Journal of Operations Management 30 (2012) 437453

Contents lists available at SciVerse ScienceDirect

Journal of Operations Management

journal homepage: www.elsevier.com/locate/jom

Six Sigma adoption: Operating performance impacts and contextual drivers of success
Morgan Swink a, , Brian W. Jacobs b,1
a b

Neeley Business School, TCU, PO Box 298530, Fort Worth, TX 76129, United States Broad College of Business, Michigan State University, N370 North Business Complex, East Lansing, MI 48824-1122, United States

a r t i c l e

i n f o

a b s t r a c t
We assess the operational impacts of Six Sigma program adoptions through an event study methodology, comparing nancial data for 200 Six Sigma adopting rms against data for matched rms, which serve as control groups for the analyses. We employ various matching procedures using different combinations of pre-adoption return on assets (ROA), industry, and size as matching criteria. By comparing performance outcomes across a hierarchy of operating metrics, we establish a pattern of Six Sigma adoption effects that provides strong evidence of a positive impact on ROA. Interestingly, these ROA improvements arise mostly from signicant reductions in indirect costs; signicant improvements in direct costs and asset productivity are not evident. We also nd small improvements in sales growth due to Six Sigma adoption. Cross-sectional analyses of the performance results reveal that distinctions in Six Sigma impacts across manufacturing and service rms are negligible. Interestingly, we nd that the performance impact of Six Sigma adoption is negatively correlated to the rms quality system maturity (indicated by prior ISO 9000 certication). Further analyses of manufacturing and service rms reveals that Six Sigma benets are signicantly correlated with intensity in manufacturing, and with nancial performance before adoption in services. We discuss the implications of these ndings for practice and for future research. 2012 Published by Elsevier B.V.

Article history: Received 18 October 2011 Received in revised form 24 February 2012 Accepted 25 May 2012 Available online 4 June 2012 Keywords: Six sigma Process innovation Operating performance Event study

1. Introduction Since its origins in the mid-1980s, the Six Sigma program for process improvement has become widely embraced. One report suggests that many Fortune 500 rms have adopted Six Sigma (Nakhai and Neves, 2009). Early successes in high prole companies such as Motorola, Allied Signal (now Honeywell), and General Electric helped to both popularize and legitimize the approach, and dozens of books have been devoted to the topic. The practitioner literature documents substantial cost savings and other benets from Six Sigma program adoptions (Pande et al., 2000; Harry and Schroeder, 2000). However, some still question whether these benets sufciently exceed the costs of adoption. Corporate-wide adoption of Six Sigma often involves considerable investments in consulting support, training, organizational restructurings, and associated information and reporting systems. For example, over a four year period (19961999)

Corresponding author. Tel.: +1 817 257 7463. E-mail addresses: m.swink@tcu.edu (M. Swink), jacobsb@bus.msu.edu (B.W. Jacobs). 1 Tel.: +1 517 884 6370. 0272-6963/$ see front matter. 2012 Published by Elsevier B.V. http://dx.doi.org/10.1016/j.jom.2012.05.001

General Electric reportedly spent more than $1.6 billion on Six Sigma investments. Researchers report that training costs are typically as much as $50,000 per trained worker (Antony, 2006; Fahmy, 2006). The net operating effects of these types of investments have not been rigorously examined. Most scholarly work to date involves perceptual data from surveys, or nancial studies of a few select companies (Goh et al., 2003; Zu et al., 2008; Gutierrez et al., 2009; Braunscheidel et al., 2011). In fact, some writers have even questioned the validity and originality of Six Sigma, calling it repackaging, a fad, and a PR ploy (Clifford, 2001; Rowlands, 2003). Other questions pertain to the types of benets provided by Six Sigma, and their limitations. A number of researchers discuss the potential for capability gains in one area of performance to be offset by added constraints or losses in another. In particular, Six Sigma potentially creates a trade-off between gains in efciency versus growth. Several important studies suggest that process improvement regimes can stie innovative exploration in favor of exploitation, thus impeding sales growth (Abernathy, 1978; Tushman and OReilly, 1996; Benner and Tushman, 2002, 2003; Naveh and Erez, 2004). Moreover, recent anecdotes from companies like General Electric and 3 M indicate that managers believe Six Sigma practices may severely constrain innovation needed to drive growth (Brady, 2005; Hindo, 2007).

438

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

Limitations might also stem from the context within which Six Sigma is adopted. Like many process improvement programs, Six Sigma originated in manufacturing rms; many of its principles and tenets were developed in a setting of asset-intensive, repeatable processes. The name itself, Six Sigma, refers to limits in measurable variations of outputs that were established in Motorolas manufacturing processes. In addition, researchers maintain that a rm must possess certain resources and make certain commitments in order to make Six Sigma successful (Antony et al., 2008; Schroeder et al., 2008). Hence, Six Sigma methods and tools may be more or less effective in certain technological and operational contexts. In this article, we examine the operating performance impacts of Six Sigma adoptions. The study seeks answers to the following three research questions. First, does Six Sigma adoption consistently produce a signicant net effect on operating performance? Given the widespread adoption and continued popularity of this program, we consider this a very important question. A sizable literature on the efcacy of other process management strategies exists, providing mixed results. However, researchers argue that Six Sigma is different from other process management approaches; it is distinguished by its requisite organizational structures, structured methods, and emphasis on customer-oriented metrics (Linderman et al., 2003; Sinha and Van de Ven, 2005; Schroeder et al., 2008). Given these proposed distinctions, it is important to determine whether or not managers should have reason to expect that Six Sigma will provide benets that exceed alternative programs for improvement. Our second research question addresses the nature of Six Sigmas impacts. What types of benecial impacts are manifested in the operating data of Six Sigma adopters? By examining the components of both prot and growth-oriented nancial outcomes of Six Sigma adopters, we develop insights into the types of impacts provided by the program. These results serve to inform the debate over the roles of process management programs in creating competitive advantages for their adopters; they also point to some interesting propositions for future research. Our third research question is: are Six Sigma impacts related to operating contexts? As Six Sigma adoptions have grown to include a wider scope of businesses, researchers have begun to question the applicability and effectiveness of related tools and techniques in certain contexts. In addition, case studies and anecdotal evidence is suggestive of factors that may be critical to successful implementation. We study differences in Six Sigma success associated with industry (manufacturing or service), labor intensity, R&D intensity, prior operating performance, and quality maturity. Our examination of these factors provides insights into the sources of, and constraints on, process improvements emerging from Six Sigma adoption. We address the foregoing questions through an event study methodology, comparing nancial data for about 200 Six Sigma adopting rms against data for matched rms, which provide control groups for the analyses. We employ various matching procedures using different combinations of pre-adoption operating performance (measured by return on assets (ROA)), industry, and size as matching criteria. By comparing performance outcomes across a hierarchy of operating metrics, we establish a pattern of Six Sigma adoption effects that provides strong evidence of a positive impact on ROA. Interestingly, these ROA improvements arise mostly from signicant reductions in indirect costs. Improvements in direct costs and asset productivities are not evident. We also nd small improvements in sales growth due to Six Sigma adoption. From cross-sectional analyses, we determine that performance improvement due to Six Sigma adoption is not signicantly related to industry (manufacturing or service) or R&D intensity. However, changes in performance are signicantly correlated with the quality maturity of the adopting rms. Interestingly, rms with

greater quality experience (as indicated by ISO 9000 certication) appear to benet less from Six Sigma. For rms in service industries, operating performance before Six Sigma adoption is a signicant determinant of performance changes. However, labor intensity is the most signicant driver of performance benets in manufacturing rms. In the next section, we formulate hypotheses relating Six Sigma adoption to operating performance by drawing upon the literatures on process improvement in general, and Six Sigma in particular. Section 3 describes the sample data and event study method. Section 4 presents the results. Section 5 discusses the ndings and their implications. Section 6 summarizes the conclusions and limitations of the study, and identies opportunities for future research. 2. Theory development and hypotheses Researchers have placed Six Sigma in the realm of operational improvement programs that are oriented toward improvements in quality or variability of process outcomes (Zu et al., 2008). There are several scholarly studies of the impacts of process improvement programs, yet none provide a rigorous examination of Six Sigma adoptions. The existing literature can be classied into three streams addressing the performance impacts of: (1) general process management strategies (Ittner and Larcker, 1997; Schmenner, 1991), (2) Total Quality Management (TQM) implementations (Hendricks and Singhal, 1996, 1997, 2001a,b; Ittner et al., 2001; Powell, 1995; Sila, 2007; York and Miree, 2004; Nair, 2006), and (3) ISO 9000 and other quality certications (Corbett et al., 2005; Martinez-Costa et al., 2009; Westphal et al., 1997; Yeung et al., 2006; Naveh and Erez, 2004; Benner and Tushman, 2002; Benner and Veloso, 2008; Levine and Toffel, 2010). These research streams provide an overall positive, though mixed, set of conclusions regarding the effectiveness of respective process improvement programs. Importantly, however, researchers have argued that Six Sigma is distinguished from these other programs by several characteristics. 2.1. The distinctive characteristics of Six Sigma Researchers describe Six Sigma as a data driven approach to problem solving, as a business process, as a disciplined statistical approach, and as a management strategy (Blakeslee, 1999; Hahn et al., 1999; Harry and Schroeder, 2000; Braunscheidel et al., 2011). While these monikers have been applied to other process improvement strategies as well, proponents and researchers argue that Six Sigma is different than other process improvement programs because it is exclusively a customer-driven and data-dened system (Breyfogle, 2003). Schroeder et al. (2008) suggest that Six Sigma must be different by virtue of the fact that it has been adopted by many rms that had already possessed quite mature quality management processes (e.g., 3M, Ford, Honeywell, American Express). Perhaps more compellingly, Schroeder et al. (2008) and Zu et al. (2008) argue that, while Six Sigma shares some philosophical underpinnings and techniques with other quality and process management approaches, it is distinguished by four attributes of its unique organizational approach. Schroeder et al. (2008, p. 540) dene Six Sigma as an organized, parallel-meso structure used to reduce variation in organizational processes by employing improvement specialists, a structured method, and customeroriented performance metrics with the aim of achieving strategic objectives. The typical parallel-meso structure for Six Sigma includes a centralized ofce within the rm that oversees a dispersed training and project execution hierarchy. The central ofce has several purposes. It creates an authority structure that acquires, develops, and assigns resources for training and improvement

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

439

projects. It also usually assembles an executive team (or teams) that sets criteria and guides improvement project selection (Carnell, 2003; Snee and Hoerl, 2003). In addition, this structure enables top management engagement and status reviews (Schroeder et al., 2008). By engaging high level managers in a centralized way, Six Sigma projects are thought to be less myopic and more aligned with business strategy. Finally, the structure purportedly affords more effective diffusion of lessons learned from projects, thus creating greater multi-level understanding (Sinha and Van de Ven, 2005). Six Sigma programs involve a variety of both part-time and full-time improvement specialists, including champions (executive project sponsors), master black belts, black belts, green belts, and lower level designations. The different belts denote different levels of training and experience with Six Sigma methods. Master black belts are typically full-time trainers and project mentors, while black belts and green belts are workers who may apply Six Sigma concepts and tools to drive improvements in their respective areas of functional responsibility. Black belts often have the same leadership characteristics as heavyweight project managers (Clark and Fujimoto, 1991). This hierarchy of improvement specialists is thought to enhance the coordination of work across organizational levels (Sinha and Van de Ven, 2005; Barney, 2002), and again to ensure the matching of tactical tasks with business strategy (Henderson and Evans, 2000; Linderman et al., 2003). Six Sigma improvement projects follow a structured method which has been recognized as a variation of the Plan, Do, Check, Act (PDCA) cycle (Shewhart, 1931; Schroeder et al., 2008). The Six Sigma method includes ve steps known as Dene, Measure, Analyze, Improve, and Control (DMAIC). A variant of the method used in design-oriented processes is Dene, Measure, Analyze, Design, and Verify (DMADV). Choo et al. (2007) suggest that the Six Sigma method provides an effective learning framework to guide knowledge acquisition and to ensure that project team members execute a more complete search of problem solving alternatives. It also provides a common language enabling workers to effectively communicate project status and to make comparisons across improvement efforts. Finally, Six Sigma projects are guided and assessed by a mixture of common and unique performance metrics. In addition to using typical nancial and operational project metrics, Six Sigma applies unique measures including process sigma, criticalto-quality (CTQ) attributes, and defects per million opportunities (DPMO). Researchers argue that such metrics establish challenging goals and guidance for project teams (Linderman et al., 2003; Pande et al., 2000), that they focus on objective data which tend to mitigate political agendas (Brewer, 2004), that they embody outcomes at business, process, and project levels, and ultimately that they prioritize a customer focus. 2.2. Six Sigmas operating performance impact While the popular press contains examples of both positive and negative Six Sigma impacts on performance (e.g., Pande et al., 2000; Harry and Schroeder, 2000; Chakravorty, 2010), few rigorous studies exist. A study of twenty rms by Goh et al. (2003) indicates that announcements of Six Sigma adoption produce no signicant changes in stock returns on announcement day. A further analysis of six of the rms shows that their long run stock performance is not better than the S&P 500. Zu et al. (2008, p. 643) nd that Six Sigma practices and traditional QM practices work together to generate improved quality performance, which then leads to higher business performance. However, they base these conclusions on self reported, perceptual measures of quality and performance. Also using a survey with perceptual measures, Gutierrez et al. (2009) nd that Six Sigma improves shared vision, but relationships to self-reported organizational performance are

not signicant. Braunscheidel et al. (2011) conducts case studies of seven rms and concludes that Six Sigma leads to documented savings and perceived innovation benets. The fundamental argument for net positive nancial impacts of Six Sigma adoption is that it creates new learning and adaptation capabilities within the rm. In short, Six Sigma is thought to both provide a structure and promote a culture that fosters problem/opportunity identication, process analysis, and the creation of sustained improvements. Gowen and Tallon (2005) use the dynamic capabilities theoretical perspective to describe the learning and adaptation capabilities associated with Six Sigma adoption. This perspective emphasizes the need for organizations to dynamically align their processes with changes in the business environment. Dynamic capability is dened as a learned and stable pattern of collective activity through which the organization systematically generates and modies its operating routines in pursuit of improved effectiveness (Zollo and Winter, 2002, p. 340). Dynamic capabilities enable organizations to efciently adapt, integrate, and recongure resources (Teece et al., 1997). Gowen and Tallon (2005) argue that, by addressing both technical designs and human resources, the structured approach of Six Sigma imbues the adopting organization with greater dynamic capabilities. In essence, the structured attributes of best practice identication, customer focus, and disciplined project selection and execution provide organizational architecture needed for these capabilities. Gowen and Tallon (2005) further suggest that effective Six Sigma implementations embody the value, rareness, inimitability, and non-substitutability (VRIN) characteristics associated with resources that provide competitive advantage, as specied in the resource-based view (Barney, 2002). Their survey data indicate that managers perceive this to be true; they nd signicant correlations between various Six Sigma program design factors and each of the VRIN elements. More generally, Anand et al. (2009) describe infrastructural elements of continuous improvement programs that foster dynamic capabilities. Indeed, they represent continuous improvement itself as a dynamic capability, when it is embedded in a comprehensive organizational context (p. 445). They further identify and study Six Sigma as a particular continuous improvement initiative that provides such a context. Their case study analyses indicate that infrastructural elements such as balanced innovation and improvement, a constant change culture, standardized improvement processes, and training are important enablers of organizational learning and dynamic capabilities. Consistent with the above arguments, other researchers also suggest that structured improvement methods lead to better organizational learning and knowledge transfer (Ittner et al., 2001; Choo et al., 2007; Molina et al., 2007), as well as overall improved job quality (Levine and Toffel, 2010). Linderman et al. (2003, 2006) demonstrate that the interaction of the structured method and rigorous goal setting of Six Sigma explains its impact on the performance of specic projects. Other researchers argue that advantages from process improvement programs derive mainly from social aspects (Powell, 1995), including a supportive learning culture (Detert et al., 2000; Schroeder et al., 2008; Naor et al., 2008) and cooperative values (Kull and Narasimhan, 2010). Gutierrez et al. (2009) maintain that Six Sigma adoption creates a shared vision that impels team members to work together to achieve common goals. An integration of these arguments and ndings suggests that Six Sigma adoption provides a structured approach to organizational learning that creates dynamic capabilities, specically, capabilities to consistently improve current processes (Ittner and Larcker, 1997), thereby raising quality and lowering costs. Teece (2007) maintains that such capabilities are critical for business success; due to the increasing pace and complexity of business

440

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

environments, organizations no longer compete on processes, but on the ability to continually improve processes. In order to test this proposition, our starting point is to test the hypothesis that Six Sigma adoption positively impacts overall profitability, as commonly measured by ROA (e.g., Corbett et al., 2005; Hendricks and Singhal, 1997), and dened as operating income per total assets, assessed before depreciation. Hypothesis 1. Six Sigma adoption produces a signicant positive effect on protability (ROA). Protability (ROA) is determined by return on sales (ROS, dened as operating income/sales) and asset turnover (ATO, dened as sales/assets). In turn, both ROS and ATO can also be broken down into their components, including cost of goods sold (COGS), sales, general, and administrative expenses (SG&A), current and xed assets, etc. Consistent with our second research question, should hypothesis H1 be supported, we plan to examine signicant differences in the constituent elements of ROA in order to develop insights into the nature of contributions attributable to Six Sigma adoption. In addition to prot, companies commonly seek growth, often measured by year-on-year increases in sales. A debate has emerged over the effects of process management programs on the growth of adopting rms. Researchers suggest that the disciplined structure of process management programs tends to crowd out growth oriented innovation in favor of exploitation (Benner and Tushman, 2002, 2003). Emphases on waste reduction, standardization, and continuous improvement are sometimes considered incompatible with the slack resources, exibility, and risk taking propensities needed to support more explorative efforts. At least two studies provide evidence of such effects from ISO 9000 certication. Benner and Tushman (2002) posit that an emphasis on process management biases innovation project selection toward incremental improvements and away from more exploratory efforts. In a longitudinal study of patent activity in the photography and paint industries, they document an increased share of the rms total innovations that are exploitative and build upon existing rm knowledge, post ISO 9000 certication. Similarly, Naveh and Erez (2004) nd that ISO 9000 certication is positively associated with greater process control, but negatively associated with innovation outcomes. As a process management strategy, Six Sigma has been criticized as being narrowly designed to improve existing processes, and not being helpful in the development of new products or disruptive technologies needed to drive sales growth (Morris, 2006). Hence, the dynamic capabilities stemming from Six Sigma adoption could be considered to be limited to continuous improvement, rather than also applying to more radical changes (Anand et al., 2009). Indeed, reports from well-known companies such as General Electric and 3M document managers feelings that they need to protect growth-oriented functions in the rm (e.g., marketing, R&D) from possible strictures imposed by Six Sigmas disciplined structure (Brady, 2005; Hindo, 2007; Parast, 2011). However, Schroeder et al. (2008) counter these concerns by maintaining that Six Sigma provides switching structures that simultaneously promote the conicting demands of exploration and control in the improvement effort (p. 537). Further, they identify Six Sigma black belts as boundary spanning actors, who integrate strategic concerns with tactical improvement efforts, thus facilitating exploration (Manev and Stevenson, 2001). Finally, the active engagement of top managers in project selection and monitoring are thought to foster more strategic and exploratory efforts. These conicting views leave open the question of whether Six Sigma adoption aids growth from innovation, whether it sties it, or whether it has no signicant effect at all.

Six Sigma adoption may produce other effects on sales growth that are independent of its effects on innovation and exploration. Six Sigma adoption can also serve as a signal to customer markets of improved product quality. First, as Six Sigma has gained notoriety (Gowen and Tallon, 2005), the reputational effects of adoption have potentially created greater pricing power for adopting rms. Second, real increases in product quality can be expected to lead to higher customer satisfaction. As a result, customers may be willing to pay more or to buy more; both outcomes increase sales revenue. These expectations are consistent with the ndings of Hendricks and Singhal (1997) and Corbett et al. (2005); these two studies show signicant increases in sales for rms that won quality awards and obtained ISO 9000 certications, respectively. In sum, the foregoing discussion describes three mechanisms through which Six Sigma adoption might foster greater sales growth: (1) through supporting product innovation (though this is contested in the literature), (2) through reputational enhancements that improve product and brand image, and (3) through process improvements that create better product quality (fewer defects). All three effects presumably lead to improved customer satisfaction and associated growth in sales. Hypothesis 2. Six Sigma adoption produces a signicant positive effect on sales growth. 2.3. Contextual drivers of Six Sigma implementation success Our nal research question addresses potential relationships between Six Sigma impacts and the operating context of adoption. The nature of our study affords us the opportunity to examine a number of contextual factors that can enhance or impede an adopting rms abilities to extract real benets from Six Sigma implementation. 2.3.1. Manufacturing versus service Of particular interest are differences in adoptions by primarily manufacturing versus primarily service rms. Case studies document Six Sigma adoptions in a wide variety of service rms including hospitals, government, banks and nancial services, utilities, tness clubs, retailers, and so on. Some researchers study Six Sigma programs and projects in both manufacturing and service rms as if they are universal (Schroeder et al., 2008; Nair et al., 2011). However, others identify limitations and challenges of Six Sigma methodology and tools in service settings. Some argue that it is harder to measure outcomes, collect reliable data, and control service processes (Hensley and Dobie, 2005; Antony et al., 2007). The ambiguous and customer-specic nature of critical-to-quality service features can make them difcult to dene, such that Six Sigma metrics such as DPMO are unnecessarily stringent and difcult to apply (Nakhai and Neves, 2009). Moreover, common Six Sigma tools and training topics may not adequately address differences between service customers expectations and perceptions. For example, Six Sigma does not typically address marketing communications or other inuencers of customers expectations. In addition, Six Sigma rarely addresses softer dimensions of service quality such as empathy (Nakhai and Neves, 2009). A survey by Antony et al. (2007) indicates that core Six Sigma methods such as statistical tools, process capability, and design of experiments are among the least commonly used tools in services. Their research also suggests that service process improvements are more dependent on organizational changes than manufacturing improvements are, and service organizations are at the same time more resistant to change because of the higher personal involvement of workers. Accordingly, we posit: Hypothesis 3. The positive effect on protability from Six Sigma adoption is greater for manufacturing rms than for service rms.

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

441

2.3.2. Labor intensity Rather than drawing contrasts between manufacturing and service rms, the more salient differentiator might be whether rms in heavier, more asset-intensive industries experience different results from Six Sigma adoption than those of rms in more laborintensive industries. Hendricks and Singhal (2001b) argue that labor-intense rms provide more fertile ground for quality process improvements because they have more process options and because they depend more on training and skills. Because laborintensive processes are inherently more variable, they likely offer a greater range of variance reduction opportunities. Indeed, variance reduction is the essential value in the Six Sigma paradigm. On the other hand, highly automated processes tend to be less variable. Furthermore, because of their high xed costs, processes employing heavy and automated equipment typically already receive specialized attention by highly trained process and manufacturing engineers, even in the absence of Six Sigma. Therefore, the impacts of Six Sigma adoption in these contexts may be muted by comparison. Hypothesis 4. The effect on protability from Six Sigma adoption is positively associated with the rms labor intensity. 2.3.3. R&D intensity Given the debate over the effects of Six Sigma adoption on innovation and growth, we are motivated to evaluate the relationship between R&D intensity and the performance effects of Six Sigma adoption. As noted in the discussion for hypothesis H2, a stream of research indicates that process management programs such as Six Sigma can exert a stiing effect on more exploratory innovation efforts. We would expect this effect to be particularly damaging to overall rm performance if the rm has strategically positioned itself as a leading innovator. As was the case documented at 3M (Hindo, 2007), if advanced R&D is the rms lifeblood, then Six Sigmas highly structured approach to continuous improvement might be regarded negatively by employees, potentially raising resistance to change and hampering implementation. On the other hand, if Six Sigma is embraced, organizational units might be at least implicitly encouraged to favor incremental exploitation projects over more radical exploratory efforts, thus destroying the innovative rms competitive advantage. Similarly, the benets of Six Sigmas program structure might be heightened or lessened by the importance of staying technologically current. Again, if a rm positions itself as a technology leader, danger is associated with becoming too focused on the status quo. In such a context, continuous improvement efforts that focus on process renement might be less important to success than efforts aimed at uncovering replacement technologies and entirely new opportunities. As mentioned earlier, researchers argue that Six Sigma builds dynamic capabilities and an organizational learning infrastructure that enables adopting rms to adapt more readily to changing environmental conditions (Gowen and Tallon, 2005; Anand et al., 2009). The central question becomes whether these abilities are limited to incremental changes, or whether they also apply to more dynamic, technologically intensive contexts. We forward the following hypothesis as a frame for testing these competing propositions. Hypothesis 5. The effect on protability from Six Sigma adoption is associated with the rms R&D intensity. 2.3.4. Prior nancial performance Following the logic of Hendricks and Singhal (2008), we recognize that prior operating performance can potentially affect the impacts of Six Sigma on performance, in two different ways. First, implementation resources can be a function of the rms preadoption protability. Highly protable rms likely have greater

reserves of cash and other needed resources to invest in process management infrastructural changes, training, and administration (Hendricks and Singhal, 2008). Therefore, such rms are likely to affect broader and more complete implementations. In addition, available resources enable adoption on a larger scale. For example, protable rms will likely have the capital needed to fund a larger number of simultaneous improvement projects. In these ways, protable rms can attain greater leverage from Six Sigma adoptions, thus leading to greater abnormal operating performance. On the other hand, poor performing rms may be better positioned for the changes required by Six Sigma adoption. Poor protability can be a source of motivation; i.e., employees in a loss-making rm might have a greater sense of urgency needed to implement organization changes like Six Sigma. Kotter (1995) suggests that poor business results can increase the probability of successful implementation of organizational changes, because the need for change is more apparent, and consequently the urgency and motivation required for successful implementations is more readily found in poor performers. As a result, levels of top management and organizational commitment may be higher, leading to more aggressive goal setting and a more effective implementation. Numerous writers highlight the importance of such commitment to Six Sigma success (e.g., Chakravorty, 2009; Antony et al., 2008; Linderman et al., 2006; Kumar and Antony, 2009) Hypothesis 6. The effect on protability from Six Sigma adoption is associated with the rms prior nancial performance.

2.3.5. Quality maturity Schroeder et al. (2008) note that some rms adopting Six Sigma already have quite mature quality management processes. This prompts the question of whether the impacts of Six Sigma on rm performance are contingent upon the rms prior quality management knowledge. If Six Sigma simply replicates the capabilities engendered by other quality management programs, then we might expect little additional performance gain from Six Sigma adoption. If, on the other hand, Six Sigmas program attributes are truly distinctive, as a number of researchers assert (Breyfogle, 2003; Schroeder et al., 2008; Zu et al., 2008), then we might expect unique performance gains. Operating on the premise that Six Sigma is related to, but truly distinctive from, other programs, an absorptive capacity perspective would suggest that more quality mature rms possess greater abilities to acquire, evaluate, assimilate, and exploit Six Sigma process knowledge (Zahra and George, 2002). Cohen and Levinthal (1990) describe absorptive capacity as an organizational ability to embrace and exploit new knowledge. Further, they argue that this ability depends on prior knowledge and experience. Related knowledge and experience provides a foundation for new knowledge absorption, as it creates familiarity and lessens causal ambiguities. For example, organizations experienced in quality management programs would likely speak much of the language of Six Sigma, even before adoption. In addition, employees who have gone through similar organizational transformations are likely to feel less threatened by Six Sigma-driven changes. For these reasons, we expect that rms experienced with quality oriented process management programs will implement Six Sigma faster, more completely, and perhaps more effectively. As a result, they should enjoy greater performance benets from Six Sigma program adoption than their less experienced counterparts. Hypothesis 7. The effect on protability from Six Sigma adoption is positively associated with the rms quality maturity (prior quality program experience).

442

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453 Table 1 Sample description for 214 rms with specic Six Sigma adoption years. Panel A: Frequency of Six Sigma adoption years Year 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 Frequency 2 0 2 1 1 1 1 0 0 2 3 Year 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Frequency 16 12 18 38 40 18 18 15 9 13 4

3. Research method 3.1. Sample collection and description We used multiple sources web searches, books, practitioner journals, and academic journals to identify a preliminary list of over 600 companies named as adopters of the Six Sigma philosophy and methodologies. While the list is certainly not exhaustive, it appears to be fairly representative, as it includes a wide range of industries, rm sizes, and adoption years. Of the identied rms, 421 are publicly traded companies with nancial data in Compustat. To corroborate whether the identied rms actually adopted Six Sigma, and to determine their Six Sigma adoption dates, we used key words such as Six Sigma in conjunction with history or adoption to thoroughly search publicly available documents (e.g., all publication sources in the Factiva database, corporate 10K reports, corporate websites, the Internet) for each of the 421 publicly traded rms. We retained in the sample only rms that had adopted Six Sigma in 2007 or earlier, in order to have sufcient data to establish post-adoption performance. Using these sources, we found denite, pre-2008 adoption years for 214 of the 421 rms (50.8%). For 143 rms (34.0%), we could not determine a specic adoption year, but we found enough evidence of Six Sigma activity to establish a late bound for adoption. For the remaining 64 rms (15.2%), we did not nd sufcient information to establish either an adoption date or a late bound for adoption. Because the available public information sometimes required interpretation and/or was conicting from different sources, each rm was researched independently by two members of our research team. The two independent researchers agreed on 351 of 421 ndings (83.4%) of adoption dates, late bounds, or no adoption dates. For the remaining 70 rms with disputed adoption dates or late bounds, the mean (median) difference in designated adoption years was 0.6 (1.0) years. To resolve the differences, we supplied the data sources discovered by the two researchers to a third researcher who independently weighed the evidence and determined the specic adoption years for use in our analyses. Early on in this research, we sent a survey to each publicly held Six Sigma adopter for which we could identify a credible contact person. The survey asked about adoption date and extent of adoption of the aforementioned practices that are distinct to Six Sigma (centralized team structure, improvement specialists, structured methods/DMAIC, and customer-focused metrics). We secured survey responses from 58 of the 214 publicly traded rms with identied adoption dates (23.8% of our sample). Of the 58 single respondents: 38 (65.5%) agreed with our identied adoption years; 9 (17.6%) were unable to provide a specic adoption year; 7 (12.1%) supplied an adoption date one year earlier than our nding; 3 (5.9%) supplied an adoption date more than one year later than our nding; and 1 (1.7%) supplied an adoption date one year later than our nding. We note that all three respondents with adoption dates greater than one year later from our nding were reporting only for their division within the overall rm. To be conservative, we used the earliest adoption year identied. Furthermore, the survey data indicate a remarkably uniform application of Six Sigma practices across the respondents. For example, over 90% of the respondents indicated that they employed a black/green belt structure, and over 95% designated that DMAIC and other Six Sigma tools were used on at least 80% of improvement projects. These results reinforce our overall condence in the accuracy of our estimates for both the timing and extent of adoptions in our sample rms. For the 214 rms with specic adoption years, Panel A of Table 1 presents the number of adopting rms by year. The earliest

Panel B: Occurrence (percentage) of most-frequent SIC codes 2-Digit code 35 36 28 37 38 Frequency 24 (11.2%) 23 (10.7%) 21 (9.8%) 21 (9.8%) 14 (6.5%) 3-Digit code 371 602 357 283 367 Frequency 12 (5.6%) 11 (5.1%) 10 (4.7%) 6 (2.8%) 6 (2.8%)

adoption year in our sample is 1986 and the most frequently occurring adoption year is 2001. We note the drop-off in Six Sigma adoptions in our sample post-2001. Given the continued interest and relevance of Six Sigma, as evidenced by academic publications, current business school textbooks and curriculums, and practitioner seminar offerings, we suspect that the drop-off of Six Sigma adoptions in our sample is indicative of non-newsworthiness. In other words, Six Sigma has become an accepted part of everyday business, much like TQM or Lean. This highlights the importance of rigorously studying the impact Six Sigma adoption on operating performance. Table 1 Panel B presents the most frequently occurring SIC codes within the sample rms. The sample contains rms from 47 unique two-digit SIC codes and 101 unique three-digit SIC codes. Though the majority of rms represent manufacturing industries, about one-third of the rms are services. Table 1 provides more information on the most frequently represented industries. Table 2 Panel A presents descriptive statistics for our sample based on the 2001 scal year, the most common Six Sigma adoption year in our sample. The median observation in the sample represents a rm with $5.6B in market value, $7.5B in total assets, $6.2B in annual sales, $0.8B in annual operating income, and 28,300 employees. For comparison, Table 2 Panel B presents descriptive statistics for the 207 suspected Six Sigma adopters for which we could not determine a specic adoption year. In addition, Table 2 Panel C presents descriptive statistics for all rms listed in the New York Stock Exchange (NYSE), also for the 2001 scal year. In summary, our sample represents a wide variety of industries, and is not signicantly different from the suspected Six Sigma adopters for which we could not determine a specic adoption year. However, when compared to all NYSE rms, our sample is not representative of smaller enterprises. This outcome raises a question regarding the generalizability of our ndings, as the cause of the difference is not known. Research indicates that small and medium sized rms are less likely to adopt Six Sigma, mainly because they lack requisite resources and knowledge (Antony et al., 2008; Kumar and Antony, 2009). Thus, our sample rms might be larger because of sampling bias (i.e., larger rms are more likely to be identied by our sources), but the sample rms might also be larger because they truly represent the population (i.e., larger rms are more likely to adopt Six Sigma). We note that large-rm bias is common in OM research. We discuss this limitation further in Section 6.

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453 Table 2 Descriptive statistics for 2001, the most frequent sample Six Sigma adoption year. Market value ($M) Total assets ($M) Sales ($M) 6204.9 14,205.1 21,799.3 162,412.0 49.1 3986.3 11,625.6 24,409.1 218,529.0 <0.01 864.8 4507.9 13,149.7 218,529.0 <0.01 Operating income ($M) 760.2 2337.4 4695.6 37,966.0 (5062.0) 494.0 1754.2 3429.2 29,602.0 (326.3) 137.4 838.5 2743.6 45,732.0 (5743.0)

443

Employees (000s) 28.3 51.9 69.2 395.0 0.4 17.8 47.6 119.6 1383.0 <0.01 3.9 17.9 50.7 1383.0 <0.01

Panel A: Sample rms (N = 214) Median 5616.4 7477.1 Mean 19,394.2 38,370.3 Std Dev 42,211.0 102,806.9 693,575.0 Max 397,831.6 Min 0.1 44.2 Panel B: Suspected Six Sigma adopting rms without known adoption years (N = 207) 4764.1 4684.2 Median 24,763.0 Mean 23,233.4 Std Dev 52,321.7 76,197.4 695,877.0 Max 392,959.0 0.4 0.2 Min Panel C: All NYSE rms (N = 2704) 673.2 1228.3 Median 13,040.7 Mean 5311.1 19,581.8 59,367.3 Std Dev 397,831.6 1,051,450.0 Max 0.1 Min 0.2

3.2. Event study method This section describes our choices for the period over which to measure the operating performance effects due to Six Sigma adoption, the approach to estimate the performance effects, and the statistical analyses we conducted to determine the signicance of the results. To pool data over time, we translated each rms scal years into event years as follows. We designated the scal year of Six Sigma adoption as year 0. The subsequent (previous) scal year is year +1 (year 1), etc. We analyzed data for the 6-year period from year 1 through year +4. Process management programs such as Six Sigma are often implemented over a period of months or years. Accordingly, we examined operating performance effects at two levels, on a yearto-year basis, and over aggregated multi-year periods, beginning with the change in performance from year 1 to year 0. In order to isolate the nancial impacts of Six Sigma adoption from the impacts of economic conditions and other external factors, we examined the abnormal operating performance of adopting rms. We dene abnormal operating performance for a rm as: Abnormal operating performance = Actual operating performance with Six Sigma adoption Expected operating performance if Six Sigma adoption had not occurred

Step 1. For each sample rm, we identied all rms within the same two-digit SIC code as the sample rm, and whose ROA in year 2 (the scal year immediately prior to the study period) was within 90110% of the sample rms ROA. The 10% requirement provides a tight match on operating performance. Step 2. If no rms were found in Step 1, we then matched performance within the 90110% performance range using all rms in the same one-digit SIC code. Step 3. If no rms were found in Step 2, we then matched performance within the 90110% performance range regardless of SIC code. Since we were interested in determining the effects of Six Sigma adoption, we excluded the 421 identied public Six Sigma-adopting rms from our benchmark groups. A limitation of our study is that we could not denitively determine that all benchmark rms were not also Six Sigma-adopters in the time frame studied. Because it is possible that at least some of the benchmark rms were adopters, our estimates of abnormal performance should be considered conservative. In order to dilute the effects of adopters who may be present in the benchmark groups, we set our matching criteria to maximize the number of rms in each group, thus limiting the number of one-on-one matching comparisons. In addition, we perform a separate analysis with one-to-one matching of adopters and conrmed non-adopters for a subset of the sample (described later). In order to evaluate the robustness of the results, we used two other matching methods that added controls to the matching criteria specied above. First, rather than matching on ROA only in year 2, we matched on the median ROA in years 2, 3, and 4, with the requirement that we had data for the sample and benchmark rms in at least two of the three years. We labeled this approach as the median-performance-industry matching method; this approach provides an even stricter performance control. Second, we added the matching criteria of rm size to each step. We measured rm size as median year-end total assets in years 2, 3, and 4, again with the requirement that we had data for the sample and benchmark rms in at least two of the three years. We selected benchmark rms with total assets within a factor of 25 of the assets of the sample rm. We labeled this matching method as median-performance-size-industry. Of the 214 public rms with identied Six Sigma adoption dates, we dropped rms with insufcient data in Compustat in either year 2, or years 2, 3, and 4 to compute the baseline ROA or size metrics. We eliminated ve rms from the performance-industry matching analysis, and eight rms from

The conventional approach to establishing expected operating performance is to use rms similar to the sample rms as benchmarks (controls). In their comprehensive, simulation-based research, Barber and Lyon (1996) described their three most important ndings to accurately determine expected operating performance and abnormal operating performance: (1) nonparametric Wilcoxon test statistics are more powerful than parametric t-statistics; (2) test statistics for the change in operating performance relative to an appropriate benchmark are more powerful than those based on the level of operating performance relative to the same benchmark; and (3) benchmark rms are best determined by matching on prior performance and industry. We employed each of these recommendations in our analyses. Using multiple criteria such as prior performance and industry presents a challenge when establishing benchmark groups, and sometimes require trade-offs. The multi-step process we used to establish our benchmark groups is as recommended by Barber and Lyon (1996). We label it as our performance-industry matching method:

444

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

Table 3 Matching process and benchmark group statistics. Matching method Performance in year 2a , and industry Step 1 matchesd Step 2 matchese Step 3 matchesf Total rms matched Mean group size Median group size Maximum group size No. of groups with a single rm
a b c d e f

Performance in years 2, 3, 4b , and industry 200 6 0 206 34.1 21.5 423 2

Performance and size in years 2, 3, 4c , and industry 198 7 1 206 14.9 9.0 86 8

203 6 0 209 34.3 18.0 470 6

Performance dened as ROA in year 2; matching range is 90110%. Performance dened as median ROA in years 2, 3, and 4; data required in at least two years. Size dened as median Total Assets in years 2, 3, and 4; data required in at least two years; matching range is within factor of 25. All rms within the same two-digit SIC code as the sample rm, and whose performance and/or size are within the specied range of the sample rm. All rms within the same one-digit SIC code as the sample rm, and whose performance and/or size are within the specied range of the sample rm. All rms whose performance and/or size are within the specied range of the sample rm regardless of SIC code.

Table 4 Median descriptive statistics for the matching period (years 2, 3, and 4) prior to Six Sigma adoption. ROA Total assets ($M) Market value ($M) Sales ($M) Operating income ($M) 1882.1 761.9 3672.0 32,291.0 (22.2) 625.2 122.0 2014.6 31,750.0 (9.8) Employees (000s) 47.9 25.5 64.5 413.0 0.3 11.1 3.2 28.4 775.1 <0.01

Panel A: Sample rms (N = 214) 0.1362 12,932.2 23,086.6 10,450.9 Mean 0.1333 4303.1 5356.9 4889.7 Median 0.0755 23,901.5 52,784.0 16,311.7 Std Dev 250,138.5 303,989.0 146,991.0 Max 0.4627 Min (0.2629) 22.1 8.3 15.7 Panel B: Benchmark rms (obtained from median-performance-size-industry matching method (N = 3077) 0.1313 9869.5 3981.7 2928.1 Mean 0.1307 915.1 735.9 738.1 Median 0.0605 44,271.1 20,646.7 8732.1 Std Dev 0.4331 933,559.1 911,494.2 174,694.0 Max 5.1 1.5 0.2 Min (0.2736)

both the median-performance-industry and median-performancesize-industry matching analyses. Table 3 presents the results of each matching process. In each process, the vast majority of matches were accomplished in Step 1, using the tightest matching criteria. In the performance-industry matching, for example, 203 of the 209 benchmark groups were generated using Step 1 of our matching process, and the remaining six benchmark groups were generated using Step 2. The mean (median) benchmark group size was 34.3 (18.0), and only six rms were benchmarked against a single rm. The other matching methods produced similar results, though notably in the median-performance-size-industry matching, the mean (median) matching group size dropped to 14.9 (9.0), and one rm required Step 3 for matching. Table 4 presents median descriptive statistics for the pre-adoption matching period (years 2, 3, and 4) for both the sample rms and the benchmark rms obtained using the median-performance-size-industry matching method. The sample rms and benchmark rms are wellmatched on ROA and assets within the stated criteria (10% and factor of 25, respectively). Due to the common occurrence of extreme outliers in accounting-based performance measures, we used the median (versus the mean) change in the benchmark rms to estimate the sample rms expected operating performance without Six Sigma adoption. The sample rms expected operating performance without Six Sigma adoption is the sum of its actual performance in the base year(s) plus the median change in operating performance for the rms in the benchmark group over the period of interest. The sample rms abnormal operating performance with Six Sigma adoption is then the difference between the sample rms actual performance (with Six Sigma adoption) and the sample rms expected performance (without Six Sigma adoption). As an

example, if a sample rms benchmark group experienced a 3.0% median increase in ROA from year 1 to year 0, we would expect the sample rm to also increase its ROA by 3.0%. If the sample rm actually experienced a 3.5% ROA increase over this period, we would estimate its abnormal performance due to Six Sigma adoption to be 0.5%. To control for outliers and non-normality, we trimmed the data at the 2.5% level in each tail for all analyses. Although we report means and changes in means using t-statistics, we emphasize nonparametric statistics and tests. We used the Wilcoxon signed-rank test to determine whether the median of a single group was significantly different from zero, and the binomial sign test to examine whether the percentage of sample rms experiencing positive operating performance was signicantly greater or less than the null of 50%. Given that we hypothesize an improvement in operating performance due to Six Sigma, we report one-tailed tests of signicance except where noted. 3.3. Measures Using Compustat and other sources, we collected data describing the contextual factors addressed in hypotheses H3H7. First, to test hypothesis H3, we employ a dummy variable to denote rms that are primarily focused on the production of goods (manufacturing) from those that are primarily focused on the provision of services. The industry dummy variable has the value 1 if the rm is in a manufacturing industry, and the value 0 if the rm is in a service industry. We identify rms as being primarily manufacturers if their SIC codes begin with 2 or 3. The service rms include rms in the transportation, communications, and utilities industries (SIC codes beginning with 4), wholesale trade and retail trade industries

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

445

(SIC codes beginning with 5), nance, insurance, and real estate industries (SIC codes beginning with 6), and other service industries (SIC codes beginning with 7 and 8). We evaluated each of the remaining rms in the agriculture, forestry, and sheries industries (SIC codes beginning with 0), mining and construction industries (SIC codes beginning with 1), and the public administration industry (SIC codes beginning with 9) to determine whether its primary business is the production of goods or the provision of services. Thus, we identied 144 rms in manufacturing industries and 70 rms in services industries. To test hypothesis H4, we consider the rms labor intensity. We measure labor intensity as the rms number of employees in year 2 divided by its sales in year 2. We use the rms R&D intensity to provide testing of hypothesis H5. We dene R&D intensity as the rms R&D expenses in year 2 divided by its sales in year 2. We note that this analysis is limited to manufacturing rms, because few service rms report R&D expenditures. To examine the possibility of different response coefcients for prot- and loss-making rms (H6), we follow the method of Hendricks and Singhal (2008). For each adopting rm, we compute industry-adjusted ROA as the rm ROA in year 2 minus the median ROA of its industry in year 2. Using these scores, we created two prior operating performance variables as follows. If the industryadjusted ROA is positive, we used the value as our positive nancial performance variable, and 0 otherwise. If the industry-adjusted ROA is negative, we used the value as our negative nancial performance variable, and 0 otherwise. Finally, we use ISO 9000 certication status as an indicator of quality maturity (H7). Using the method employed by Corbett et al. (2005), we determined the year of initial ISO certication for each rm in our sample. We then computed the number of years between initial ISO certication and Six Sigma adoption as a proxy for the rms maturity in applying quality management principles and techniques. We coded rms as 1 if they were not ISO certied, or if they were certied only after Six Sigma adoption. In addition to the above factors, we include three control variables in the analysis. Larger rms likely have more implementation resources, and economies of scale to better exploit improvements. Conversely, smaller rms tend to be more focused and agile, easing adoption of new improvement philosophies. In addition, if wellimplemented, the relative impact of any one improvement program such as Six Sigma is likely to be greater in a small rm. Hence, we control for the effects of rm size, operationalizing it as the natural log of the rms market value at the end of year 2. Market value provides a suitable measure of size, given that rm sales, assets, and number of employees are already represented in the independent and dependent variables described earlier. We also include the year of Six Sigma adoption as a control for changes in economic, business environment, and other time-related variables that might confound the results. Lastly, we control for leadership changes in adopting rms. It can be argued that, if Six Sigma adoption is instigated by new leadership, then observed benecial effects might be attributable to new leadership competencies or motivational factors, rather than to Six Sigma adoption per se. To control for this possibility, we used the Execucomp database to determine sample rm CEO changes in years 0, 1 and 2. For the 32 sample rms not listed in Execucomp, we searched all sources in the Factiva database for evidence of CEO turnover in the three-year time period. In total, 90 rms in our sample (42.4%) experienced a CEO change within the specied period. We note that the mean (median) CEO turnover rate in Execucomp is 40.4% (52.9%) for a three-year period, suggesting that our sample rms are not more prone to CEO changes than other rms. Sample rms with a CEO change were coded as 1; other rms were coded as 0.

4. Results We begin the discussion of our empirical results by focusing on the effects of Six Sigma adoption on the rm protability measure, ROA, on both an annual basis and for the aggregated multiple-year periods. We then decompose ROA into its constituent components to determine whether changes in ROA are due to changes in ROS, COGS/sales, SG&A/sales, ATO, or a combination thereof. We also examine the impact of Six Sigma adoption on rm growth by considering year-on-year changes in sales. Lastly, we present our results contingent on the contextual factors of industry, and preadoption labor intensity, R&D intensity, nancial performance, and quality maturity.

4.1. Overall performance effects Table 5 presents the results for abnormal changes in the level of ROA on an annual basis, for each of our three matching methods. Given that our baseline period to establish benchmark groups is year 2 for the performance-industry match, and years 2, 3, and 4 for the median-performance-industry and medianperformance-size-industry matches, we consider the ve annual changes beginning with the change from year 1 to year 0 and continuing through the change from year +3 to year +4. As aggregate measures, we also consider the change per rm for three multipleyear periods: from year 1 to year +4; year 0 to year +4; and year +1 to year +4. By comparing results across the annual and multiple-year periods, we can see whether the operating performance impacts of Six Sigma adoption begin immediately or are lagged, and whether the effects are persistent. Our rst observation is that, regardless of matching method, the annual changes in abnormal ROA for the sample rms are consistently positive. For performance-industry matching, 5 of 5 (4 of 5) median (mean) annual changes are positive, and the % rms with positive changes is greater than 50% for 5 of 5 annual changes. Two of the 5 annual changes (from year 0 to +1, and year +3 to +4) are statistically signicant in all three tests for median, mean, and % positive. For median-performance-industry matching, 4 of 5 median and mean annual changes are positive, and the % rms with positive changes is greater than 50% for 4 of 5 annual changes. Three of the 5 annual changes (from year 0 to +1, year +1 to +2, and year +3 to +4) are statistically signicant in the median, mean, and % positive test. For median-performance-size-industry matching, 4 of 5 median and mean annual changes are positive, and the % rms with positive changes is greater than 50% for 5 of 5 annual changes. However, only 1 of the 5 annual changes (from year +3 to +4) is statistically signicant for the median, mean, and % positive. We next consider the results for the multiple-year periods. For all three matching methods, the change per rm over the 6-year period (from year 1 to +4) is signicantly positive for both the mean and median, and the % rms with positive 6-year changes are signicantly greater than 50% for two of the three matching methods. For performance-industry matching, the median and mean changes per rm over the 6-year period are 0.355% and 1.214%, respectively, both signicantly positive at the 1% level, and 55.69% of sample rms experience positive changes, signicantly greater than 50% at the 10% level. For median-performance-industry matching, the median (mean) change per rm is 0.703% (1.294%) over the 6-year period, signicantly positive at the 1% (1%) level, and 56.10% of sample rms experience positive median annual changes, signicantly greater than 50% at the 10% level. For medianperformance-size-industry matching, the median (mean) change per rm is 0.433% (1.029%) over the 6-year period, signicantly positive at the 5% (5%) level, and 53.99% of sample rms experience positive changes, insignicantly greater than 50%.

446

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

Table 5 Annual abnormal changes in ROA for sample rms for year 1 through year +4. From year: N Median Z-Statistic Mean t-Statistic % Positive Z-Statistic 0.641 2.585*** 0.941 0.520 2.173** 1.470* 0.542 0.077 1.580* 2.388*** 1.818** 0.520 2.030** 1.562* 1.718** 1.406* 0.287 0.217 0.438 0.075 2.428*** 1.018 0.705 1.018 0.686 1.480* 1.441* 1.000 1.029 0.686 1.521* 2.000**

Panel A: Performance-industry matching 197 0.121% 0.674 0.001% 0.004 52.28% 1 to 0 0 to +1 194 0.469% 2.741*** 0.600% 2.884*** 59.28% +1 to +2 191 0.180% 1.040 0.129% 0.633 53.40% 181 0.113% 0.578 0.115% 0.491 51.93% +2 to +3 166 0.423% 2.192** 0.447% 1.947** 58.43% +3 to +4 1 to +4 167 0.355% 2.390*** 1.214% 2.729*** 55.69% 0 to +4 167 0.077% 1.639* 0.994% 2.314** 52.10% * +1 to +4 167 0.030% 0.917 0.540% 1.406 50.30% Panel B: Median-performance-industry matching 194 0.256% 1.002 0.130% 0.568 55.67% 1 to 0 0 to +1 191 0.378% 2.208** 0.391% 1.904** 58.64% +1 to +2 189 0.241% 2.120** 0.367% 1.834** 56.61% +2 to +3 181 0.099% 0.634 0.034% 0.170 48.07% +3 to +4 164 0.552% 1.679** 0.336% 1.497* 57.93% *** *** 1 to +4 164 0.703% 2.632 1.294% 3.108 56.10% ** ** 0 to +4 164 0.530% 2.019 0.901% 2.262 56.71% +1 to +4 164 0.411% 1.541* 0.559% 1.730** 55.49% Panel C: Median-performance-size-industry matching 194 0.055% 0.415 0.049% 0.190 51.03% 1 to 0 0 to +1 191 0.080% 0.482 0.021% 0.103 50.79% 188 0.061% 0.989 0.223% 1.071 51.60% +1 to +2 179 0.000% 0.172 0.017% 0.086 50.28% +2 to +3 163 0.465% 1.995** 0.471% 1.897** 59.51% +3 to +4 1 to +4 163 0.433% 1.900** 1.029% 2.322** 53.99% 0 to +4 163 0.479% 1.569* 0.914% 2.097** 52.76% 163 0.395% 1.696** 0.826% 2.223** 53.99% +1 to +4 Panel D: One-on-one median-performance-size-industry matching using only early adopters matched against sample rms that adopted at least ve years later 1 to 0 34 0.150% 0.244 0.111% 0.222 55.88% 0 to +1 34 0.592% 1.362* 0.932% 1.488* 62.16% +1 to +2 34 0.172% 0.904 0.717% 1.244 61.54% +2 to +3 34 0.413% 0.962 0.280% 0.588 41.67% +3 to +4 34 1.197% 2.039** 1.220% 2.370*** 58.82% 1 to +4 34 1.383% 1.406* 1.784% 1.628* 55.88% 0 to +4 34 1.740% 1.863** 2.025% 2.220** 62.86% +1 to +4 34 1.759% 2.101** 2.211% 2.362*** 66.67% All samples trimmed at 2.5% each tail. Z-Statistics for medians are obtained using Wilcoxon Signed-Rank tests. Z-Statistics for % positive are obtained using Binomial Sign tests. * Signicance is one-tailed: p .10 ** Signicance is one-tailed: p .05. *** Signicance is one-tailed: p .01.

The change per rm over the 5-year period (from year 0 to +4) is also signicantly positive for both the mean and median, using all three matching methods. The % rms with positive 5-year changes are signicantly greater than 50% using the median-performanceindustry matching method. As expected, the magnitudes of the 5year changes are generally less than those of the 6-year changes. The change per rm over the 4-year period (from year +1 to +4) follows a similar pattern and is again generally lesser in magnitude and signicance than the 5-year changes. As we noted in Section 3.2, a limitation of our study is that we could not denitively determine that all benchmark rms were not also Six Sigma-adopters during the sampling time frame. If many of the benchmark rms were truly adopters, then estimates of abnormal performance would be muted, making signicant differences difcult to detect. The fact that we do nd signicant differences suggests that, in the worst case, our ndings of abnormal performance due to Six Sigma adoption are conservative. To address this limitation, we would ideally match known adopters only with known non-adopters. Such a pure comparison would produce a more reliable estimate of expected performance improvement from Six Sigma adoption. To approximate this pure comparison, we identied known non-adopters as rms from our sample that adopted Six Sigma at least ve years later than earlier sample adopter rms. The ve-year delay allows us to consider operating performance impacts to the early adopters during the post-adoption period through year +4. Given that we are limited to

only the 214 rms in our sample, and to permit the greatest number of comparisons, we matched each adopter against only a single rm, and did not use data from any one rm more than once; this method permitted 41 matches using the ROA, assets, and industry criteria from median-performance-size-industry matching. Table 5 Panel D presents the results for abnormal changes in ROA for year 1 through year +4. The ROA improvements are signicant and generally stronger than in our other matching methods. These results should be regarded as somewhat tentative, given the small sample size, and the confounding with time due to this matching methodology (the adopters all adopted prior to 2003). However, the results do strongly reinforce the conclusion that the estimated improvements from our larger analyses are real, albeit conservative. Considering both the annual changes and multiple-year changes indicates that Six Sigma adoption produces an immediate and persistent positive effect on ROA. These results provide support for hypothesis H1. 4.2. Decomposition of ROA effects For brevity in presenting and discussing the remainder of our results, we concentrate on our most conservative matching method, median-performance-size-industry, noting that the pattern of results is similar regardless of matching method. Table 6 Panel A presents the results for the abnormal changes in the level of ROS on an annual basis and for multiple-year periods. For the 6-year

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

447

Table 6 Annual abnormal changes in ROS, COGS/sales, SG&A/sales, and ATO for sample rms for year 1 through year +4 using the median-performance-size-industry matching method. From year: N Median 0.117% 0.048% 0.127% 0.167% 0.482% 0.899% 0.632% 0.892% 0.528% 0.260% 0.214% 0.200% 0.173% 0.095% 0.533% 0.336% 0.438% 0.007% 0.096% 0.155% 0.149% 0.219% 0.842% 0.639% 0.604% 0.295% 0.055% 0.633% 0.605% 0.130% 0.004% 0.027% 0.253% Z-Statistic 0.320 0.009 1.151 0.249 1.955** 0.937 1.363* 1.761** 2.063** 0.106 0.193 1.927** 0.344 0.738 0.374 0.225 0.018 0.319 0.061 0.762 2.277** 1.979** 2.846*** 2.140** 2.476*** 0.882 0.512 1.299* 0.626 0.802 0.715 0.367 0.095 Mean 0.324% 0.117% 0.328% 0.065% 0.364% 0.342% 0.490% 0.596% 0.645% 0.222% 0.006% 0.447% 0.086% 0.211% 0.281% 0.150% 0.033% 0.051% 0.123% 0.023% 0.342% 0.297% 1.094% 0.771% 0.747% 0.200% 0.866% 0.680% 0.164% 1.067% 1.944% 1.120% 0.026% t-Statistic 1.406* 0.492 1.449* 0.287 1.639* 0.774 1.114 1.563* 1.976** 0.987 0.026 2.346*** 0.407 0.952 0.539 0.317 0.087 0.370 0.855 0.156 2.639*** 2.190** 2.832*** 2.108** 2.707*** 0.203 1.112 0.795 0.184 1.199 0.992 0.634 0.018 % Positive 51.55% 50.79% 51.60% 52.51% 59.51% 52.76% 56.44% 56.44% 56.44% 45.00% 45.91% 45.86% 46.98% 48.89% 53.03% 51.13% 46.97% 49.38% 55.35% 47.13% 41.61% 40.00% 38.64% 42.86% 38.64% 44.33% 49.74% 44.68% 45.25% 51.53% 49.69% 49.08% 48.47% Z-Statistic 0.431 0.217 0.438 0.673 2.428*** 0.705 1.645* 1.645* 1.645* 1.265 1.031 1.038 0.737 0.258 0.696 0.260 0.696 0.158 1.348* 0.718 2.048** 2.324** 2.611*** 1.648** 2.611*** 1.580* 0.072 1.459* 1.271 0.392 0.078 0.235 0.392

Panel A: Changes in abnormal ROS 194 1 to 0 0 to +1 191 188 +1 to +2 179 +2 to +3 163 +3 to +4 1 to +4 163 0 to +4 163 +1 to +4 163 +1 to +4 effecta 163 Panel B: Changes in abnormal COGS/sales 160 1 to 0 159 0 to +1 +1 to +2 157 149 +2 to +3 135 +3 to +4 132 1 to +4 0 to +4 133 132 +1 to +4 Panel C: Changes in abnormal SG&A/sales 160 1 to 0 159 0 to +1 +1 to +2 157 +2 to +3 149 +3 to +4 135 1 to +4 132 0 to +4 133 +1 to +4 132 Panel D: Changes in abnormal ATO 194 1 to 0 0 to +1 191 +1 to +2 188 +2 to +3 179 163 +3 to +4 163 1 to +4 163 0 to +4 +1 to +4 163

All samples trimmed at 2.5% each tail. Z-Statistics for medians are obtained using Wilcoxon Signed-Rank tests. Z-Statistics for % positive are obtained using Binomial Sign tests. a Effect on ROA computed per rm as ROS Change +1 to +4 Firm ATO in year 0. * Signicance is one-tailed: p .10. ** Signicance is one-tailed: p .05. *** Signicance is one-tailed: p .01.

period from year 1 to +4, the median and mean change per adopting rm, and the % of sample rms experiencing positive change, are all positive but insignicant. For the 5-year period from year 0 to +4, the median change is 0.632%, signicantly positive at the 10% level. The mean change is 0.490%, positive but insignicant, and 56.44% of rms experience positive changes, signicantly greater than 50% at the 10% level. For the 4-year period from year +1 to +4, the median and mean changes are 0.528% and 0.645%, respectively, both signicantly positive at the 5% level, and 56.44% of rms experience positive changes, signicantly greater than 50% at the 10% level. To determine whether the improvement in ROS from year +1 to +4 contributes signicantly to the improvement in ROA, we employ the method of Kinney and Wempe (2002). For each adopting rm, we compute the effect of ROS on ROA by multiplying the change in abnormal ROS from year +1 to +4 with the rms ATO at adoption (year 0). The median and mean ROS effects over the 4year period are 0.528% and 0.645%, respectively, both signicantly positive at the 5% level, and 56.44% of sample rms experience positive ROS effects, signicantly greater than 50% at the 10% level. These results indicate that the ROS improvement from Six Sigma adoption signicantly contributes to the overall improvement in ROA.

The primary cost components that impact operating income (and hence, ROS and ROA) are COGS and SG&A. To determine the contributions of both components, we examined abnormal changes in COGS/sales and SG&A/sales. Given that all rms do not consistently report COGS and SG&A separately each year, we included only adopting rms and benchmark rms in our analyses that did report both accounts. This eliminated approximately 35 adopting rms from our sample. Table 6 Panels B and C present the results for the abnormal changes in the level of COGS/sales and SG&A/sales, respectively, on an annual basis and for the multiple-year periods of interest. Although the median annual changes in COGS/sales are all negative, only the change from year +1 to +2 is signicant. The mean (median) abnormal change from year +1 to +2 is 0.200% (0.444%), signicant at the 5% (1%) level. For the 6-year period (year 1 to +4) and the 5-year period (year 0 to +4), the median, mean, and % positive changes are all positive but insignicant. For the 4-year period (year +1 to +4), the change statistics are negative but insignicant. The evidence provides no support that Six Sigma adoption produces signicant reductions in COGS. Considering the annual abnormal changes in SG&A/sales for the sample rms, we see that they are consistently negative. Four of 5 (5 of 5) median (mean) annual SG&A/sales changes are negative, and the % rms with positive changes is less than 50% for 4 of 5

448

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

Table 7 Annual abnormal changes in sales growth for sample rms for year 1 through year +4. From year: N Median Z-Statistic 1.693 ** 0.570 0.753 1.278 0.250 0.229 0.103 0.378 0.260 0.225 0.856 0.396 2.681*** 1.725** 1.876** 1.952** 0.575 0.326 0.158 0.046 1.410* 1.483* 1.078 1.403* Mean 0.555% 1.148% 0.498% 1.613% 0.749% 3.179% 1.985% 0.034% 0.957% 1.101% 0.417% 0.208% 3.138% 11.897% 8.118% 5.344% 2.127% 0.774% 0.237% 0.071% 2.098% 11.022% 5.006% 3.759% t-Statistic 0.589 1.284* 0.501 1.536* 0.853 0.704 0.545 0.012 0.950 1.182 0.411 0.217 3.296*** 2.803*** 2.416*** 2.028** 1.973** 0.859 0.234 0.072 2.345*** 2.485*** 1.422* 1.441* % Positive 43.65% 46.91% 47.64% 47.51% 46.39% 43.11% 49.70% 48.50% 47.94% 46.60% 48.15% 49.72% 57.32% 52.44% 54.88% 56.71% 50.00% 46.07% 51.06% 49.72% 53.37% 50.92% 53.99% 55.83% Z-Statistic 1.781** 0.862 0.651 0.669 0.931 1.780** 0.077 0.387 0.574 0.941 0.509 0.074 1.874** 0.625 1.249 1.718** 0.000 1.085 0.292 0.075 0.862 0.235 1.018 1.488*

Panel A: Performance-industry matching 197 1.398% 1 to 0 0 to +1 194 0.646% +1 to +2 191 0.664% 181 0.305% +2 to +3 166 0.689% +3 to +4 167 5.718% 1 to +4 0 to +4 167 0.094% +1 to +4 167 1.533% Panel B: Median-performance-industry matching 194 0.710% 1 to 0 191 0.532% 0 to +1 189 0.486% +1 to +2 181 0.030% +2 to +3 +3 to +4 164 1.540% 1 to +4 164 2.463% 164 3.485% 0 to +4 164 5.631% +1 to +4 Panel C: Median-performance-size-industry matching 194 0.008% 1 to 0 0 to +1 191 1.485% 188 0.271% +1 to +2 179 0.030% +2 to +3 163 1.259% +3 to +4 163 0.695% 1 to +4 0 to +4 163 3.502% 163 6.182% +1 to +4

All samples trimmed at 2.5% each tail. Z-Statistics for medians are obtained using Wilcoxon Signed-Rank tests. Z-Statistics for % positive are obtained using Binomial Sign tests. * Signicance is one-tailed: p .10. ** Signicance is one-tailed: p .05. *** Signicance is one-tailed: p .01.

annual changes. Two of the 5 annual changes (from year +2 to +3, and year +3 to +4) are statistically signicant in all three tests for median, mean, and % positive. For the 6-year period from year 1 to +4, the median (mean) change is 0.842% (1.094%), signicantly negative at the 1% (1%) level, and 38.64% of sample rms experience positive median annual changes, signicantly less than 50% at the 1% level. For the 5-year period from year 0 to +4, the median and mean changes are 0.639% and 0.771%, respectively, both signicant at the 5% level. 42.86% of rms experience positive changes in SG&A/sales for the 5-year period, signicantly less than 50% at the 5% level. Similarly, the mean, median, and % positive changes for the 4-year period from year +1 to +4 are all negative and signicant at the 1% level. These results suggest that Six Sigma improvements signicantly and persistently reduce SG&A costs. Table 6 Panel D presents the results for the abnormal changes in the level of ATO. Interestingly, the annual changes in abnormal ATO for the sample rms are generally negative. Five of 5 (2 of 5) median (mean) annual changes are negative, and the % rms with positive changes is less than 50% for 4 of 5 annual changes. However, only the change from year +1 to +2 is marginally signicant. For the multiple-year periods, the median and % positive changes are all negative but insignicant. The results therefore indicate no signicant relationship between Six Sigma adoption and ATO. 4.3. Sales growth effects In order to test hypothesis H2, we next consider the effect of Six Sigma adoption on changes in sales. Table 7 presents the results for abnormal sales growth on an annual basis, for each of our three matching methods. We note that, more so than for any other of

our performance measures, the results for abnormal sales growth appear somewhat sensitive to the matching method employed. There are at least two plausible reasons for this discrepancy: (1) sales growth data are inherently noisier than our other measures since they are percentage changes in absolute numbers rather than differences in ratios and (2) changes in sales are correlated with rm size, so the size control added in median-performance-size-industry matching has a greater effect. Accordingly, we again concentrate our discussion on the results from the most conservative matching method, median-performance-size-industry, presented in Panel C. The annual abnormal % changes in sales for the sample rms are neither consistently positive nor negative. Two of 5 (5 of 5) median (mean) annual changes are positive, and the % rms with positive changes is greater than 50% for 3 of 5 annual changes. Only the positive change from year +3 to +4 is statistically signicant in the tests for median and mean. For the 6-year period from year 1 to +4, the median (mean) change is 0.695% (11.022%), signicantly different from zero at the 10% (1%) level; 50.92% of sample rms experience positive changes, insignicantly different from 50%. For the 5-year period from year 0 to +4, the median and % positive changes are positive but insignicant; the mean change is 5.006%, signicant at the 10% level. For the 4-year period from year 0 to +4, the median, mean, and % positive changes are all signicantly positive at the 10% level. We conclude that the results provide only limited support for Hypothesis 2, that Six Sigma adoption positively impacts sales growth. The foregoing ndings collectively indicate that signicantly improved ROA in adopting rms is primarily due to indirect cost reductions (SG&A), and perhaps mildly due to positive sales growth. Both of these changes are reected in improved ROS, rather than to improvements in ATO.

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

449

Table 8 Estimated coefcients (standardized, t-Statistics in parentheses) from regressions of abnormal ROA change from year 1 to year +4 using the median-performance-size-industry matching method. Independent variables Operationalization Model 1 manufacturing and Services 3.875 (1.203) 0.074 (0.790) 0.242 (2.932)*** Model 2 services only 7.646 (1.565) Model 3 manufacturing only 3.830 (0.907) Model 4 manufacturing only 5.589 (1.175)

Intercept Manufacturing or services Labor intensity R&D intensity 1 if manufacturing, 0 if services Employees/sales R&D/sales

0.128 (0.852)

0.343 (3.199)***

Positive nancial performance

Negative nancial performance

ISO9000 experience

Firm size Adoption year New CEO

Industry-adjusted ROA if positive, 0 otherwise Industry-adjusted ROA if negative, 0 otherwise Six Sigma adoption year minus 1st ISO9000 certication ln(market value)a YearO 1 if new CEO in years 0, 1, or 2, 0 otherwise Number of observations Model F value R2 Adjusted R2

0.059 (0.725) 0.136 (1.664)* 0.254 (2.551)**

0.358 (2.483)*** 0.374 (2.499)** 0.201 (1.278)

0.081 (0.831) 0.103 (1.083) 0.207 (1.755)*

0.325 (2.661)*** 0.042 (0.383) 0.115 (0.934) 0.118 (1.131) 0.264 (2.024)**

0.027 (0.317) 0.113 (1.200) 0.049 (0.612) 156 2.418 11.63% 6.82%
**

0.192 (1.307) 0.219 (1.553) 0.134 (0.954) 47 2.346 29.63% 17.01%

**

0.000 (1.000) 0.110 (0.903) 0.037 (0.385) 109 2.758 16.05% 10.23%
**

0.067 (0.542) 0.168 (1.172) 0.023 (0.220)1 97 2.181** 16.55% 8.96%

All samples trimmed at 2.5% each tail. a Alternative operationalizations of rm size [ln(Sales), ln(Employees), ln(Total Assets)] yield substantively similar results. * Signicance is two-tailed: p .10. ** Signicance is two-tailed: p .05. *** Signicance is two-tailed: p .01.

4.4. Relating abnormal ROA performance to Six Sigma adopter context To examine the data for support of hypotheses H3H7, which address the potential roles of contextual factors in Six Sigma adoptions, we perform cross-sectional analyses. To assess the impacts to operating performance over the entire study period, we use the abnormal ROA change from year 1 to +4, obtained using the median-performance-industry-size matching method, as our dependent variable. Table 8 shows the results for regressions of the abnormal performance of Six Sigma adopters on the contextual variables for the entire sample (Model 1), the service and manufacturing subsamples (Models 2 and 3, respectively), and the manufacturing subsample with rms reporting R&D intensity (Model 4). We note that all four models are signicant at the 5% level. We review the results in the order of our hypotheses and discuss them further in the next section. In Model 1, the coefcient for the dummy indicator of manufacturing or service industry is not statistically signicant, indicating that the ROA benets of Six Sigma adoption are not signicantly greater for manufacturing rms than for service rms. This fails to support hypothesis H3. Despite the lack of signicant difference in the overall benet of Six Sigma adoption between manufacturing and service rms, we examine Models 2, 3, and 4 to determine whether the contextual factors that can impact an adopting rms abilities to extract benets from Six Sigma implementation differ between rms in manufacturing or services.

A signicant association indicated in the results presented in Table 8 pertains to labor intensity. The labor intensity coefcient is positive and signicantly different from zero at the 1% level for the total sample and manufacturing subsamples (Models 1, 3, and 4). Labor intensity is positive but insignicant for the service industry subsample (Model 2). Thus, hypothesis H4 is supported, but only for manufacturing rms. As noted previously, the impact of R&D intensity can only be evaluated for our manufacturing subsample as most services rms do not report R&D expenses. The results from Model 4 indicate no signicant effect of R&D intensity on abnormal ROA changes. Thus, hypothesis H5 is not supported. The results from Model 1 indicate that pre-adoption protability is signicantly correlated with abnormal ROA from Six Sigma adoption by sample rms, but only at a marginal level (p 0.10) and only for rms with negative nancial performance. Given that the values of the negative nancial performance variable are nonpositive by denition, the negative coefcient indicates greater benets for Six Sigma adopters with more negative prior nancial performance. The results from Model 2 demonstrate that both positive and negative pre-adoption nancial performance is associated with greater abnormal performance for service rms. These results suggest that, while overall effects of Six Sigma adoption are typically positive in service rms, they may be amplied by either the nancial strength or weakness of the rm at the time of adoption. Interestingly, these ndings are not signicant for manufacturing rms. Thus, hypothesis H6 is supported, but only for service rms. The cross-sectional analyses yield signicant results for the ISO 9000 experience variable. The ISO 9000 experience coefcients for

450

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

the total sample and manufacturing subsamples (Models 1, 3, and 4) are signicantly negative at the 5%, 10%, and 5% levels, respectively. To understand this counter-intuitive result better, we examined the raw data. For the 21 rms that were certied to ISO 9000 after Six Sigma adoption, the median and mean abnormal ROA changes from year 1 to year +4 are 2.615% and 3.579%, respectively, both signicantly greater than zero at the 1% level. For the 124 rms that were ISO 9000 certied prior to Six Sigma adoption, the median and mean abnormal ROA changes are 0.501% and 0.471%, neither significantly different from zero. The differences in medians and means between these two groups are both signicantly different from zero at the 1% level. This suggests that rms with greater quality maturity benet less from Six Sigma adoption, a nding counter to our hypothesis H7. We note that none of our control factors rm size, adoption year, new CEO signicantly impact the benets from Six Sigma adoption.

5. Discussion of the results Overall, the results indicate that the benets of Six Sigma adoption tend to more than compensate for associated costs and required investments. Recalling that our estimates are conservative, Six Sigma adopters should expect an addition to ROA of at least 0.20.3 percentage points each year on average. These magnitudes of change are both statistically and economically signicant. The results from the median-performance-size-industry method, which are the most conservative, indicate that the sample rms abnormal ROA increased on average by 1.029 percentage points in total over the 6-year period from year 1 to +4, or an average of 0.206 percentage points improvement per year. Given that the median sample rm had an ROA of 13.22% in year 2, this change represents a 7.8% improvement relative to non-adopting rms. For the median sample rm with $7.0B in assets in year 2, such an improvement translates into roughly $220M in additional operating income over the 6-year period for the same asset base. While quite signicant, this ROA boost nevertheless appears to be modest in comparison to results indicated in other process management event studies. Corbett et al. (2005) nd an average yearly ROA increase of 0.89 percentage points associated with ISO 9000 certication. Hendricks and Singhal (1997) nd an average yearly ROA increase of 5.01 percentage points in years 1 to +3 for winners of quality awards. We hesitate to conclude that Six Sigma actually offers less of an impact than these other programs, however. As we noted earlier, estimates from our study are conservative, given the possibility that some rms populating our benchmark groups could have also adopted Six Sigma before or during the sampled time horizon. Indeed, the average ROA boost indicated from our one-onone matching process (Table 5 Panel D) is nearly double that of the more conservative matching process. While Hendricks and Singhal (1997) cite similar potential pollution of benchmark groups as a limitation of their study, it is important to note that they study quality award winners, i.e., successful purveyors of quality initiatives. Thus, their sample is upwardly biased. There is no reason to believe that our sample is similarly upward biased; our results may represent a more varied and realistic success rate in process management implementation. Also important, our performance matching criterion (3-year median) is stricter than the criteria used by either of the other two studies. Six Sigma is a relative newcomer to the ranks of process improvement programs. It is likely that many Six Sigma adopters have already put TQM, lean, or other strategies in place prior to Six Sigma. In these rms, manager may have already targeted low hanging fruit in previous process improvements, and we should therefore not be surprised by the relatively small operating

performance improvements associated with Six Sigma reected in our overall sample. Indeed, our nding regarding quality maturity calls into question the argument that Six Sigma is truly distinctive from other quality oriented management processes. Using the absorptive capacity perspective, we argued that rms with greater experience in quality management should benet more from Six Sigma adoption, because Six Sigma is at the same time similar to, yet distinct from, prior quality management programs. Since the empirical evidence demonstrates less benet for rms with greater quality maturity, a more likely conclusion is that the capabilities stemming from Six Sigma adoption add little measurable value over and above those that emanate from ISO 9000 certication. Thus, while Six Sigma may entail distinctive organizational structures, problem solving tools, and metrics, these attributes appear to be less important for already experienced rms. The managerial implications of this nding for rms with and without mature quality systems could be far-reaching, and thus require further research. Our ndings support the theory that Six Sigma structure engenders the development of dynamic learning capabilities. However, one might still question whether the positive returns from these capability developments justify risks and opportunity costs associated with Six Sigma adoption. By using matched, presumed non-adopting rms as benchmarks in our analysis, we have constructed proxies that incorporate such risks and opportunity costs. However, a given manager considering Six Sigma adoption will want to consider the specic risks and foregone opportunities that are relevant to his/her rm. For example, the prots projected by our study might not clear the desired rate of return (hurdle rate) for a given rm. Moreover, for a given rm a Six Sigma program might be an inferior investment to a new distribution center, an ERP system, a more fuel efcient eet of trucks, or other such investments if they have more certain or faster paybacks.2 Our results do suggest that positive returns from Six Sigma adoption take time. The most signicant driver of ROA in our data, savings in SG&A, rst materializes signicantly only in years +2 through +4 (see Table 6 Panel C). Similarly, with one exception, signicant improvements in sales growth emerge only in years +3 and +4 (see Table 7). Thus, it appears that dynamic capabilities emerge gradually, or at least take time to be manifested in operating performance improvements. It is also possible that Six Sigma is rolled out sequentially across divisions, extending the time for impacts to manifest in overall corporate performance. Though these ndings should be regarded as tentative, they do suggest that managers should be willing to wait at least 23 years before net impacts of Six Sigma adoption become signicantly positive. Other insights into the nature of Six Sigmas operational impacts provided by our study are also quite interesting, and somewhat surprising. Six Sigma is widely recognized as a methodology for cutting costs and eliminating defects (Byrne and Norris, 2003; Pande et al., 2000). Six Sigma focuses organizational efforts on process improvement, especially through reducing variance for outputs of product (or service) features that are deemed to critically inuence customers perceptions of quality. At its core, the DMAIC method aims to measure and analyze the deviation of a given process from its critical-to-quality goals so that workers can install preventive measures that eliminate the root cause of defects. Such preventive measures involve the implementation of training, procedures, monitoring and control systems, tools, technologies, and product redesign. One would expect that this focus on structural control would yield improvements in process efciencies. Reductions in variation and associated defects are known to create cost savings in areas of internal product rework, inventories, capacity

We thank one of the reviewers for offering this observation.

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

451

buffers, warranties, and repair work. These also include material costs savings from reduced scrap and labor savings from reduced appraisal, material handling, and supervision. We expected that such efciency improvements would be reected in overall lowered product or service costs and associated higher margins. Surprisingly, our results clearly show that efciencies gained from Six Sigma adoption are reected more strongly in indirect cost savings (SG&A), as opposed to savings in direct operating costs (COGS). We note that COGS includes direct purchase, labor, and operating expenses, while SG&A captures indirect expenses associated with governance, logistics, advertising, overhead, and other indirect activities. It is important to note that most SG&A processes in manufacturing rms are in fact, repeatable service processes (e.g., customer service, billing, transportation, etc.). Nakhai and Neves (2009) identify a number of non-manufacturing applications of Six Sigma inside manufacturing rms. Such processes tend to be labor-intensive and repetitive. A related nding in our data is that Six Sigma benets are strongly correlated with labor-intensity in manufacturing rms, yet this same correlation is not signicant in service rms. Taken together, both ndings suggest that laborintensive, repeatable processes offer the greatest opportunities for applications of Six Sigma methods. We offer several explanations for this nding, which at rst glance may seem somewhat counter-intuitive. First, Hendricks and Singhal (2001b) argue that labor-intense rms provide more fertile grounds for quality process improvements because they have more process options and depend more on training and skills. As we explained in our motivation for hypothesis H4, processes that are more automated and less labor-intensive tend to be inherently less variable. As a result, these processes provide less overall opportunity for improvement from Six Sigma structured methods, which are mostly aimed at variance reduction. Second, back-ofce (SG&A) operations tend to be less inuenced by specic customer requirements and idiosyncrasies, i.e., they are more repeatable. Consequently, they may present more attractive targets for Six Sigma projects. Indeed, many of the service rm examples of Six Sigma applications described in the literature are actually in back-ofce or business-to-business contexts (Nakhai and Neves, 2009; Antony et al., 2007). Given the supposed challenges of implementing Six Sigma in highly personalized services, future studies that directly compare Six Sigma implementations in back-ofce versus more direct personal service contexts could reveal important differences in how Six Sigma concepts are operationalized. Finally, there is logic suggesting that process variance reductions will reduce indirect costs, perhaps even more strongly than direct costs. Schmenner (1988, 1991) notes that overhead costs often exceed direct costs. Further, he argues that slow moving and highly variable process ows are the primary drivers of indirect overhead costs. For example, variable process ows create requirements for many transactions (purchasing, inventory control, production control, quality control) as well as other added overheads (inventory, space, material handling, management attention). These arguments are echoed in swift-even-ow theory (Schmenner and Swink, 1998), which states that processes with faster and less variable ows will be more productive and efcient. Consider, for example, how quality improvements from Six Sigma variance reduction efforts would be reected in indirect costs. A rm making these improvements would enjoy lowered safety stocks, lowered quality assurance costs, lowered material handling costs, and so on. These cost savings could conceivably exceed the direct savings attributable to product quality improvements. Again, future research which explores these possible explanations is needed. The foregoing arguments connect Six Sigma efforts to indirect costs, but why would Six Sigma impacts not also be signicantly

manifested in the direct costs (COGS) of our sample rms? We offer two possible explanations. First, in most manufacturing rms the largest component of COGS is the cost of purchased items. It is likely that many Six Sigma projects, particularly in the rst few years of Six Sigma implementation, are inwardly focused, not involving suppliers directly, and thus produce only limited, if any, impacts on purchased item costs. Second, we noted earlier that many adopters of Six Sigma are already mature organizations in terms of quality management. Many of the preceding quality and process improvement approaches may have already been aimed at direct production processes. As a result, the potential additional impacts of Six Sigma project are likely lessened in these areas. Supporting this conjecture, our cross-sectional analyses show that Six Sigma benets are reduced in rms with greater quality maturity. More research is needed to investigate these and other possible explanations of this nding. Two studies, Hendricks and Singhal (1997) and Corbett et al. (2005), demonstrate signicant increases in sales associated with quality awards and ISO 9000 certications, respectively. Similarly, our results indicate that sales improvements from Six Sigma adoptions are at least marginally signicant. Such effects on sales growth may stem from improved product quality and customer satisfaction, or from reputation and image-related halo effects associated with adoption. To the extent that product innovation drives sales growth, our results suggest that Six Sigma adoption at least does no consistent harm to innovation; to the contrary, it may be compatible with or even supportive of growth-oriented innovation. This nding is consistent with the expectations of Schroeder et al. (2008) and Manev and Stevenson (2001), and provides no support for arguments that Six Sigmas narrow and disciplined focus may hamper growth (Brady, 2005; Hindo, 2007; Parast, 2011). On the other hand, the positive effects of Six Sigma on sales growth may be entirely due to improvements in product quality and the rms reputation. Future research efforts should focus on parsing out these different effects. Statistically indistinguishable positive ROA impacts of Six Sigma adoption between manufacturing and services rms suggest that adoption impacts are more a function of the programs overall structural and cultural aspects, and less about the focused statistical tools and techniques that may be more frequently applied in manufacturing contexts (Schroeder et al., 2008; Naor et al., 2008; Gutierrez et al., 2009; Nakhai and Neves, 2009; Antony et al., 2008). The positive benets of Six Sigma appear to also be robust to other contextual factors including R&D intensity, rm size, adoption year, and CEO turnover. However, our results do indicate that Six Sigma adoption impacts may be related to the rms absolute nancial performance level at the time of adoption. Both protable and loss-making service rms appear to gain more from Six Sigma adoption. We hypothesized that protable rms have ample resources to support a broader and more thorough adoption of Six Sigma, while loss-making rms may be better able to overcome inertia and resistance to change, because their managers and employees have a greater sense of urgency to implement required organizational changes. Our results indicate that effects of prior nancial performance can be signicant, but only in service rms. Perhaps the organizational changes required to implement Six Sigma are more far-reaching in services than they are in manufacturing, or at least managers may perceive them as such. This is an interesting nding that deserves further attention.

6. Conclusions, limitations, and future research Our study provides solid support for the hypothesis that Six Sigma adoption tends to produce signicant abnormal benets to rm protability. These benets appear to be persistent over

452

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453

the time horizon we studied, as positive ROA changes were frequently evidenced in latter periods (years +3 and +4). Further, the results indicate that overall protability impacts stem primarily from improved indirect cost efciencies, rather than from direct cost improvements or from improved asset productivities. The results also manifested marginally signicant positive effects on sales growth. These ndings hint at potential differences in how Six Sigma programs are possibly being applied in front-ofce versus back-ofce contexts. In addition, our ndings identify some important contextual factors. In general, our ndings hint that Six Sigma methods may be most benecial when applied to labor-intensive, repeatable processes. However, the results suggest that less labor-intensive, quality experienced, manufacturing rms will not experience the prot impact from Six Sigma adoption that others will. On the other hand, service rms contemplating Six Sigma adoption should pay close attention to how their current performance can be leveraged to maximize Six Sigmas potential. Managers in these rms would do well to identify and exploit either resource-based or motivational advantages that come from positive or negative prior nancial performance, respectively. Our study is limited in ways that can be addressed in future research. First, our sample of service rms is somewhat small and grossly aggregated across service types (e.g., logistics providers and banks). Future research that directly compares Six Sigma implementations in personal and non-personal services might extend our understanding of its applicability limits and particular sources of impact. Second, our sample may be biased toward larger rms. We cannot verify the bias, as it may be that only larger rms are likely to adopt Six Sigma due to the sizable investments required. Moreover, in our analysis, our control for rm size was not signicant, suggesting that size is not a strong driver of Six Sigma impact. However, our limited sample prevents us assessing the effects of rm size for relatively small rms. Given the difculty in obtaining secondary data on small and medium sized enterprises (SMEs), a primary data collection approach might be benecial. Third, our study is limited by its focus on operational performance impacts over a six-year time horizon. As time passes, future studies should be able to study whether Six Sigma effects persist over even longer periods. Moreover, there are a host of other performance effects that should be examined, including the stock markets valuation of future cash ows and intangible assets associated with Six Sigma, as well as impacts on corporate social responsibility and other concerns. Though we investigated effects on sales growth, this serves as a poor proxy for innovation performance. Future studies might also build on our method by comparing patent activity, new product introductions, and other more direct measures of explorative innovation across adopters and non-adopters of Six Sigma. We explored associations of abnormal performance due to Six Sigma with manufacturing and service contexts and with a number of other contextual factors. Others factors may also be important. Several researchers have pointed to important differences in the timing of adoption of such programs (Westphal et al., 1997; Yeung et al., 2006; Benner and Veloso, 2008), thus warranting a study comparing early versus late adopters. Though our results did not relate another size variable (market value) to abnormal performance, yet other size-related factors such as resource slack and inertia may moderate Six Sigma impacts (Hendricks and Singhal, 2001b). In addition, the process management literature points out other potentially important moderators, including diversication (Hendricks and Singhal, 2001b), technology strategy (Ittner and Larcker, 1997; Benner and Veloso, 2008), and technological dynamism (Benner and Tushman, 2002; Benner, 2009; Parast, 2011). Finally, numerous writers have identied implementation

factors associated with an effective Six Sigma program, including motivation, t with culture, and conformance to program structure.

References
Abernathy, W.J., 1978. The Productivity Dilemma. Johns Hopkins University Press, Baltimore. Anand, G., Ward, P.T., Tatikonda, M.V., Schilling, D.A., 2009. Dynamic capabilities through continuous improvement infrastructure. Journal of Operations Management 27, 444461. Antony, J., 2006. Six Sigma for service processes. Business Process Management Journal 12 (2), 234248. Antony, F., Kumar, M., Cho, B., 2007. Six Sigma in service organisations: benets, challenges and difculties, common myths, empirical observations and success factors. International Journal of Quality & Reliability Management 24 (3), 294311. Antony, J., Kumar, M., Labib, A., 2008. Gearing Six Sigma into UK manufacturing SMEs: results from a pilot study. Journal of the Operational Research Society 59, 482493. Barber, B.M., Lyon, J.D., 1996. Detecting abnormal operating performance: the empirical power and specication of test statistics. Journal of Financial Economics 41, 359399. Barney, M., 2002. Macro, meso, micro: Six Sigma. The Industrial Organizational Psychologist 39 (4), 104107. Benner, M.J., 2009. Dynamic or static capabilities? Process management practices and response to technological change. Journal of Production Innovation Management 26, 473486. Benner, M.J., Tushman, M., 2002. Process management and technological innovation: A longitudinal study of the photography and paint industries. Administrative Science Quarterly 47, 676706. Benner, M.J., Tushman, M.L., 2003. Exploitation, exploration, and process management: The productivity dilemma revisited. Academy of Management Review 28 (2), 238256. Benner, M.J., Veloso, F.M., 2008. ISO 9000 practices and nancial performance: a technology coherence perspective. Journal of Operations Management 26, 611629. Blakeslee Jr., J.A., 1999. Implementing the Six Sigma solution. Quality Progress 32 (7), 7785. Brady, D., 2005. Bringing innovation to the home of Six Sigma. Business Week (August), 68. Braunscheidel, M.J., Hamister, J.W., Suresh, N.C., Star, H., 2011. An institutional theory perspective on Six Sigma adoption. International Journal of Operations & Production Management 31 (4), 423451. Brewer, P.C., 2004. Six Sigma helps a company create a culture of accountability. Journal of Organizational Excellence 23 (3), 4559. Breyfogle, F.W., 2003. Implementing Six Sigma: Smarter Solutions Using Statistical Methods. John Wiley, Hoboken, NJ. Byrne, G., Norris, B., 2003. Drive Baldrige level performance. Six Sigma Forum Magazine 2 (3), 1321. Carnell, M., 2003. Gathering customer feedback. Quality Progress 36 (1), 6061. Chakravorty, S.S., 2009. Six Sigma programs: an implementation model. International Journal of Production Economics 119 (1), 116. Chakravorty, S.S., 2010. Where process improvement projects go wrong. The Wall Street Journal (January), R6. Choo, A.S., Linderman, K.W., Schroeder, R.G., 2007. Method and psychological effects on learning behaviors and knowledge creation in quality improvement projects. Management Science 53 (3), 437450. Clark, K.B., Fujimoto, T., 1991. Product Development Performance. Harvard Business School Press, Boston. Clifford, L., 2001. Why you can safely ignore Six Sigma. Fortune 143 (2), 140. Cohen, W.M., Levinthal, D.A., 1990. Absorptive capacity: a new perspective on learning and innovation. Administrative Science Quarterly 35 (1), 128152. Corbett, C., Montes-Sancho, M., Kirsch, D., 2005. The nancial impact of ISO 9000 certication in the U.S.: an empirical analysis. Management Science 51 (7), 10461059. Detert, J.R., Schroeder, R.G., Mauriel, J.J., 2000. A framework for linking culture and improvement initiatives in organizations. Academy of Management Review 25 (4), 850863. Fahmy, D., 2006. When one black belt is enough. Treasury and Risk Management 16 (3), 5559. Goh, T.N., Low, P.C., Tsui, K.L., Xie, M., 2003. Impact of Six Sigma implementation on stock price performance. TQM & Business Excellence 14 (7), 753763. Gowen, C.R., Tallon, W.J., 2005. Effect of technological intensity on the relationships among Six Sigma design, electronic-business, and competitive advantage: a dynamic capabilities model study. The Journal of High Technology Management Research 16, 5987. Gutierrez, L.J.G., Llorens-Montes, F.J., Sanchez, O.F.B., 2009. Six Sigma: from a goaltheoretic perspective to shared-vision development. International Journal of Operations & Production Management 29 (2), 151169. Hahn, G.J., Hill, W.J., Hoeri, R.W., Zinkgraf, S.A., 1999. The impact of Six Sigma improvementa glimpse into the future of statistics. The American Statistician 53 (3), 208215. Harry, M., Schroeder, R., 2000. Six Sigma: The Breakthrough Management Strategy Revolutionizing The Worlds Top Corporations. Doubleday, New York.

M. Swink, B.W. Jacobs / Journal of Operations Management 30 (2012) 437453 Henderson, K.M., Evans, J.R., 2000. Successful implementation of Six Sigma: benchmarking General Electric Company. Benchmarking: An International Journal 7 (4), 260281. Hendricks, K.B., Singhal, V.R., 1996. Quality awards and the market value of the rm: an empirical investigation. Management Science 42 (3), 415436. Hendricks, K.B., Singhal, V.R., 1997. Does implementing an effective TQM program actually improve operating performance? Empirical evidence from rms that have won quality awards. Management Science 43 (9), 12581274. Hendricks, K.B., Singhal, V.R., 2001a. The long-run stock price performance of rms with Effective TQM programs. Management Science 47 (3), 359368. Hendricks, K.B., Singhal, V.R., 2001b. Firm characteristics, Total Quality Management, and nancial performance. Journal of Operations Management 19, 269285. Hendricks, K.B., Singhal, V.R., 2008. The effect of product introduction delays on operating performance. Management Science 54 (5), 878892. Hensley, R.L., Dobie, K., 2005. Assessing readiness for Six Sigma in a service setting. Managing Service Quality 15 (1), 82101. Hindo, B., 2007. At 3M, a struggle between efciency and creativity. Business Week (August), 814. Ittner, C.D., Larcker, D., 1997. The performance effects of process management techniques. Management Science 43, 522534. Ittner, C.D., Nagar, V., Rajan, M.V., 2001. An empirical examination of dynamic quality-based learning models. Management Science 47 (4), 563578. Kinney, M.R., Wempe, W.F., 2002. Further evidence on the extent and origins of JITs protability effects. The Accounting Review 77 (1), 203225. Kotter, J.P., 1995. Leading change: why transformation efforts fail. Harvard Business Review 73 (2), 5967. Kull, T.J., Narasimhan, R., 2010. Quality management and cooperative values: investigation of multilevel inuences on workgroup performance. Decision Sciences 41 (1), 81113. Kumar, M., Antony, J., 2009. Does size matter for Six Sigma implementation? Findings from the survey in UK SMEs. The TQM Journal 21 (6), 623635. Levine, D.I., Toffel, M.W., 2010. Quality management and job quality: how the ISO 9001 standard for quality management systems affects employees and employers. Management Science 56 (6), 978996. Linderman, K., Schroeder, R.G., Zaheer, S., Choo, A.S., 2003. Six Sigma: a goal-theoretic perspective. Journal of Operations Management 21 (2), 193203. Linderman, K., Schroeder, R.G., Choo, A.S., 2006. Six Sigma: the role of goals in improvement teams. Journal of Operations Management 24 (6), 779790. Manev, I.M., Stevenson, W.B., 2001. Balancing ties: boundary spanning and inuence in the organizations extended network of communication. Journal of Business Communication 38 (2), 183205. Martinez-Costa, M., Choi, T.Y., Martinez, J.A., Martinez-Lorente, A.R., 2009. ISO 9000/1994, ISO 9001/2000 and TQM: the performance debate revisited. Journal of Operations Management 27, 495511. Molina, L.M., Llorens-Montes, J., Ruiz-Moreno, A., 2007. Relationship between quality management practices and knowledge transfer. Journal of Operations Management 25, 682701. Morris, P.W., 2006. ISO 9000 and nancial performance in the electronics industry. Journal of American Academy of Business 8 (2), 227235. Nair, A., 2006. Meta-analysis of the relationship between quality management practices and rm performance-implications for quality management theory development. Journal of Operations Management 24 (6), 948975. Nair, A., Malhotra, J.K., Ahire, S.L., 2011. Toward a theory of managing context in Six Sigma process-improvement projects: an action research investigation. Journal of Operations Management 29, 529548.

453

Nakhai, B., Neves, J.S., 2009. The challenges of Six Sigma in improving service quality. International Journal of Quality & Reliability Management 26 (7), 663684. Naor, M., Goldstein, S.M., Linderman, K.W., Schroeder, R.G., 2008. The role of culture as driver of quality management and performance: infrastructure versus core quality practices. Decision Sciences 39 (4), 671702. Naveh, E., Erez, M., 2004. Innovation and attention to detail in the quality improvement paradigm. Management Science 50 (11), 15761586. Pande, P.S., Neuman, R.P., Cavanagh, R.R., 2000. The Six Sigma Way: How GE, Motorola and Other Top Companies are Honing their Performance. McGraw-Hill, New York. Parast, M.M., 2011. The effect of Six Sigma projects on innovation and rm performance. International Journal of Project Management 29, 4555. Powell, T.C., 1995. Total Quality Management as competitive advantage: a review and empirical study. Strategic Management Journal 16, 1537. Rowlands, H., 2003. Six Sigma: a new philosophy or repacking of old ideas? Engineering Management (April), 1821. Schmenner, R.G., 1988. Escaping the black holes of cost accounting. Business Horizons 31 (1), 6672. Schmenner, R.G., 1991. International factory productivity gains. Journal of Operations Management 10 (2), 229254. Schmenner, R.G., Swink, M.L., 1998. On theory in operations management. Journal of Operations Management 17 (1), 97113. Schroeder, R.G., Linderman, K., Liedtke, C., Choo, A.S., 2008. Six Sigma: denition and underlying theory. Journal of Operations Management 26, 536554. Shewhart, W.A., 1931. Economic Control of Quality of Manufactured Product. D. Van Nostrand, New York. Sila, I., 2007. Examining the effects of contextual factors on TQM and performance through the lens of organizational theories: an empirical study. Journal of Operations Management 25, 83109. Sinha, K.K., Van de Ven, A.H., 2005. Designing work within and between organizations. Organization Science 16 (4), 389408. Snee, R.D., Hoerl, R.W., 2003. Leading Six Sigma. Prentice Hall, Englewood Cliffs, NJ. Teece, D.J., 2007. Explicating dynamic capabilities: the nature and microfoundations of (sustainable) enterprise performance. Strategic Management Journal 38 (13), 13191350. Teece, D.J., Pisano, G., Shuen, A., 1997. Dynamic capabilities and strategic management. Strategic Management Journal 18 (7), 509533. Tushman, M.L., OReilly III, C.A., 1996. Ambidextrous organizations: managing evolutionary and revolutionary change. California Management Review 38 (4), 830. Westphal, J.D., Gulati, R., Shortell, S.M., 1997. Customization or conformity? An institutional and network perspective on the content and consequences of TQM adoption. Administrative Science Quarterly 42, 366394. Yeung, A.C.L., Cheng, T.C.E., Lai, K.H., 2006. An operational and institutional perspective on Total quality management. Production and Operations Management 15 (1), 156170. York, K.M., Miree, C.E., 2004. Causation or covariation: an empirical re-examination of the link between TQM and nancial performance. Journal of Operations Management 22, 291311. Zahra, S.A., George, G., 2002. Absorptive capacity: a review, reconceptualization, and extension. Academy of Management Review 27 (2), 185203. Zollo, M., Winter, S.G., 2002. Deliberate learning and the evolution of dynamic capabilities. Organization Science 13 (3), 339351. Zu, X., Fredendall, L.D., Douglas, T.J., 2008. The evolving theory of quality management: the role of Six Sigma. Journal of Operations Management 26, 630650.