Performance effect of Lean:
a complete implementation matters
Thomas Borup Kristensen & Poul Israelsen
Abstract
To understand how the practices of Lean affect performance, we tested and validated a system-wide approach using mediating relationships in a structural equation model.
We used a cross-sectional survey of 200 Danish companies that indicated that they used Lean. Thus, this study is especially relevant to Denmark, but the approach is em- pirically more generalizable. We show that the effect of Lean standardized flow pro- duction practices on performance is mediated by analytical continuous improvement empowerment practices and by delegation of decision rights practices. Thus, standard- ized flow practices do not have direct effects on performance. Instead, standardized flow provided that foundation for companies to implement continuous improvement, which, in turn, directly affect performance positively. Another cause, in addition to flow practices, of continuous improvement was the delegation of decision rights. The paper provides evidence that supports the view that middle managers’ actions further enhance performance in Lean companies. The right Lean behavior by middle manag- ers increases the level of analytical continuous improvement empowerment. In total, high-performing Lean companies implement a complete package of Lean practices.
1. Introduction
Since Womack et al. (1991) published their seminal book on the Toyota production system, many companies around the world, including some in Denmark, have adopted the methods in question under the label “Lean.” Lean is a dominant management phi- losophy companies use to achieve sustainable profits (Fullerton et al 2014). Therefore, whether the promising words of Lean actually help enhance a company’s performance should be investigated.
Liker (2004) warned against perceiving Lean systems as piecemeal technical projects.
This will provide the organization with an incentive to pick low-hanging fruit, thus missing the chance to acquire a long-term sustainable system. In another central study of Toyota that examined Lean as a package, Kobayashi (1995) argued that the Lean sys- tem could work only if all the pillars of the system function. In the present study, we
address the importance of relations between Lean bundles of tools. These relations con- stitute a package approach to Lean, as we are interested in the holistic picture of multi- ple relations between multiple Lean practices. We are able to identify how Lean affects performance based upon multiple relations between standardized flow, decentraliza- tion, analytical continuous improvement, and Lean behavior of middle management.
In this study, we investigated how elements of Lean are implemented to enhance performance. The elements are standardized flow production, decentralization, continuous improvement, the role of middle management, and performance effects.
Standardized flow production is a prerequisite for increased decentralization and con- tinuous improvement in Lean companies. These two elements enhance performance.
We include the behavior of middle management as this is another important cause of increased continuous improvement. Moreover, decentralization increases continuous improvement, as the latter cannot function without a partially self-controlled group that has the authority to continuously experiment with potential improvements.
To provide empirical evidence of these multiple relations between these elements of Lean implementation and performance, we used survey data from 200 Danish com- panies from 2008. We tested all the relations in a structural equation model. The relations were empirically significant, and the overall model fit of the full model was acceptable. Therefore, we concluded that these elements should not be perceived in isolation but in a holistic way. Therefore, a full implementation matters in terms of enhanced performance.
This study contributes to the literature on Lean in several ways. First, we show that continuous improvement is a primary driver of performance. Second, continuous improvement is supported by decentralization and standardized flow production. The latter relationship is important because continuous improvement cannot function properly without standards as a baseline for comparison and because standardized flow production provides transparency on the shop floor where potential improve- ment may be harvested. Third, the role of middle managers has not been previously explored, and they constitute an important cause of continuous improvements as these managers facilitate appropriate Lean behavior. Fourth, to our knowledge, this is the first Scandinavian, including Danish, study to present large-scale empirical evi- dence that Lean functions as a package of multiple elements. The previous empirical evidence on Lean performance in Scandinavia and was from single a company as in Kristensen and Israelsen (2014).
The managerial implication of this paper is evidence-based takeaways. Managers may be informed that Lean implementation can enhance performance. However, perfor- mance is enhanced based on a full implementation of Lean. If managers omit some of
the elements of Lean in their implementation, they will not harvest the full potential performance effects of Lean. Therefore, managers should think carefully about a com- plete Lean implementation; a scattered approach is less effective. Thus, the decision to implement Lean should be strategic and long term, as these multiple elements of Lean take many years to implement well.
The remainder of the paper is structured as follows. In Section 2, we describe the literature review and the hypothesis development. In Section 3, we introduce the re- search method, and in Section 4, we present the analysis. In Section 5, we present the conclusion and limitations.
2. The performance effect of Lean practices
Several studies have explored the performance effects of various Lean practices (includ- ing just-in-time [JIT] and world-class manufacturing). Hofer et al. (2012) presented a list of previous studies that concluded Lean led to mixed financial results. These studies included a work by Huson and Nandag (1995). They discovered that JIT manufactur- ing, compared to non-JIT manufacturing, had significant effects on some performance measures but not on all. Although the unit cost rose and the operating margin per sales dollar declined, earnings and earnings per share still improved due to a mix of revenue increase and interest reduction caused by a decrease in working capital. A decrease in inventory was also found by Balakrishnan et al. (1996). The latter study, however, found no performance effect linked to JIT adoption when measured in terms of return on as- sets as the dependent variable and compared to that of firms in the control group. Both studies measured JIT adoption as a binary item, as being either present or not present.
In contrast, Fullerton and McWatters (2002) measured Lean in terms of degree of imple- mentation, not just as a binary variable, as omitting this variable may have caused the mixed results in Balakrishnan et al. (1996) and Fullerton and McWatters (2002). Kinney and Wempe (2002) showed that adopting JIT (again measured only as a binary variable) had a positive effect on profit margins. However, the demonstrated effect was unsus- tainable, as the control firms leveled out the difference within 5 to 6 years.
Other studies employed a more detailed approach to measuring the construct of JIT/
Lean in order to gain a better understanding of the complexity of these manufactur- ing strategies. A case-based survey, carried out by Young and Selto (1993) of one large company, revealed Lean had no performance effect on six different items that meas- ured performance.
Fullerton and McWatters (2001) used a combined JIT construct that consisted of the average score of three JIT factors: JIT manufacturing, JIT quality, and JIT unique, where the latter expresses Kanban and JIT purchasing. The authors found significant
performance effects, which emphasize the need to address JIT, or Lean, as a multidi- mensional phenomenon.
Shah and Ward (2003) utilized the concept of complementarities (their label) between Lean practices. The authors categorized practices in terms of four bundles: total qual- ity management (TQM), total productive maintenance (TPM), JIT, and human resource management (HRM). Each bundle had a performance effect on unit costs and non-fi- nancial measures. The approach is compelling. Therefore, we used bundles of prac- tices in our approach to understanding how Lean practices contribute to performance.
Shah and Ward did not examine mediator effects between the bundles. Therefore, the Shah and Ward (2003) study seems to have supported the additive effect of practices instead of the dependencies (system) between the bundles of practices on which we focus. Looking more closely into the study by Fullerton and McWatters (2001), it failed to focus on the importance of including the HRM bundle elements (self-directed teams and the delegation process). Unfortunately, they focused narrowly on the tan- gible elements of the Lean package, the elements we categorize as standardized flow production in our model. Moreover, Shah and Ward (2007) recognized the need to understand enhanced performance effects from Lean as an effect where the practices are mutually dependent.
Hofer et al. (2012), Furlan et al. (2011), and Bortolotti et al. (2014) also found positive effects using “a bundle approach.” Bortolotti et al. (2014) utilized a model that arranged the bundles as independent variables or mediator variables. Meaning, the authors real- ized that bundles of practices are related. This is not done by Shah and Ward (2003).
We also used this approach in which bundles are related (they depend on each other) to understand the complexity of Lean.
In sum, we employ two important learnings from this literature review. Lean should be treated as bundles of practices and should be understood system-wide with inter- dependencies between the bundles. Lean consists of a package of practices, such as principles for governing a company by delegating decision making, the use of stand- ardized flow techniques in the processes, and analytical tools to sustain continuous improvement. In the present study, we included 12 practices for measuring the Lean package of principles; see Table 1. These practices constitute the main Lean prac- tices. Table 1 presents an overview of the 12 practices and how they are grouped into constructs.
Table 1. The 12 Lean practices
3 Main constructs 12 Practices
Standardized flow production (physical operations)
U-cell Production Kanban 5 S SMED TPM
Analytical continuous improvement (information empowerment)
Kaizen
Whiteboard Meetings Value Stream Mapping (VSM) Six Sigma
Delegation of decisions rights
(decision empowerment) Partially Self-controlled Teams Policy Deployment
Co-influence on tasks
The grouping into the constructs in Table 1 was inspired by Shah and Ward (2003) grouping into bundles and by Furlan et al. (2011). However, we disentangled Shah and Ward’s HRM bundle into two constructs (delegation of decision rights and continuous improvement founded in information empowerment) to better reflect the management perspectives in Lean and focus less on the technical aspects, which is in line with Liker (2004). Moreover, Furlan et al. (2011) represented the bundles as binary vari- ables, which we consider reductionist, and they also presented this as a limitation of their study.
2.1. Standardized flow production is a prerequisite for delegation and for continuous im- provement
Standardized flow production consists of multiple practices that enforce standardiza- tion and flow, as shown in Table 1. U-cell production consists of the basic shop floor layout of the Lean organization. This means that the workstation, machinery, or labor set-up is shaped like the letter U (Liker 2004). U-cell production requires more than just having a U-shaped shop floor layout; it also requires identification of product families with almost the same production flow, thus, standardized flow (Bicheno and Holweg 2009; Kennedy and Widener 2008). Standardized flow from U-cells is fur- ther supported by 5S. The objective of 5S-cells is to keep the production cell clean and organized. The most important technique in 5S is the standardization practice known as “standard work” (Bicheno and Holweg 2009). Standardized flow provides the foundation for continuous improvements (Kaizen etc.), as the standards constitute the baseline against which one can compare improvements (Liker 2004). Standardized work also stabilizes the process, thus providing the opportunity to easily recognize
and correct abnormalities and locate improvements. Therefore, 5S is one of the pillars of continuous improvements, which is reflected as hypothesis 1b below.
Standards are also important in terms of aligning actions with company goals (Ken- nedy and Widener 2008), which is reflected in hypothesis 1a, as standardization thus provides the foundation for delegation, but only partly self-controlled delegation. In addition, in the U-cell production setup, the cell members function as a team. This team structure provides the basis for delegating decision rights, as there are formal teams to actually delegate to, which is represented in hypothesis 1a.
Standardized flow is reinforced by Kanban, Single-Minute Exchange of Die (SMED), and TPM. Kanban (Bicheno and Holweg 2009) is done by standardization via a visual inventory pull system (Liker 2004), and this helps reveal the improvement potential of processes that do not actually follow standard time. This is the basis for hypothesis 1b. SMED is a practice that may be included when creating an organization that can deliver flow production. SMED consists of the systematic work involved in reducing changeover times (standards for changeovers), which, in turn, is an important ele- ment in creating flow alongside small inventories (Bicheno and Holweg 2009). TPM is the last practice in the group of practices we label “flow production.” Total productive maintenance is about avoiding the breakdown of resources in order to create stabil- ity in machine availability by standardizing maintenance (Bicheno and Holweg 2009).
These other types of standardized flow also create the foundation for continuous improvements as argued for standardization (hypothesis 1b) and the foundation for partial delegation of the decision (hypothesis 1a). We proposed the following:
Hypothesis 1a: Flow production is positively related to partial delegation of deci- sion rights.
Hypothesis 1b: Flow production is positively related to continuous improvement based on analytical information empowerment.
The direct relation between flow production and performance was tested to verify that this should not be significant when mediator variables (analytical continuous im- provement and delegation of decision rights) are included. This non-relation between flow production and performance shows that Lean is a system-wide practice package based on full mediation as opposed to partial mediation. This non-relation is the nov- elty of this paper. We label this mediation relation Hypothesis 1c.
2.2. Delegation of decision rights is needed to create continuous improvement
As Table 1 shows, the construct analytical continuous improvement consists of four practices. Value stream mapping (VSM) is a core analytical tool of the Lean package of practices. The purpose of VSM is to map all processes and categorize them as value- added or non-value-added activities and to base continuous improvement decisions on this categorization (cf. Rother and Shook 1999). The resulting map, or VSM, shows how the production units flow through the processes, while providing additional in- formation regarding lead times, stock turns, changeover times, and cycle times.
Six Sigma was developed, or employed under the name of Six Sigma, by Toyota.
Toyota built many of the Six Sigma elements into the company’s quality improve- ment programs; therefore, Six Sigma may still be considered part of the Lean program (Bicheno and Holweg 2009). Six Sigma is a practice that aims to measure variability within processes by focusing on quality issues such as scrapped parts per million.
Kaizen is a practice that aims to ensure sustainable continuous improvements (Liker 2004), which are often achieved through events that take place on the shop floor. The current performance measures are analyzed, and plans are made to define the objec- tives of the event in question. The event is then carried out, and its results are meas- ured. This process is repeated numerous times to ensure that the results are sustain- able (Bicheno and Holweg 2009). In short, Kaizen is about providing employees with an analytical tool for improving operations and thus remove waste (Liker 2004). The objective is to enhance current operations by measuring an area of interest to find avenues of improvement, carry them out, and sustain them systematically. Although Kaizen is dedicated to helping employees enhance their knowledge, the practice also entails a standardized procedure that ensures this knowledge becomes incorporated in daily work. Whiteboard meetings are somewhat similar to Kaizen events, as these meetings aim to ensure that all employees follow up on the production measures and take corrective actions. Whiteboard meetings are typically conducted once a week by shop floor employees and more frequently by foremen and middle managers (Car- reira 2005). These meetings usually take place in the middle of the production floor and consist of a few standard measures of how well operations are running. This way, the meetings almost constitute a real-time follow-up on the production problems and deviations that crop up periodically. Employees are expected to engage actively with the figures and to present ideas for improvement based upon them.
As hypothesis 2 states below, continuous improvement is conducted on the shop floor and in teams only if decision rights are delegated to this level. If the teams have no delegated decisions rights, then improvements are not possible as the actors have no room within which to act.
The delegation of decision rights construct consists of three practices. Essential to the Lean package approach is the involvement of every worker within the given organiza- tion (Lind 2001) not in self-controlled groups, but in groups that have access to the help of their foremen to overcome problems using standards and methods set by sen- ior management. Decentralized decisions are dominated by a group agenda approach known as the enabling formalization (Ahrens and Chapman 2004), a key element of Lean (Liker 2004). The primary Lean organizational structure consists of teams. This structure is necessary due to the high degree of interdependency in JIT organizations where there are very low stock levels to buffer problems that arise from a lack of co- ordination or an isolated focus of responsibility. Each team must make decisions on a wide range of issues, with the help of the principles, practices, and philosophy formu- lated by senior management. Similarly, the team employs the company’s chosen tools to implement improvements (Liker 2004). In other words, the teams possess a large degree of decision-making authority while at the same time being expected to follow company guidelines. Thus, the teams are labeled as partially self-controlled teams. The teams must apply the basic methods used by the company (Huntzinger 2007).
Policy deployment is a practice in which consensus within the organization is built up. The key idea here is that it is better to reach less optimal solutions with consensus than to have perfectly optimal solutions imposed by a technocratic expert regime. The objective is to inculcate commitment to company goals among all employees by com- municating a set of common goals. This is mainly achieved with the help of very short reports in A3 paper format and by participation in the organization. All stakeholders in a decision are involved through participation, and those who have to implement the decision also have a strong influence on the design of the plan. It is not just a plan of the outcomes but also of the means, and they have to be aligned with the rest of the organization (Bicheno and Holweg 2009). Policy deployment involves a long process in order to bring about consensus; however, decisions can subsequently be implemented rapidly. Thus, we proposed the following:
Hypothesis 2: The delegation of decision rights is positively related to continuous improvement based on analytical information empowerment.
2.3. Middle managers’ behavior affects continuous improvement and performance Ouchi and McGuire (1975) concluded that personal controls are effective when the processes (means) are well-known. In the Lean organization, the means are well- known, and a certain behavior is expected, i.e., that described by Lean practices.
Furthermore, middle and team managers are expected to spend a lot of time on the shop floor in order to detect problems and to rectify them immediately in cooperation with the team members (Lind 2001). The implementation of Lean practices takes place where value is created, called “Gemba” in Toyota terminology (Bicheno and Holweg
2009). Therefore, leaders control whether employees carry out the correct Lean behav- ior. Additionally, leaders train on the Lean practices themselves, thus acting as behav- ioral role models on the shop floor. This “Lean management” feeds into the system of the other Lean practices, and the performance effect from this Lean management is mediated by the analytical continuous improvement empowerment variables. Middle managers ensure that the shop floor workers are working according to “right” Lean behavior and thus deploy Lean practices correctly. Furthermore, if middle managers are committed and encouraged to do more Lean, for example, by wanting to train employees in Lean, be Lean role models, and follow up on employees’ Lean behavior.
Then they also encourage the organization to increase the use of analytical continuous improvement. This leads to the following hypothesis:
Hypothesis 3a: Middle managers’ Lean behavior is positively related to continuous improvement based on analytical information empowerment.
Hypothesis 3b: Middle managers’ Lean behavior is positively related to performance.
2.4. Delegation of decision rights and continuous improvement are positively related to performance
Milgrom and Roberts’ (1995) study supports our hypotheses. The authors stated that JIT companies are required to give greater autonomy to workers (we represent this with our construct delegation of decision rights), which requires good use of local information and process improvements (we represent this with our construct con- tinuous improvement empowerment). Therefore, our set of hypotheses reflects that standardized flow production creates the foundation for the delegation of decision rights and continuous improvement empowerment. This is done by creating standard- ized production. Implementing information systems (in continuous improvements empower) is less costly, as they are based on standards. Moreover, these information systems and standards are a foundation for the teams (delegation of decisions rights) to approach process improvements in the right way. Flow production, however, affects performance only if it leads to better decisions (measured by the mediation) about process improvements by surfacing problems (with low inventories not working as buffer) and by implementing standards to facilitate improvement decisions.
In general, maintaining, sustaining, and benefitting from practices such as U-cells, Kanban, SMED, TPM, and 5S requires analytical continuous improvement, as these practices are not just one-off consulting implementations. These practices are continu- ously adjusted and calibrated to fit the condition of the customers and the environ- ment (which is often dynamic for Lean companies) using analytical information, and this continuous process should preferably be decentralized to allow quick responses
with local knowledge. Moreover, updating standards in 5S does not improve perfor- mance significantly unless the standards are continuously under pressure for sustain- able reduction, with analytically derived solutions from decentralized decision mak- ers. Johnson (1992) summed the situation up by describing business performance as best achieved by being responsive and flexible, and this requires the use of real-time problem-solving to control lead times, variation, and customer satisfaction, which in turn requires training workers in self-management. Fullerton et al. (2014) confirmed this mechanism and stated that performance can be enhanced if workers have the right information. Thus, we proposed the following hypothesis:
Hypothesis 4a: Delegation of decision rights is positively related to performance.
Hypothesis 4b: Continuous improvement based on analytical information empower- ment is positively related to performance.
To control for performance effects caused only by the size of the companies, we in- cluded a control variable, which is the number of employees. We wanted to control for any effects from size such as economies of scale.
Figure 1 illustrated the hypotheses collected in one system. They should be perceived as a system-wide approach where each component marginally contributes to perfor- mance direct or indirectly, and their interdependent relations are presented.
3. Methodology
3.1. Survey design and sample
The survey was completed with the help of Danish Industry (DI) in 2008. DI is a private organization that works in the interest of more than 10,000 Danish member companies. The organization provides guidance through consulting and courses to enhance productivity.
The main advantage of using the DI membership database in mailing out the ques- tionnaire is that we ensured that the respondents were familiar with the topics of the questionnaire by using their direct email address. The questionnaire was distributed to 1,517 members, of whom 459 responded to the survey. This is a 30% response rate, which is higher compared to a recent study on Lean performance, where the response was 8.6% (Hofer et al. 2012). The questionnaire and data collection were performed by researchers at DI, not the authors.
We used only 200 of the 459 responses for the analysis. We selected companies that indicated that they used Lean and have reported their performance,1 which had a posi- tive effect on the validity of the survey answers. The respondents are representatives
of companies that work to some degree with Lean, and therefore, they are knowledge- able about the Lean practices that they were asked about in the questionnaire.
Table 2 provides descriptive statistics for the respondents’ positions in their organi- zations and presents the distribution among industries for the 200 companies that used Lean and for the total 459 companies. The “other” category consisted almost exclusively of production companies. Therefore, the companies mainly represented traditional production companies, except perhaps for construction and healthcare.
To ensure that the construction and healthcare companies did not differ significantly from the other companies, we performed a comparison test. For all questions, we compared the 25 healthcare and construction companies’ means with those of the 175 other companies on t-tests and non-parametric tests (Mann-Whitney U) and found no statistically significant differences. Furthermore, there was no statistically significant difference in the means for the performance questions when we compared the differ- ent types of positions (chief executive officer, department leader, production director, etc.) of the respondents.
Table 2. Background of the respondents
200 selected Lean cases 459 all cases Industry:
Iron & Metal 80 178
Electronics 20 41
Graphical 6 10
Construction 13 26
Healthcare 12 23
Biotech 2 4
IT 1 4
Business services 0 1
Other, Production 66 172
Position:
CEO, VP 34 74
Department director 59 107
COO 87 206
Production manager 7 25
Other 13 47
3.2. Statistical technique, non-response test, and sample size
Figure 1 and the related hypotheses suggest a structural equation model (SEM) ap- proach as the statistical technique. Multiple paths and mediations among constructs
can be tested using structural equation modeling (Kline 2005) instead of using mul- tiple linear regression models to test each path before and after a mediator variable.
Moreover, the structural equation model includes observed manifest variables and latent variables in the same test run (Kline 2005). Therefore, with the SEM approach, we get an overall assessment of the fitness of the data to the ex-ante model and an as- sessment of the significance and direction for each path in the model. Thus, the SEM approach is attractive when addressing Lean as a system-wide model as shown in Figure 1.
Figure 1. Hypotheses
Flow
Delegaon
Cont. Improv empow
No. Of Employees (Size)
Performance
Management
1b
1a 2
4a
4b
3a 3b
A sample size of 200 answers is acceptable in SEM (Kline 2005). To further ensure that the sample size was adequate compared to the number of free parameters to be estimated, we used the CFI and the TLI as the fit indices in the test section, as they penalize overly complex models. We also tested parsimony adjusted fit indices, the PRATIO, the PNFI, and the PCFI, on our model. They were above the threshold value of 0.6 (Blunch 2008). The number of free parameters to be estimated in relation to the sample size was well above the minimum ratio recommended of 1:5 (Worthington and Whittaker 2006), and above the average of 7.4 found in leading operations manage- ment literature (Shah and Goldstein 2006).
To test whether the dataset was subject to non-response bias, a t-test compared the first 10% of answers with the last 10% as recommended by Amstrong and Overton (1977). The result was statistically non-significant. Thus, we find no sign of non-re- sponse bias.
3.3. Survey constructs and variables
Performance was measured as a latent variable using four measures as indicators. The first was the measure of cost per product unit reduction over the previous fiscal year.
The second measure was the reduction of worker time consumption per unit within the previous fiscal year. The third measure was the lead time reduction within the previous fiscal year, and the fourth measure was the reduction of materials per unit.
Each measure was coherent with the measures that Lean companies are supposed to improve. They were measured on a 6-point reversed scale, where 6 represents the lowest (poorest) performance (deterioration of performance), 5 represents unchanged performance;, 4 represents a reduction of 0–5%, 3 represents a reduction between 5%
and 10%, 2 represents a reduction of 10–25%, and 1 represents a reduction of more than 25%.
The four measures were chosen in keeping with Fullerton and McWatters’ (2001) find- ing of a JIT performance effect on reduced scrap and rework. This was subsequently reflected in our cost per unit, time consumption, and materials per unit measures. The effect of reduced inventory and queue time in production (Fullerton and McWatters, 2001) was reflected in the lead time reduction. Shah and Ward (2003) used the manu- facturing cost per unit and customer lead time in their study, where they also found an effect on scrap and rework.
As expected a priori, factor analysis of the four performance items revealed only one factor (i.e., an eigenvalue greater than 1). The Cronbach alpha for the performance was .833, which is more than the 0.5–0.6 recommended by Nunnally (1978) for explora- tive research. Table 3 presents the factor analysis of the performance items. Moreover, Pearson correlations between the three measures revealed that they were significantly positively correlated.
Table 3. Factor analysis: performance (n = 200)
Component Matrix
Component 1 Cost per unit
Time per unit Lead time Materials per unit
.740 .745 .567 .643
The questionnaire collected information on the companies’ use of 12 Lean practices shown in Table 1. These Lean practices constituted the 12 items on the questionnaire, with binary answers for each of the items.2 A measure for each of the three constructs (flow, analytical continuous improvement, and delegation of decision rights) was generated by summing the number of practices a company used within each con- struct/group, which led to three manifest variables. For example, if a company used two of the five flow practices, the score was 2 for the company’s flow variable. Thus, the three variables measured the degree of implementation for these Lean elements.
Although most variables used in structural equation modeling are latent variables, it is also acceptable to use such manifest variables according to Kline (2005). Moreover, to reduce the number of free parameters, 12 Lean practices, to be estimated it is suit- able to treat them as manifest variables. This approach was recommended by Van der Stede (2000) when the sample size is small in an SEM context.
The last main construct concerned the behavior of middle managers regarding Lean.
This construct consisted of the three questions shown in Table 4. These questions were specifically developed for this questionnaire. To assess these questions, a factor analysis was conducted. Here, Cronbach’s alpha was .909 and, thus, was higher than the recommended levels.
Table 4. Factor analysis – middle management behavior (n = 200) Component Matrix
Component 1 The daily managers are good role models for their
subordinates .890
The daily managers are good at being on the shop floor and
training their subordinates in Lean behavior .944
The daily managers are good at following up on whether or
not their subordinates are showing Lean behavior .929
4. Research results
In this section, we present the statistical test results. First, we show the fitness of the structural equation model, and then we assess the path coefficients to indicate whether the hypotheses were supported or not.
4.1. Fitness of the model
The model in Figure 1 was tested using SPSS AMOS 22.0 with the choice of maximum likelihood. We used multiple fit indexes as recommended by Kline (2005). As shown in Table 5, the goodness-of-fit statistics indicated acceptable fit to the data. The Chi square ratio (divided by degrees of freedom) was less than two, indicating an accept-
able fit (Bollen 1989; Kline 2005). The other fit indices (IFI, TLI, and CFI) exceeded the acceptable level of .90, and the RMSEA did not exceed the acceptable fit measure of .08 (Browne and Cudeck 1993). The AIC measure was lower for the default model com- pared to the saturated model, indicating the parsimony of the model (Kline 2005).
Table 5. Fitness of SEM
Fit measure Level
RMSEA .071
IFI .956
TLI .939
CFI .956
CMIN/DF (NC) 1.990
AIC (saturated) 154.000
AIC (Default) 153.602
Mardia’s test coefficient of multivariate normality was 5.7. This was below the thresh- old value of 8, which is required to avoid violating the assumption of normality in maximum likelihood models (on which our SEM was based). Further, we assessed the univariate normality of the constructs in the SEM model, and they revealed no prob- lems with skewness or kurtosis.
4.2. Test results
Test results for the model presented in Figure 1 are shown in Figure 2.
Figure 2 presents the squared multiple correlations coefficient for the variables that are dependent in a relation. They are presented in brackets just below the label of the variable. The squared multiple correlations coefficient can be interpreted as the explained variance of the dependent variable. Figure 2 also shows the standardized regressions weights for each path (arrow) and the level of significance marked with
“***” representing a p value of less than .01. Table 6 expands the test results in Figure 2 by showing the standard errors, the critical ratio, and an overview of the support for the hypotheses. All hypotheses were supported in the structural equation model.
However, hypothesis 4b had a path coefficient with a p value of .064.
Figure 2. Test results
Flow
Delegaon
(0.13)
Cont. Improv empow
(0.26)
No. Of Employees (Size)
Performance (0.16)
Management
Follow up
(0.8) Floor training
(0.90) Rolemodels
(0.63)
Materials pr. unit
(0.47)
Lead me
(0.37) Worker
time/unit
(0.72)
Cost/unit
(0.72) 0.33***
0.41***
0.13(.04*)
-0.22(.003**)
-0.14(.064*)
0.19(.011**)
0.90***
0.95***
0.80***
0.68*** 0.61*** 0.85*** 0.85***
0.05
-0.21***
N=200
Table 6. Test results
Relation Hypothesis Supported P-value Standard error C.R. Standard coefficient
F D 1a Yes *** .036 4.916 .33
F C.I.A 1a Yes *** .042 6.371 .41
D C.I.A 2 Yes .040 .079 2.050 .13
M C.I.A 3a Yes *** .052 -3.282 -.21
M P 3b Yes .011 .066 2.529 .19
D P 4a Yes .003 .098 -2.994 -.22
C.I.A P 4b Yes .064 .082 -1.851 -.14
Abbreviations: F = standardized flow production; M = Lean middle management; C.I.A = continuous improvement empowerment grounded in analytics; D = delegation of decision rights based on a group agenda; P = performance
The results show that higher use of the delegation of decision rights and higher use of analytical continuous improvement practices positively affect performance.3 Leading up to this is higher use of standardized flow production practices, as they are posi- tively related to these variables, but flow production is not directly related to perfor-
mance. Moreover, firms that have middle managers who are committed to Lean and encourage the use of Lean will positively affect performance directly and indirectly through analytical continuous improvement.
To confirm hypothesis 1c, we also needed to test the indirect effects on performance from standardized flow production. This test supported hypothesis 1c, as the stand- ardized path coefficient showed the correct direction with –.096 and had a statis- tically significant p value of .007 (two-tailed significance, PC). While testing the indirect effects, we also assessed the indirect effect of management on performance.
This overall indirect effect was also statistically significant, with a p value of .011 and a path coefficient of .035, correct direction. This supported the general approach that the management’s role was somewhat mediated by the analytical continuous improvement variable, and, consequently, by the delegation of decision rights, as they are related.
In sum, all hypotheses were supported empirically in the structural equation model.
5. Conclusions
This research provides some of the first empirical evidence that Lean practices func- tion in a system-wide model to increase performance. Previously, Lean practices have primarily been studied as separate elements or without accounting for indirect rela- tions when the survey approach has been used. We sought to close this gap by using a structural equation model approach with several variables in a system, based on a survey. The data came from 200 Danish companies, which makes the study particu- larly interesting in a Danish context.
Testing our structural equation model representing Lean as a system-wide model showed that changing the operations of a business so they are more standardized flow oriented does not affect performance. Standardized flow production’s performance effects are mediated by continuous improvement based on analytical information support and by the delegation of decision rights with a group agenda. Thus, standards must be put in the context of kaizen (continuous improvement) behavior in order to enhance performance. Further, continuous improvement cannot be established with- out having standards in place and a transparent flow in which potential improvement areas are easier to identify.
Organizing in value stream and U-cells gives the option to delegate decisions rights to partially self-controlled teams. These teams are needed to deploy continuous improve- ment, and without them, there is less room within which to act. The survey showed that Lean organizations rely on teams to drive continuous improvements. These path dependencies from standard flow production to the delegation of decision to continu-
ous improvement and to performance is an important finding to support managers in their implementation of Lean.
Another conclusion is that confirmation of middle management’s role is important, as this increases performance and is related to increased use of analytical continuous improvement. Middle managers must participate in continuous improvement through Gemba and motivate line workers to perform appropriate Lean behavior. Thus, middle management behavior is not sufficient to enhance performance but supports continu- ous improvements and, in that respect, enhances performance.
In sum, implementing Lean as a piecemeal approach may result in incomplete utiliza- tion of the performance potential, as our study indicated the need to perceive Lean as a system-wide model where individual Lean practices may not affect performance di- rectly but are mediated by other Lean practices. Thus, there are indirect effects on per- formance. Therefore, senior management needs to be patient regarding performance expectations for Lean implementations as many practices need to be implemented, which is not done overnight.
The main limitation of the study is that we can only claim to have found evidence of an association between constructs, and not causal relations. The directions of the as- sociations are supported by a priori theory, however.
We want to thank DI for allowing us access to the survey data they collected.
References
Ahrens, T. and Chapman, C.S. (2004). Accounting for flexibility and efficiency: a field study of management control systems in a restaurant chain, Contemporary Accounting Research, Vol. 21, pp. 271-301.
Amstrong, J.C. and Overton, T.S. (1997). Estimating nonresponse bias in mail surveys, Journal of Marketing Research, Vol. 14, pp. 396-402.
Balakrishnan, R., Linsmeier, T.J. and Venkatachalam, M. (1996). Financial benefits from JIT adoption: effects of customer concentration and cost structure, The Accounting Review, Vol. 71, pp. 183-205.
Bicheno, J. and Holweg, M. (2009). The Lean toolbox, Picsie Books: Buckingham.
Blunch, N.J. (2008). Introduction to structural equation modelling using SPSS and AMOS, Sage: London.
Bollen, K.A. (1989). Structural equation with latent variables, Wiley: New York.
Bortolotti, T., Danese, P., Flynn, B.B. and Romano, P. (2014). Leveraging fitness and lean bundles to build cumulative performance sand cone model, International Journal of Production Economics, doi:10.1016/j.
ijpe.2014.09.014
Browne, M.W. and Cudeck, R. (1993). Alternative ways of assessing model fit. In Bollen, K.A. and J.S. Long (Eds.) Testing structural equation models, Sage: Newbury Park, pp. 136-162.
Carreira, B. (2005). Lean manufacturing that works: powerful tools for dramatically reducing waste and maxi- mizing profits, AMACON: New York.
Fullerton, R.R. and McWatters, C.S. (2001). The production performance benefits from JIT implementation, Journal of Operations Management, Vol. 19, pp. 81-96.
Fullerton, R.R. and McWatters, C.S. (2002). The role of performance measures and incentives systems in rela- tion to the degree of JIT implementation, Accounting, Organizations and Society, Vol. 27, pp. 711-735.
Fullerton, R.R., Kennedy, F.A. and Widener, S.K. (2014). Lean manufacturing and firm performance: the incre- mental contribution of Lean management accounting, Journal of Operations Management, doi:10.1016/.
jom.2014.09.002
Furlan, A., Vinelli, A. and Dal Pont, G. (2011). Complementarity and lean manufacturing bundles: an empirical analysis, International Journal of Operations & Production Management, Vol. 31, pp. 835-850.
Hofer, C., Eroglu, C. and Hofer, A.R. (2012). The effects of Lean production on financial performance: the medi- ating role of inventory leanness, International Journal of Production Economics, Vol. 138, pp. 242-253.
Huntzinger, J.R. (2007). Lean cost management: accounting for lean by establishing flow, J. Ross Publications:
Fort Lauderdale.
Huson, M. and Nanda, D. (1995). The impact of just-in-time manufacturing on firm performance in the US, Journal of Operations Management, Vol. 12, pp. 297-310.
Johnson, T.H. (1992). Relevance regained: from top-down control to bottom-up empowerment, The Free Press: New York.
Kennedy, F.A. and Widener, S.K. (2008). A control framework: insights from evidence on Lean accounting, Management Accounting Research, Vol. 19, pp. 301-323.
Kinney, M.R. and Wempe, W.F. (2002). Further evidence on the extent and origins of JIT’s profitability effects, The Accounting Review, Vol. 77, pp. 203-225.
Kline, R.B. (2005). Principles and practice of structural equation modeling, 2nd ed., The Guildford Press: New York.
Kobayashi, I. (1995). 20 keys to workplace improvement, Productivity Press: Portland.
Kristensen, T.B. and Israelsen, P. (2014). Performance effects of multiple control forms in a Lean organization: a quantitative case study in a systems fit approach, Management Accounting Research, 25, pp. 45-62.
Liker, J. (2004). The Toyota way, McGraw-Hill: New York.
Lind, J. (2001). Control in world class manufacturing: a longitudinal case study. Management Accounting Research, Vol. 12, pp. 41-74.
Milgrom, P. and Roberts, J. (1995). Complementarities and fit: strategy, structure, and organizational change in manufacturing, Journal of Accounting and Economics, Vol. 19, pp. 179-208.
Nunnally, J. (1978). Psychometric theory, McGraw-Hill: New York.
Ouchi, W.G. and Maguire, M.-A. (1975). Organizational control: two functions, Administrative Science Quarterly, Vol. 20, pp. 559-569.
Rother, M. and Shook, J. (1999). Learning to see: value stream mapping to add value and eliminate MUDA, Lean Enterprise Institute: Cambridge, USA.
Selto, F.H., Renner, C.J. and Young, S.M. (1995). Assessing the organizational fit of a just-in-time manufactur- ing system: testing selection, interaction, and systems model of contingency theory, Accounting Organiza- tions and Society, Vol. 20, pp. 665-684.
Shah, R. and Goldstein, S.M. (2006). Use of structural equation modeling in operations management research:
looking back and forward, Journal of Operations Management, Vol. 24, pp. 148-169.
Shah, R. and Ward, P.T. (2003). Lean manufacturing: context, practice bundles, and performance, Journal of Operations Management, Vol. 21, pp. 129-149.
Shah, R. and Ward, P.T. (2007). Defining and developing measures of lean production, Journal of Operations Management, Vol. 25, pp. 785-805.
Van der Stede, W.A. (2000). The relationship between two consequences of budgetary controls: budgetary slack creation and managerial short-term orientation, Accounting, Organization and Society, Vol. 25, pp.
609-622.
Womack, J.P., Jones, D.T. and Ross, D. (1991). The machine that changed the world, HarperCollins Publishers:
New York.
Worthington, R.L., and Whittaker, T.A. (2006). Scale development research, a content analysis and recommen- dations for best practices, The Counseling Psychologist, Vol. 34 (6), pp. 806-838.
Young, S.M. and Selto, F.H. (1993). Explaining cross-sectional workgroup performance differences in a JIT facility: a critical appraisal of a field-based study, Journal of Management Accounting Research, Vol. 5, pp.
300-326.
Notes
1. The respondents were asked whether their companies worked with Lean. Of the 459, 244 companies answered they did not work with Lean. Thus, they were excluded from further analysis. Five companies did not answer the performance questions, and they were also excluded.
2. The exception is “co-influence,” which is measured on a 7-point scale. We converted this to a binary item.
3. Notice that performance is scaled in reverse to analytical continuous improvement and delegation of decision rights; thus, a negative path coefficient means that they are positively related.
