be used.
b
If latent variable is ordinal, estimate an order-restricted model.
c
Time (t = 1–24) would have been the predictor of usage; we were unable to estimate the model and thus divided the 24 years in equal lengths (epochs) that we use as
predictors.
Per formance ¼ b0 þ b1 factor þ bnþ1 C 1 þ þ bnþk C K þ e, management to identify management styles (DeSarbo et al., 2006;
which in a longitudinal framework may be written as: Lewis, 2005; McCutcheon, 1987; Rost and Langeheine, 1997a;
Wood and de Menezes, 1998). The standard latent class model
Per formancet ¼ b0 þ b1t factor t þ bnþ1t C 1t þ þ bnþkt C Kt assumes that binary indicator variables are associated because of
þ et : (1) discrete points (classes) in the latent space. In the absence of these
classes, the indicator variables are no longer associated. Accord-
Testing Proposition 1: Establishing evidence of integration and ingly, we let pij be the probability of the ith practice being used by a
developing yearly measures of the construct. We need to assess firm in category j, hj be the prior probability that a randomly
whether a factor underlies the association between the seven chosen firm is in class j or the size of a class, and x = (x1, x2, . . ., x7)
practices in every year (t) of our data. If practices were used represent the response of a firm to the use of these seven practices.
independently, we would have no evidence of integration and As we only observe x, inference must be based on the joint
Proposition 1 would be rejected. Hence, we first establish that distribution, whose density for the K-class model is given by:
there is significant correlation in practice use, so that we can
X
K p
Y
estimate for each firm the value of this factor in the year. Since our 1xi
f ðxÞ ¼ hj pxi ji ð1 pxi ji Þ (2)
practice data are binary variables, we use chi-square tests. Having j¼1 i¼1
established association, our second step is to determine whether
this association is due to a common continuous factor. We The posterior probability that a firm with response vector x is in
therefore fit a logit-probit latent trait model that is analogous to class j is therefore:
the traditional factor model but is specifically for binary variables, p
Y
as proposed by Bartholomew and Knott (1999: chapter 4) and used hj pxi ji ð1 pi j Þ1xi
by de Menezes and Wood (2006) in their analysis of management i¼1
hð jjxÞ ¼ ; j ¼ 1; . . . K (3)
practices. For brevity, we report that these fits were poor and from f ðxÞ
now on we consider models that assume the latent variable is
categorical (discrete). By maximizing the log-likelihood function, parameters
Latent class models were initially proposed by Lazarsfeld and (h j sandpi j s) are estimated. Multiple sets of random starting values
Henry (1968) in order to identify homogeneous groups of are generally used to avoid local solutions. A firm is allocated to the
individuals based on the correlation structure between several class that it has the highest posterior probability (3) of belonging
binary or categorical variables. Several variants of latent class to. Consequently, in a latent class we have firms that do not
models have been used in sociology to identify types of attitude necessarily make use of exactly the same subset of practices, but
structures from survey responses, in marketing to unveil consumer have the same probability of adopting each practice and as such
segments from demographic and preference variables and in have a similar approach to management.
A log-linear reformulation of this model, as shown by Rost and growth cluster models (latent class growth analysis or LCGA:
Langeheine (1997b), whose parameters are not probabilities and as Andruff et al., 2009; Jung and Wickrama, 2008; Muthén, 2001,
such not restricted to the interval [0,1] enables restrictions to be 2004), using LatentGold 4.0 as described by Vermunt and
applied to the parameters, the inclusion of predictors and more Magidson (2005). These models are increasingly being used in
flexible inference for categorical indicator variables (Heinen, 1996; social sciences and psychology with the aim of capturing
Vermunt, 1999). This reformulation has been adopted by Vermunt information over time about inter-individual differences and
and Magidson (2005) in developing LatentGold 4.0, the software intra-individual change (Nesselroade, 1991; Wang, 2007). In
that we use here, to classify cases into homogeneous groups that contrast to conventional growth models that would assume firms
they call latent class clusters. come from a single population and that a single growth trajectory
If a categorical latent variable is responsible for the association in can adequately approximate the entire population, LCGA enables
practice use, we can fit a latent class model to the data. LatentGold the identification of homogeneous clusters of firms that describe
4.0 is used to estimate the model parameters and allocate firms into similar growth trajectories of practice use.
latent class clusters for each year in the dataset. Standard goodness- We first examine the growth pattern of each practice
of-fit and model selection statistics are then used to judge the separately, by using time (year) as a predictor of usage in a latent
number of latent class clusters that best fits the data. Residual class growth model. The aim is to find the best model (correct
correlations are analyzed in order to detect any dependency number of latent classes) that describes the evolution of each
between pairs of practices (violation of the local independence practice. However, after attempting to model the growth using the
assumption), which, if necessary, can then be modeled. Since the 24 points in time, the results did not converge to unique solutions.
probabilities of a firm adopting each practice are estimated, given Therefore, we divided the period 1980–2003 into five epochs of
that it is in a cluster, we can also identify the characteristics of each equal length as follows: epoch 0 = 1980–1984, epoch 1 = 1985–
cluster in terms of how firms are likely to use practices. 1989, epoch 2 = 1990–1994, epoch 3 = 1995–1999, epoch
As we describe in the next section, we fit up to five latent classes 4 = 2000–2003. If a practice was introduced at any time during
per year and the best solutions imply either two- or three-classes; an epoch, or was already in use at the beginning of an epoch, the
we opted for three. However, the yearly classification varied and practice code for that epoch was set to 1. Otherwise, the practice
we then restricted the models so that the probability of a firm using code for that epoch was 0.
a practice changes monotonically as we move along the sequence We estimate latent class growth models for each individual
of latent class clusters, i.e.: pi j pi j1 . In short, we estimate models practice: the dependent variable is the practice-use indicator (xi = 0
that are known in the statistical literature as order-restricted latent or 1) and the predictor is the epoch coded 0–4, as defined above.
classes (Goodman, 1974; Croon, 1990), which represent an Practice-use indicators are binary variables and thus binary logistic
intermediate stage between the latent trait model where the regression models are estimated. For each practice, we fit up to five
latent space is a continuum, as in factor analysis, and the standard latent classes and choose the model with the lowest Bayesian
latent class model described above. This procedure results in three Information Criterion (BIC) that describes the growth or trajectory
ordered homogeneous clusters of firms per year and a firm is for each practice. A mean growth curve is estimated for each class.
allocated to the cluster that it has the highest estimated probability We then use the identified growth classes as inputs to a second
of belonging to. In each cluster, the probabilities of a firm adopting latent class model, whose classification summarizes the overall
a practice i, where i = 1–7, is the same, and as specified these do not evolution of the seven practices. In short, this is a two-stage process
decrease from cluster 1 to 3. that leads to a single classification of firms (trajectory class) based on
Testing Proposition 2: Establishing the link between integration the diffusion of the seven practices over the 24 years.
and performance. The association between cluster membership and Testing Proposition 4: Establishing the link between stages of
productivity is investigated via a two-level hierarchical regression adoption and performance. Finally, we investigate whether the
procedure, using Mixed Models with SPSS14. The first level is time evolution is associated with firm performance, by examining
(t) and the second is the company (i). Cluster 2, the medium-usage productivity in the last available accounting year as a function of
cluster, is chosen as the baseline, so that if there is a link, we would trajectory class that categorizes firms according to the identified
expect a positive coefficient for Cluster 3 and/or a negative stages of adopting (integrating) practices. A General Linear Model
coefficient for Cluster 1. Our model is the following: is estimated using SPSS14, where the dependent variable is final
performance and the covariates are initial productivity, final
productiyit ¼ b0 þ bs Sizeit þ bA Assetsit þ bt t þ bCl1 Cluster1it company size, final fixed assets and accounting period. The
þ bCl3 Cluster3it þ eit (4) trajectory class is the categorical predictor. We also estimate an
alternative model, as a validity check, where the dependent
where Cluster1it is 1 if firm i is low-usage at time t and 0 otherwise, variable is an average of productivity in the last two years and the
Cluster3it is 1 if firm i is high-usage at time t and 0 otherwise, and eit initial productivity is replaced by its average in the two initial
is the error term. Time (t) varies from 1 to 22 (1982–2003, since years.
1982 is the first year for which performance data is available). The
correlation structure of the first level residuals is identified and 4. Empirical results
estimated from a baseline model that excludes the cluster
membership predictors. Improvement in model fit between our 4.1. The link between integration and performance
model, as described in equation (4), and the baseline model is
assessed by the change in 2 times the log-likelihood, which is a 4.1.1. Establishing the evidence of integration
likelihood ratio statistic that in our case is distributed as a chi- Testing for association in practice use. Chi-square tests of the
square with two degrees of freedom (due to two additional cross-tabulations of the seven practices for the 24 years showed
parameters that our model has, relative to the baseline model). that independency was generally rejected (P-values < 0.05). We
also observed that in 1980, 14% of firms reported using one or more
3.2.2. Assessing evolution and its link with performance OM practice and only 9% reported using one or more HRM practice;
Testing Proposition 3: Establishing the evidence of evolution and by 2002 these figures had grown to 99% and 98%, respectively. In
classifying firms according to the different stages. In order to establish order to confirm that HRM practices were not being adopted
the trajectory (evolution) of each practice, we estimate latent independently of OM practices, we created two binary indicator
variables. For each year, the first variable indicated whether a firm was initially quite small, but increased significantly in the 1990s so
had adopted at least one OM practice, and the second whether a that at the end of the study period it comprises more than 50% of
firm had adopted at least one HRM practice. Cross-tabulation of the firms. The observed frequencies of adopting each practice, as
these two variables suggested that in every year prior to 2002, shown in Fig. 2c, indicate that the typical growth curve that we
firms that had adopted at least one OM practice were more likely to associate with diffusion processes is only completely observable
be using at least one HRM practice than firms that had not adopted for Cluster 3.
any OM practices. With the exception of 2002 and 2003, by which Firms in Cluster 1 are characterized by a consistently low use of
time almost all firms had adopted at least one OM and one HRM practices, which only increases by around the year 2000. This
practice, chi-square tests of independence with continuity cluster comprises about 90% of the firms at the start of the study
correction were strongly significant (P-values 0.002). There is period, but membership declined steadily. By contrast, Cluster 2,
no evidence that firms emphasized a type of practice, or that two which contains firms with an intermediate usage of practices (an
separate clusters representing HRM and OM practices are likely to average use of two practices until 1990), increased in size up to
emerge from these data. In short, there is significant correlation in 2000, and then decreased, such that by 2001 the data could be
practice use over the 24 years. reasonably represented by a high usage and a low usage group.
Estimating the underlying construct: Identifying yearly latent class The observed frequencies of using a practice and the estimated
clusters or homogeneous groups of firms (1980–2003). As the data did probabilities of using a practice vary significantly from cluster to
not fit either a one- or two-factor latent trait model, and failed the cluster (see Fig. 2c). Overall, JIT is used less and is not used in
goodness-of-fit tests described by Bartholomew et al. (2002: 184– Cluster 1; but in more recent years, the estimated likelihood of its
188), there is no evidence of a continuous common factor. We use, according to the latent class models, increased significantly in
therefore tested for a discrete common factor, as outlined in Clusters 2 and 3 (Fig. 2d—probability of usage is approximately
Table 1, and initially considered up to five latent classes–K = 2–5 in 0.70).
(2). However, some models resulted in multiple solutions, Fig. 2d shows the evolution of the estimated probability of using
indicating that either sample sizes were not sufficiently large to each practice if a firm belongs to a cluster over the years. Clusters 2
capture the detailed patterns that stem from large models or that and 3 are closer with regards to the estimated probabilities of using
fewer homogeneous clusters are present in the data. Hence, we OM practices, whereas Clusters 1 and 2 are closer with respect to
estimated two- and three-latent class models (up to 23 parameters their likelihood of making use of empowerment in the early and
in a model) in each year, and used model selection criteria to decide late years. The relative gaps between each cluster in the figures
which best represented the data. suggest that ICT and supply-chain partnering discriminate
The log-likelihood-based model selection criteria (AIC, BIC, between Clusters 1 and 2, whereas empowerment and teamwork
CAIC) suggested a three-cluster (latent class) solution in some discriminate between Clusters 2 and 3. All figures indicate that
years and a two-cluster solution in others. This is illustrated in the firms in Cluster 3 adopt an integrated management approach and
plot of the Akaike Information Criterion (AIC) that is shown in are more likely to mirror those that have been described in the
Fig. 1: the smaller the AIC statistic, the better is the model. It can be literature as early implementers. If there were benefits from
seen in the first decade of the period that two- and three-latent having such an approach, we would expect this group of firms to
cluster solutions fit the data equally well. From 1989, however, outperform the others.
three-cluster solutions fit better and, towards the end of the period, Testing Proposition 2: The link between cluster membership and
the quality of fit of the two- and three-cluster solutions begins to performance (1982–2003). The association between productivity,
converge again. It appears that the pattern of co-existence of our performance measure, and yearly cluster membership is
practices evolves in the period investigated. Yet the ordering of examined by a linear mixed model with time defined as the first
clusters oscillated between years. In order to achieve consistency level, and company as the second level. A baseline growth model
year to year, we use order-restricted models that, as described in (i.e. with time and the controls as fixed effects) suggested a first-
the previous section, constrain the probabilities of using a practice order autoregressive error covariance structure for the effect of
in each class so that they do not decrease. Consequently, the latent time; its fit statistics are: 2log LL = 1775, BIC = 1814, which show
variable that we estimate is ordinal. a significant improvement from a null model that estimates the
Although the three-class solution might be worse than the two- intercept (2log LL = 5195, BIC = 5210). To this baseline model we
class solution in a few years, using a three-class model throughout added the cluster membership predictors as fixed effects, so that
is suitable to most years and simplifies tracking the evolution of we have the model that we defined in equation (3). Results are
practice usage over time. We therefore concentrate on three-class summarized in Table 2. This model is not an improvement in
models, which are specified so that the first cluster makes the quality of fit (2log LL = 1779, BIC = 1818) when compared to our
minimum use of practices and the third the maximum use. All baseline growth model, but indicates that belonging to the high-
estimated models are identifiable (converge to unique solutions) usage cluster (Cluster 3) is positively associated with performance
and satisfy the goodness-of-fit criteria, thus Proposition 1 is (t = 2.35, p = 0.019), while belonging to the low-usage cluster is not
supported by the data. (t = 0.44, p = 0.66). That is, a high-usage firm outperforms
Our results are summarized in Fig. 2a–d, which show, medium-usage firms, though there is no difference between those
respectively, for each year the average number of practices used within low- and medium-usage groups nor is cluster membership
in each cluster, the cluster sizes (i.e. number of firms in each per se a strong predictor of productivity.
cluster), the number of companies per cluster that use each The lack of difference in productivity within the two lower
practice (i.e. the diffusion of each practice per cluster), and the clusters tallies with our initial finding that in more recent years a
(estimated) probability of usage of each practice per cluster. two-class model would best fit the data. In addition, given our
As shown in Fig. 2a, practice use increases significantly in every observations on the extent of practice use in each cluster, we might
cluster during the period studied. Firms in Cluster 3 appear to be conclude that Cluster 3 is the only one making a fully integrated
the innovators, who are then followed by the others. At the start of use of practices. Indeed, this is confirmed when we consider the
the study period, Cluster 3 firms had already adopted an average of estimated probabilities of using each practice if a firm is in Cluster
about four different practices, which increased to just above six at 3 that resulted from the yearly models, as illustrated in Table 3. We
the end of the period, suggesting that most firms in this cluster use observe that not only the estimated size of Cluster 3 significantly
all seven practices in the final year. As Fig. 2b indicates, Cluster 3 increases in the period, but also that by the year 2003 all seven
practices are very likely to be adopted within this group of firms. teamwork it is a four-class model. The latent class growth models are
All in all, those that made a more integrated use of practices are summarized in Table A1 (Appendix), where we report the estimated
also more likely to be high performers. These results may indicate regression coefficients, the quality of fit of the model and the size for
support for Propositions 3 and 4, which we now formally test by each class per practice. These describe the different growth curves
following the steps that were described in Table 1. per practice that were identified; a higher coefficient for epoch
indicates higher growth. For learning culture and supply-chain
4.2. Assessing the link between stages of adoption and performance partnership, the fourth class contained less than 1% of cases, so we
also adopt three-class solutions for these practices. For JIT and
Testing for Proposition 3: Establishing the evidence of evolution and teamwork, four classes were identified whose sizes were greater
classifying firms according to the different stages. We first estimated than 5%; we kept the 4-class solution, though we note that in both
growth models for each practice separately. For TQM, empower- cases one of the growth classes explains a very small proportion of
ment and ICT, the best latent class growth model is a three-class the variance (low r-square). These results indicate that there are
model, and for JIT, learning culture, supply-chain partnership and three clearly identifiable stages in the evolution of these practices.
Fig. 2. (a) Evolution of practice usage. (b) Evolution of cluster membership. (c) Frequency of practice use by cluster and year. (d) Estimated probability of practice-use per Cluster.
Fig. 2. (Continued ).
Table 2
Cluster membership and productivity—estimates of fixed effects.
Baseline model With cluster membership added
Estimate Std error t Estimate Std error t
Intercept 9.35 0.181 51.62** 9.42 0.185 50.98**

Assets 0.1 0.016 6.47** 0.1 0.016 6.34**
Size 0.85 0.023 36.49** 0.84 0.023 36.51**
Time 0.01 0.003 3.22** 0.01 0.003 2.00*
Cluster 1–low usage 0.01 0.027 0.44
Cluster 3–high usage 0.06 0.025 2.35*
Dependent variable: log(Value-added).

*
p .05.
**
p .01.
Identifying firms according to stages of evolution of an integrated other practices. In order to identify the number of trajectory
approach. Our next step was to classify the firms based on the clusters across all practices, models with up to five latent classes
growth models. The indicators for clustering were the seven were estimated. Goodness-of-fit chi-square statistics were non-
growth class membership variables that are categorical variables significant (p = 1.00) for models with more than three latent
ranging from 1 to 4 in the case of JIT and teamwork, or 1 to 3 for all classes. Three clusters are sufficient to describe the whole data and
Table 3
High-usage cluster 3—estimated probabilities of practice use and size.
Practice Year
1980 1985 1990 1995 2000 2003
Empowerment 0.84 0.70 0.71 0.95 0.98 1.00

ICT 0.13 0.26 0.52 0.89 0.95 1.00
JIT 0.42 0.50 0.70 0.94 0.90 0.84
Learning Culture 0.67 0.74 0.62 0.94 0.96 0.99
Supply-chain 0.39 0.42 0.62 0.93 0.96 0.95
Teamwork 0.57 0.78 0.79 0.96 0.95 0.97
TQM 0.46 0.35 0.54 0.96 0.97 0.98
Size of Cluster 3 0.03 0.05 0.14 0.30 0.44 0.60
Table 4
Final three-class solution based on the evolution of practice use.
Management practice Cluster (Trajectory) Late adopter Intermediate adopter Early adopter
Cluster size 46.8% 40.6% 12.6%
Growth class Probabilitya
Empowerment 1 0.29 0.87 0.25

2 0.69 0.06 0.40
3 0.01 0.07 0.36
ICT 1 0.36 0.64 0.39

2 0.61 0.08 0.25
3 0.03 0.28 0.36
JIT 1 0.61 0.10 0.37

2 0.12 0.63 0.25
3 0.23 0.20 0.02
4 0.04 0.07 0.36
Learning culture 1 0.35 0.88 0.12

2 0.64 0.02 0.23
3 0.01 0.10 0.65
Supply-chain 1 0.42 0.79 0.26

2 0.53 0.06 0.22
3 0.05 0.15 0.52
Teamwork 1 0.70 0.38 0.01

2 0.11 0.55 0.26
3 0.17 0.03 0.36
4 0.02 0.04 0.37
TQM 1 0.37 0.82 0.31

2 0.62 0.04 0.33
3 0.02 0.15 0.36
a
Probability of belonging to growth class conditional upon Cluster membership.
companies can be classified in three trajectory clusters that adoption of practices only begins to rise in the middle of the study
describe the evolution across the seven practices. The classification period. From this point, however, adoption is fairly rapid, so that by
error rate for this model is 7%, which is quite good and we have the end of the period studied, the adoption rate in this cluster is
support for Proposition 3. comparable to that in Cluster 3. We will refer to firms in Cluster 1
Results are summarized in Table 4, which shows the size of each as late adopters.
trajectory cluster and the probabilities of belonging to each of the Testing Proposition 4: Establishing the link between stages of
practice growth classes when a firm is within a trajectory cluster. adoption and performance. We investigated a possible effect of the
According to this table, the composition of each trajectory cluster three trajectories on performance, by examining productivity in
in relation to the growth classes of each practice can be inferred. the last available accounting year as a function of the trajectory
For example, firms in the second trajectory cluster are more likely class, controlling for company size, fixed assets in the final
to belong to Growth class 1 for empowerment (probability = 87% accounting year, productivity in the initial (first available)
when Growth class is equal to 1 for empowerment), which accounting year, and the number of years (accounting period)
corresponds to companies that experienced the highest growth for between the initial and final accounting years. A general linear
empowerment since its regression coefficient for epoch in Table A1 model with final productivity as the dependent variable was
is equal to 6.85 and is the highest among the three growth classes. estimated. Initial productivity, workforce size, final assets and
Similarly, it can be seen that higher growth in JIT use (Growth Class accounting period were entered as covariates, and the trajectory
2) is also associated with a higher likelihood of company being in class was the categorical predictor.
the second trajectory cluster (probability = 0.63). The results are shown in Table 5, where we observe a significant
Yet, in line with Proposition 3 and aiming to test Proposition 4, effect of trajectory class on final performance, which is indicated
we focus on the evolution across all practices. By plotting the
observed practice use per trajectory cluster, as shown in Fig. 3, we Table 5
have a clearer picture of the year-on-year evolution that enables Productivity and the overall evolution of practice use.
the identification of the three adoption stages.
Source Mean F Partial
Firms in Cluster 3 are early adopters who tend to make an square Eta-square (%)
integrated use of practices early on, the smallest group (12.6% of
Corrected model 41.3 181.8*** 87.9
the sample, as shown in Table 4). In this cluster, however, the rate
Intercept 70.7 311.3*** 67.5
of further introduction of a practice is relatively slow; by midway Initial productivity 0.1 0.6 0.4
through the study period, adoption of the three HRM practices and Value of assets 2.2 9.8** 6.1
JIT reaches a plateau, although adoption of TQM, ICT and supply- Final size of workforce 45.9 202.0*** 57.4
Time (accounting year) 0.0 0.0 0.0
chain partnering continues to rise. Firms in Cluster 2 start from a
Trajectory class 0.8 3.3* 4.2
lower base, but show rapid change over a relatively short period. Error 0.2
These are firms that we will call intermediate adopters, based on
Note: Dependent variable is log(Value-added) in the final accounting year.
their initial status. In fact, by the end of the study period, firms in
R-Sq = .88.
this cluster have a higher estimated probability of using every *
p .05.
practice than those in the other two clusters. Cluster 1 firms also **
p .01.
***
start from a low base, but are reluctant to get started, and their p < .001.
Fig. 3. Practice evolution trajectories: proportion of firms using each practice.
by its partial eta-squared value (last column). After accounting for successes of quality management and lean production suggest
the control variables, trajectory class membership explains 4.2% of that integrating OM and HRM practices is crucial (e.g. Scherrer-
the remaining variation in final firm performance. Pairwise Rathje et al., 2009; Terziovski et al., 1996). Yet few authors have
comparison tests showed that early adopters significantly empirically addressed the integration of management practices
outperform both late and intermediate adopters (p < 0.05) in and even fewer its link with productivity.
the final accounting year, but there is no significant difference In the present study, we observed that the use of the seven
between late and intermediate adopters. In spite of the similarity practices evolved in a similar fashion and three homogeneous
in performance between late and intermediate adopters, we have groups that described three levels of usage were identified for each
an indication that early adoption is beneficial to the firm. of the 24 years studied. The assumption that practices are only
These results were confirmed by a similar model where the manifest of a latent philosophy and that they would be
initial productivity was measured as the average of the first two independent in the absence of this philosophy was supported by
available years, and similarly the final productivity was the the data. The choice of practices is not driven by expectations on
average of the last two years. Furthermore, hierarchical regres- the performance of specific combinations but by having an
sions, where a baseline model had the covariates as controls and integrated system in place that reflects a managerial philosophy.
the second model had additional dummy variables that indicated Hence, the focus switches from potential efficient bundles of
early and late adopters, also showed that early adopters were practices to this unobserved philosophy or management approach.
positively associated with final (and average final) productivity. The implication for the practitioner is that there are no magical
Given that intermediate adopters experienced rapid growth and by solutions or quick-fixes; the philosophy needs to be in place.
the end of the period studied make a similar or even greater usage Indeed, this may not be surprising when we consider how much
of the practices than early adopters (as shown in Fig. 3), this academics and practitioners have written about Toyota. The
productivity differential supports Proposition 4. Furthermore, it challenge remains to fully embrace the philosophy, be it quality
suggests first-movers’ advantage, which may reflect the continu- management, integrated manufacturing, just-in-time manufactur-
ous improvement achieved through a longer period of integrating ing or lean production. In fact, the need for integration between
practices. OM and HRM has been acknowledged in descriptions of success as
well as failures of quality management strategies. Boudreau et al.
5. Discussion (2003) argued that in spite of a practical need for integration
because they are fundamentally related, the disciplines of HRM
Our review of the literature has shown that many studies have and OM tend to be taught, researched and practiced separately,
addressed the potential gains from combining two or more often drawing on distinct disciplinary foundations. Consequently,
elements of lean production, and that often such combinations more can be learned from addressing the interface, since OM and
entailed both OM and HRM practices. We therefore expect a HRM can explain and moderate the effect of each other, in which
significant correlation in the use of practices and thus a focus on case integration is crucial for competitive advantage.
separate elements may hide the potential outcomes from Given that we had no evidence of a continuous latent variable
integration. Indeed, the evidence from case studies on the that would measure the underlying philosophy, latent class
models enabled the identification of clusters and significantly usage that we found to be the most productive cluster, as
reduced the number of performance equations that were analyzed illustrated in Table 3 in the last year studied (2003), all firms had
when compared to previous studies on synergies, without the need introduced empowerment and teamwork, while 15.6% had not
for either arbitrary criteria or pre-specified bundles. This type of introduced JIT and 4.6% had not introduced supply-chain partner-
model may also capture the non-linear association between ing.
practice and performance that has often been mentioned. The We specifically examined whether there were sustainable
analysis can be extended to ordinal or continuous manifest effects by modeling the stages of adoption. Based on the diffusion
variables, allowing for different levels of measurements and of each practice in the sample, we classified firms according to
restrictions to be added. If necessary, the assumption that the when they adopted practices. The data unveiled three groups:
correlation between practices is only due to the philosophy being early adopters, intermediate adopters and late adopters. At the end
measured can be relaxed. Hence, latent class models are powerful of the period, intermediate adopters had not only caught up with
clustering tools that can be used in future studies on large samples. early adopters but were even making more use of practices, as we
In fact, if the pattern of low versus high implementation that has illustrate in Table A2 where the actual use of practice by the
been observed in the literature persists, such models may be more different groups in different years is shown. In contrast with 1990,
robust than factor analysis when analyzing data that measure the when late and intermediate adopters rarely used the practices
extent of use. This is particularly important when addressing the (median number of practices used were, respectively, 0 and 1), in
link with performance, as recent empirical work has used 2003 more practices were widespread in these groups. Neverthe-
structural equation models, which assume that the data are less, early adopters, a relatively small group (12.28% of the sample),
reasonably well distributed. outperformed the others. This final result supports the initial
We observed that some firms moved between clusters so that in claims that led to the spread of quality management as a
the final years only two clusters might as well have been identified. philosophy, which is that a cohesive management system and
Indeed, this pattern, which is observed after 2000, is consistent continuous improvement can become sources of competitive
with previous research where samples were generally split into advantage. It could be that some of the early adopters in our
two groups: basic–advanced users, early–late implementers. In the sample were initially high performers, but since we controlled for
present study, however, this pattern stems from an integrated initial productivity and considered alternative models, we do have
(correlated) use of practices and its evolution, and not from an an indication of potential first-mover’s advantage and long-lasting
arbitrary cutting-point (e.g. use of 2/3 of practices at least 50% of effects. It seems that benefits are sustainable over the long term
times—Callen et al., 2003) or quality criterion (e.g. award winners). and thus we were unable to observe the productivity gains of the
It is possible that such a pattern might have resulted from focusing intermediate adopters because their use of practices only caught
on the set of the most widely used practices in a single industry up with early adopters near the end of the period studied. This
sector, because similarities between firms could have facilitated should be further investigated within the more encompassing
the spread of practices and thus favor a two-way classification: use, concept of lean production, and preferably beyond manufacturing.
no-use. Although the lower-usage cluster became smaller with In which case, we would like to stress the need for future
time, some firms have lagged behind the majority. In which case, longitudinal studies that include all elements of lean production,
they are either more resistant to innovations or lack the other factors that may influence the learning experience, and the
investment capacity. Given our data, we are unable to identify diffusion of management practices.
factors that could explain this perceived stagnation and we look
forward to further work on different industries with a wider set of 6. Conclusion
practices and firm characteristics to clarify these findings.
Our results indicate that high usage and integration of OM and This is the first longitudinal study on the integration between
HRM practices are linked with productivity. They support previous OM and HRM practices and firm performance in British
work on similar data (Birdi et al., 2008), which concluded that manufacturing. It contrasts with previous work, which tended
expectations based on the lean production concept were supported to investigate potential effects of either individual practices or
by the data: pairwise combinations of the OM and HRM practices specific combinations, and focuses on the association between
tended to be positively associated with productivity, while OM practices and its link with productivity.
practices on their own were not; supply-chain partnering We argued that the integration of practices would reflect a
positively interacted with the other OM practices; and teamwork managerial approach that combines OM and HRM, as advocated in
interacted with most practices. In all, Birdi and his colleagues the quality management and lean production literatures. Conse-
concluded that adopting empowerment and extensive training quently, it is the integration of practices that would have the ability
was the key to productivity. In our view, the implication that to achieve multiple goals and result in superior firm performance.
empowerment alone is best practice is misleading, since a clear We used ordered restricted latent class models to identify
operational focus can be critical to its success. Nonetheless, they homogeneous clusters of firms and established a positive link
also argued that ‘‘there is no practice that does not appear to have with performance. In addition, our analysis of the diffusion of each
some effect when combined with one or more practices’’ (Birdi practice resulted in an alternative classification that describes
et al., 2008: 496). We cannot directly compare the size of effects, three stages of adoption, and we found that earlier integration is
due to sampling variations in the models. There is no evidence, indeed associated with performance. Overall, the benefits appear
according to them, to suggest that firms using any of the seven to be sustainable. An integrated managerial philosophy is
practices without an individual effect on productivity should stop potentially a source of competitive advantage, highlighting the
adopting them. Hence, they seem to have recommended that importance of continuous improvement and learning that is often
practices should be used as a package, in spite of some having allied to the lean production concept.
larger effects. In our results, as indicated by the gaps between the We lack information as to the extent of the use of practices and
curves of each cluster in Fig. 2d, the estimated probabilities of rely on single-item measures, but generally our results are quite
adopting practices in different clusters suggest that empowerment positive and give us insights into the potential evolution of lean
and teamwork may discriminate between medium- and high- production in British manufacturing. Moreover, we rely on just one
usage firms, whereas supply-chain partnership may discriminate measure of performance – productivity – though it may be more
at the bottom end. As matter of fact, when we consider the higher- reliable than financial performance indices, which are more likely
to reflect several economic factors that cannot be controlled by any potential performance effects, but are fit to address the integration
individual company and thus lead to far more complex models. We of management practices in large samples. It gives evidence into
regret not having access to data on quality and lead-time, which in the evolution of core OM and HRM practices in British
our view are key outcomes. manufacturing and shows that integration, early adoption and
An additional limitation is the focus on core practices that are continuous improvement may be linked to organizational perfor-
the most used in British manufacturing and as such by the end of mance.
the study period are widespread in the sample. Yet our results of a
positive link between an integrated use of OM and HRM practices
Acknowledgements
and performance are consistent with more recent findings by
academics and consultants based on a cross-section of 730
For giving access to the data, the authors are very grateful to
manufacturing firms in France, Germany, the UK and the US
Kamal Birdi, Chris Clegg, Malcolm Patterson, Andrew Robinson and
(Bloom et al., 2005). An index of 18 practices (OM and HRM) was
Toby Wall. The authors also thank for their comments on previous
positively associated with productivity, return on capital
versions, Canan Kocabasoglu-Hilmer, Mohan Sodhi, four anony-
employed, sales per employee, sales growth and growth in market
mous referees, and an anonymous associate editor.
share. These findings should encourage future work on more
detailed data that would allow for multi-item measures of a wider
range of practices and implementation. Appendix A
In all, this research introduces a framework and statistical
models that are not traditionally used in empirical analysis of See Tables A1 and A2.
Table A1
Latent class growth models.
Practice Class Estimated regression coefficients R-square Class size
Intercept Epoch
Empowerment 1 0.65 6.85 0.80 50.40%

2 0.43 3.68 0.51 40.60%
3 1.08 1.63 0.40 9.00%
ICT 1 0.46 3.82 0.77 48.50%

2 0.13 5.89 0.64 34.50%
3 0.59 4.88 0.59 17.00%
JIT 1 0.36 5.08 0.19 32.40%

2 0.30 6.48 0.82 30.40%
3 0.21 2.79 0.72 27.80%
4 0.87 2.52 0.06 9.40%
Learning Culture 1 0.52 6.96 0.80 48.80%

2 0.28 2.57 0.54 38.20%
3 0.80 4.24 0.41 13.00%
Supply-chain 1 0.66 3.95 0.78 55.80%

2 0.03 5.56 0.53 29.60%
3 0.68 4.39 0.39 14.60%
Teamwork 1 0.94 3.16 0.71 50.30%

2 0.38 6.43 0.86 28.50%
3 0.35 1.42 0.01 13.80%
4 0.97 2.84 0.10 7.40%
TQM 1 0.60 6.98 0.80 51.10%

2 0.30 3.02 0.57 37.60%
3 0.90 4.46 0.58 11.30%
Table A2
Observed percentage of practice use per type of adopter in three years of the study.
Practice 1980 1990 2003
Late Intermediate Early Late Intermediate Early Late Intermediate Early
Empowerment 0.00 2.72 20.00 2.87 18.48 43.64 77.03 97.28 69.09
ICT 0.48 4.89 7.27 5.74 35.87 45.45 84.21 95.65 87.27
JIT 2.87 5.98 23.64 9.57 35.33 54.55 63.64 88.04 65.45
Learning CULTURE 0.00 2.72 36.36 2.87 22.16 72.73 83.73 98.91 85.45
Supply-chain partnering 2.39 4.89 23.64 9.57 17.39 81.82 81.34 95.11 83.64
Teamwork 1.44 2.72 27.27 4.31 19.57 58.18 81.82 91.85 67.27
TQM 0.48 2.17 16.36 2.39 13.59 69.09 85.17 98.91 78.18
Mean no. of practices 0.08 0.24 1.55 0.37 1.82 3.78 5.55 6.65 6
Median no. of practices 0 0 1 0 1 4 6 7 6
Minimum no. of practices 0 0 0 0 0 1 2 4 4
Maximum no. of practices 2 3 6 4 7 7 7 7 7
Sample size = 448; No. of early adopters = 55; No. of intermediate adopters = 184; No. of late adopters = 209.
References Deming, W., 2000. Out of the Crisis: Quality, Productivity and Competitive Position.
The MIT Press, Cambridge MA.
Abdi, F., Shavarini, S., Hoseini, S., 2006. Glean lean: how to use lean approach in Dennis, P., 2002. Lean Production Simplified. Productivity Press, New York.
services industries? Journal of Services Research 6, 191–206. DeSarbo, W.S., Di Benedetto, C.A., Jedidi, K., Song, M., 2006. identifying sources of
Abernathy, F., Dunlop, J., Hammond, J., Weil, D., 2000. Control your inventory in a heterogeneity for empirically deriving strategic types: a constrained finite-
world of lean retailing. Harvard Business Review, November/December, 169– mixture structural-equation methodology. Management Science 52 (6), 909–
176. 924.
Anderson, J.C., Rungtusanatham, M., Schroeder, R., 1994. A theory of quality Diamantopoulos, A., Siguaw, J.A., 2006. Formative versus reflective indicators in
management underlying the Deming management method. Academy of Man- organizational measure development: a comparison and empirical illustration.
agement Review 19 (3), 472–509. British Journal of Management 17 (4), 263–282.
Andruff, H., Carraro, N., Thompson, A., Gaudreau, P., Louvet, B., 2009. Latent class Dow, D., Samson, D., Ford, S., 1999. Exploding the myth: do all quality management
growth modelling: a tutorial. Tutorials in Quantitative Methods for Psychology practices contribute to superior quality performance? Production and Opera-
5 (1), 11–24. tions Management 8 (1), 1–27.
Arthur, J.B., 1994. Effects of human resource systems on manufacturing perfor- Dwyer, G., 2002. Quality management initiatives and people management strate-
mance and turnover. Academy of Management Journal 7, 670–687. gies. International Journal of Quality and Reliability Management 19 (3), 524–
Bailey, T., 1993. Discretionary Effort and the Organization of Work: Employee 539.
Participation and Work Reform since Hawthorne. Mimeo, Teachers College, de Menezes, L.M., 1999. On fitting latent class models for binary data: the estima-
Columbia University. tion of standard errors. British Journal of Mathematical and Statistical Psychol-
Balakrishnan, R., Linsmeier, T.J., Venkatachalam, M., 1996. Financial benefits from ogy 52 (2), 149–168.
JIT adoption: effects of costumer concentration and cost structure. The Ac- de Menezes, L.M., Wood, S., 2006. The reality of flexible work systems in Britain.
counting Review 71 (2), 183–205. International Journal of Human Resource Management 17 (2), 106–138.
Bartholomew, D.J., Knott, M.K., 1999. Latent Variable Models and Factor Analysis, Easton, G.S., Jarrell, S.L., 1998. The effects of total quality management on
2nd ed. Arnold, London. corporate performance: an empirical investigation. Journal of Business 71
Bartholomew, D.J., Steele, F., Moustaki, I., Galbraith, J.I., 2002. The Analysis and (2), 253–307.
Interpretation of Multivariate Data for Social Scientists. Chapman & Hall/CRC, Flynn, B.B., Flynn, E.J., 2005. Synergies between supply chain management and
Boca Raton. quality management: emerging implications. International Journal of Produc-
Becker, B.E., Huselid, M.A., 2006. Strategic human resources management: where do tion Research 43 (16), 3421–3436.
we go from here? Journal of Management 32 (6), 898–925. Flynn, B., Sakakibara, S., Schroeder, R., 1995. Relationship between JIT and TQM:
Becker, G., Gerhart, B., 1996. The impact of HRM on organizational performance: practices and performances. Academy of Management Journal 38 (5), 1325–
progress and prospects. Academy of Management Journal 39 (4), 779–801. 1360.
Beer, M., Spector, B., Lawrence, R., Mills, D., Walton, R., 1984. Managing Human Freitas, I.M.B., 2008. Sources of differences in the pattern of adoption of organiza-
Assets. The Free Press, New York. tional and managerial innovations from early to late 1990s in the UK. Research
Birdi, K., Clegg, C., Patterson, M., Robinson, A., Stride, C.B., Wall, T.D., Wood, S.J., Policy 37 (1), 131–148.
2008. The impact of human resource and operational management practices on Fullerton, R.R., McWatters, C.S., Fawson, C., 2003. An examination of the relationship
company productivity: a longitudinal study. Personnel Psychology 61, 467– between JIT and financial performance. Journal of Operations Management 21
501. (4), 383–404.
Bloom, N., Dorgan, S., Dowdy, J., Rippin, T., Van Reenen, J., 2005. Management George, S., 1992. The Baldrige Quality System. Wiley, New York.
practices: the impact on company performance. CentrePiece, summer issue, 2– Goldstein, H., 2003. Multilevel Statistical Models, Third Edition. Edward Arnold,
6. London.
Bou, J.C., Beltran, I., 2005. Total quality management, high-commitment human Goodman, L.A., 1974. Simple models for the analysis of association in cross-
resource strategy and firm performance: an empirical study. Total Quality classifications having ordered categories. Journal of the American Statistical
Management 16 (1), 71–86. Association 74 (367), 537–552.
Boudreau, J., Hopp, W., McClain, J.O., Thomas, L.J., 2003. On the interface between Hart, C., Schlesinger, L., 1991. Total quality management and the human resource
operations and human resources management. Manufacturing & Service Opera- professional: applying the Baldrige framework to human resources. Human
tions Management 5 (3), 179–202. Resource Management 30 (4), 433–454.
Bryson, A., Gomez, R., Kretschmer, T., Willman, P., 2007. The diffusion of workplace Heinen, T., 1996. Latent Class and Discrete Latent Trait Models: Similarities and
voice and high-commitment human resource management practices in Britain, Differences. Sage Publications, Thousand Oaks.
1984-1998. Industrial and Corporate Change 16 (3), 395–426. Hendriks, K.B., Singhal, V.R., 2001. Firm characteristics, total quality management
Callen, J.L., Fader, C., Krinsky, I., 2000. Just-in-time: a cross-sectional plant analysis. and financial performance. Journal of Operations Management 19 (3), 269–
International Journal of Production Economics 63 (3), 277–301. 285.
Callen, J.L., Morel, M., Fader, C., 2003. The profitability-risk tradeoff of just-in- Hopp, W.J., Spearman, M.L., 2004. To pull or not to pull: what is the question?
time manufacturing technologies. Managerial and Decision Economics 24, Manufacturing and Service Operations Management 6 (2), 133–148.
393–402. Huselid, M.A., 1995. The impact of human resource management practices on
Capelli, P., Neumark, D., 2001. Do ‘high-performance’ work practices improve turnover, productivity, and corporate financial performance. Academy of Man-
establishment-level outcomes? Industrial Labor Relations Review 54 (4), agement Journal 38 (3), 635–672.
737–775. Jung, T., Wickrama, K.A.S., 2008. An introduction to latent class growth analysis and
Carter, J.R., Narasimhan, R., 1996. Purchasing and supply chain management: future growth mixture modeling. Social and Personality Psychology Compass 2 (1),
directions and trends. International Journal of Purchasing and Materials Man- 302–317.
agement 32 (1), 2–12. Juran, J., 1993. Made in USA: a renaissance in quality. Harvard Business Review,
Carter, P., Carter, J.R., Monczka, R.M., Slaight, T.H., Swan, A.J., 2000. The future of July–August, 42–50.
purchasing and supply: a ten year forecast. Journal of Supply Chain Manage- Kato, T., Morishima, M., 2002. The productivity effects of participatory employment
ment 36 (1), 14–26. practices: evidence from new Japanese panel data. Industrial Relations 41 (4),
Challis, D., Samson, D., Lawson, B., 2002. Integrated manufacturing, employee and 487–520.
business performance: Australian and New Zealand evidence. International Kaynak, H., 2003. The relationship between total quality management practices and
Journal of Production Research 40 (8), 1941–1964. their effects on firm performance. Journal of Operations Management 21 (4),
Chandler, G.N., McEvoy, G.M., 2000. Human resource management, TQM, and firm 405–435.
performance in small and medium enterprises. Entrepreneurship Theory and Kinney, M.R., Wempe, W.F., 2002. Further evidence on the extent and origins of JIT’s
Practice 25 (1), 43–47. profitability effects. The Accounting Review 77 (1), 203–225.
Croon, M.A., 1990. Latent class analysis with ordered latent classes. British Journal Kollberg, B., Dahlgaard, J.J., Brehmer, P., 2007. Measuring lean initiatives in health
of Mathematical and Statistical Psychology 44, 315–331. care services: issues and findings. International Journal of Productivity &
Cua, K.O., McKone, K.E., Schroeder, R.G., 2001. Relationships between implementa- Performance Management 56 (1), 7–24.
tion of TQM, JIT, and TPM and manufacturing performance. Journal of Opera- Koufteros, X.A., Vonderembse, M.A., 1998. The impact of organizational structure on
tions Management 19 (2), 675–694. the level of JIT attainment: towards theory development. International Journal
Datta, D., Guthrie, J., Wright, P., 2005. Human resource management and labor of Production Research 36 (10), 2863–2878.
productivity: does industry matter? Academy of Management Journal 48, 135– Kufidu, S., Vouzas, F., 1998. Human resource aspects of quality management:
145. evidence from MNEs operating in Greece. International Journal of Human
Dean Jr., J.W., Bowen, D.E., 1994. Management theory and total quality: improving Resource Management 9 (5), 818–830.
research and practice through theory development. Academy of Management Lawler, E.E., Mohrman, S.A., Ledford Jr., G.E., 1995. Creating High Performance
Review 19 (3), 392–418. Organizations. Jossey-Bass, San Francisco CA.
Dean, J.W., Snell, S.A., 1991. Integrated manufacturing and job design: moderating Lazarsfeld, P.F., Henry, N.W., 1968. Latent Structure Analysis. Houghton-Mifflin,
effects of organizational inertia. Academy of Management Journal 34 (4), 776– New York.
804. Lewis, M., 2005. Research note: a dynamic programming approach to customer
Delery, J.E., Shaw, J.D., 2001. The strategic management of people in work relationship pricing. Management Science 51 (6), 986–994.
organizations: review, synthesis and extension. In: Rowland, K.M., Ferris, Li, S., Rao, S.S., Ragu-Nathan, T.S., Ragu-Nathan, B., 2005. Development and valida-
G.R. (Eds.), Research in Personnel and Human Resource Management. JAI tion of a measurement instrument for studying supply management practices.
Press, Greenwich CT, pp. 165–197. Journal of Operations Management 23 (6), 618–641.
MacDuffie, J.P., 1995. Human resource bundles and manufacturing performance: Shah, R., Ward, P.T., 2003. Lean manufacturing: context, practice bundles, and
organizational logic and flexible production systems in the world auto industry. performance. Journal of Operations Management 21 (2), 129–150.
Industrial and Labour Relations Review 48 (2), 197–221. Shah, R., Ward, P.T., 2007. Defining and developing measures of lean production.
Martinez-Lorente, A., Dewhurst, F., Dale, B., 1998. Total quality management: Journal of Operations Management 25 (4), 785–805.
origins and evolution of the term. The TQM Magazine 10 (6), 378–386. Snell, S.A., Dean, J.W., 1992. Integrated manufacturing and human resource man-
McCutcheon, A.C., 1987. Latent Class Analysis. Sage Publications, Beverly Hills. agement: a human capital perspective. Academy of Management Journal 35 (3),
Mehra, S., Inman, A.R., 2004. Purchasing management and business competitive- 467–504.
ness in the coming decade. Production Planning and Control 15 (7), 710–718. Snell, S.A., Dean, J.W., 1994. Strategic compensation for integrated manufacturing:
Muthén, B.O., 2001. Second-generation structural equation modeling with a com- the moderating effects of jobs and organizational inertia. Academy of Manage-
bination of categorical and continuous latent variables: new opportunities for ment Journal 37 (5), 1109–1140.
latent class/latent growth modeling. In: Collins, L.M., Sayer, A. (Eds.), New Spear, S., Bowen, H.K., 1999. Decoding the DNA of the Toyota production system.
Methods for the Analysis of Change. American Psychological Association, Harvard Business Review 77 (9), 97–106.
Washington DC, pp. 291–322. Swank, C., 2003. The lean service machine. Harvard Business Review 81 (10), 123–
Muthén, B.O., 2004. Latent variable analysis: growth mixture modelling and related 129.
techniques for longitudinal data. In: Kaplan, D. (Ed.), Handbook of Quantitative Terziovski, M., Sohal, A., Samson, D., 1996. Best practice implementation of total
Methodology for the Social Science. Sage Publications, Newbury Park CA, pp. quality management: multiple cross-case analysis of manufacturing and ser-
345–368. vice organizations. Total Quality Management 7 (5), 459–482.
Nahm, A., Vonderembse, M.A., Koufteros, X., 2003. The impact of organizational Vermunt, J.K., 1999. A general class of non-parametric models for ordinal categori-
structure on time-based manufacturing and performance. Journal of Operations cal data. In: Sobel, M., Becker, M. (Eds.), Sociological Methodology, vol. 29, 187–
Management 21 (3), 281–306. 224.
Nahm, A., Vonderembse, M.A., Koufteros, X., 2004. The impact of organizational Vermunt, J.K., Magidson, J., 2005. Technical Guide for LatentGold 4. 0: Basic and
culture on time-based manufacturing and performance. Decision Sciences 35 Advanced. Statistical Innovations Inc., Belmont MA.
(4), 579–607. Wall, T.D., Wood, S.J., 2005. The romance of HRM and business performance, and the
Narasimhan, R., Swink, M., Kim, S.W., 2004. Disentangling leanness and agility: an case for big science. Human Relations 58 (4), 429–462.
empirical investigation. Journal of Operations Management 24 (5), 421–439. Walton, M., 1986. The Deming Management Method. Pedigree, New York.
Nesselroade, J.R., 1991. Interindividual differences in intraindividual change. In: Wang, M., 2007. Profiling retirees in the retirement transition and adjustment
Collins, L.A., Horn, J.L. (Eds.), Best Methods for the Analysis of Change. American process: examining the longitudinal change patterns of retirees’ psychological
Psychological Association, Washington DC, pp. 92–106. well-being. Journal of Applied Psychology 92 (2), 455–474.
Ohno, T., 1988. Toyota Production System: Beyond Large Scale Production. Produc- Waterson, P., Clegg, C.W., Bolden, R., Pepper, K., Warr, P.B., Wall, T.D., 1999. The use
tivity Press, Cambridge MA. and effectiveness of modern manufacturing practices: a survey of UK industry.
Parker, S., 2003. Longitudinal effects of lean production on employee outcomes and International Journal of Production Research 37 (10), 2271–2292.
the mediating role of work characteristics. Journal of Applied Psychology 88 (4), Wejnert, B., 2002. Integrating models of diffusion of innovations: a conceptual
620–634. framework. Annual Review of Sociology 28 (1), 297–326.
Patterson, M.G., West, M.A., Wall, T.D., 2004. Integrated manufacturing, empower- Wickens, P., 1987. The Road to Nissan: Flexibility, Quality, Teamwork. Macmillan,
ment and company performance. Journal of Organizational Behavior 25 (5), Houndmills.
641–665. Womack, J., Jones, D., 1994. From lean production to the lean enterprise. Harvard
Powell, T.C., 1995. Total quality management as competitive advantage: a review Business Review 72 (1), 93–104.
and empirical study. Strategic Management Journal 16 (1), 15–37. Womack, J., Jones, D., Roos, D., 1990. The Machine that Changed the World. Rawson,
Reed, R., Lemak, D.J., Montgomery, J.C., 1996. Beyond process: TQM content and firm New York.
performance. Academy of Management Review 21 (1), 173–202. Wood, S.J., 1996. How different are human resource practices in Japanese ‘‘trans-
Rogers, E.M., 1983. Diffusion of Innovations, 3rd ed. Free Press, New York. plants’’ in the United Kingdom? Industrial Relations 35 (4), 511–525.
Rost, J., Langeheine, R., 1997a. Applications of Latent Trait and Latent Class Models Wood, S.J., de Menezes, L.M., 1998. High commitment management in the
in the Social Sciences. Waxmann, New York. UK: evidence from the Workplace Industrial Relations Survey, and Employ-
Rost, J., Langeheine, R., 1997b. A guide through latent structure models for cate- ers’ Manpower and Skills Practices Survey. Human Relations 51 (4), 485–
gorical data. In: Rost, J., Langeheine, R. (Eds.), Applications of Latent Trait and 515.
Latent Class Models in the Social Sciences. Waxmann, New York. Wood, S.J., Stride, C.B., Wall, T.D., Clegg, C.W., 2004. Revisiting the use and effec-
Samson, D., Terziovski, M., 1999. The relationship between total quality manage- tiveness of modern management practices. Human Factors and Ergonomics in
ment practices and operational performance. Journal of Operations Manage- Manufacturing 14 (4), 415–432.
ment 17 (5), 393–409. Youndt, M.A., Snell, S.A., Dean, J.W., Lepak, D.P., 1996. Human resource manage-
Scherrer-Rathje, M., Boyle, T.A., Deflorin, P., 2009. Lean, take two! Reflections from ment, manufacturing strategy, and firm performance. Academy of Management
the second attempt at lean implementation Business Horizons 52 (1), 79–88. Journal 39 (4), 836–866.

Journal of Operations Management: Lilian M. de Menezes, Stephen Wood, Garry Gelade

Uploaded by

Copyright:

Available Formats

Journal of Operations Management: Lilian M. de Menezes, Stephen Wood, Garry Gelade

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Journal of Operations Management: Lilian M. de Menezes, Stephen Wood, Garry Gelade

Uploaded by

Copyright:

Available Formats

Journal of Operations Management 28 (2010) 455–471

Contents lists available at ScienceDirect

Journal of Operations Management

1. Introduction use of operation (OM) and human resource management (HRM)

The integration of TQM and HRM practices and its association

Proposition 2. The integration of OM and HRM practices associated

Steps in testing propositions Goal Method

Proposition 1 Establish evidence of integration and develop yearly measures

Proposition 2 Establish link between integration and performance

Discrete Level 1 = ﬁrm,

In our case, a signiﬁcant positive coefﬁcient for the high

Proposition 3 Establish evidence of evolution and classify ﬁrms

Baseline model With cluster membership added

Estimate Std error t Estimate Std error t

Intercept 9.35 0.181 51.62** 9.42 0.185 50.98**

Dependent variable: log(Value-added).

1980 1985 1990 1995 2000 2003

Empowerment 0.84 0.70 0.71 0.95 0.98 1.00

Empowerment 1 0.29 0.87 0.25

ICT 1 0.36 0.64 0.39

JIT 1 0.61 0.10 0.37

Learning culture 1 0.35 0.88 0.12

Supply-chain 1 0.42 0.79 0.26

Teamwork 1 0.70 0.38 0.01

TQM 1 0.37 0.82 0.31

Fig. 3. Practice evolution trajectories: proportion of ﬁrms using each practice.

Practice Class Estimated regression coefﬁcients R-square Class size

Empowerment 1 0.65 6.85 0.80 50.40%

ICT 1 0.46 3.82 0.77 48.50%

JIT 1 0.36 5.08 0.19 32.40%

Learning Culture 1 0.52 6.96 0.80 48.80%

Supply-chain 1 0.66 3.95 0.78 55.80%

Teamwork 1 0.94 3.16 0.71 50.30%

TQM 1 0.60 6.98 0.80 51.10%

Practice 1980 1990 2003

Late Intermediate Early Late Intermediate Early Late Intermediate Early

You might also like

Intercept 9.35 0.181 51.62 9.42 0.185 50.98