• Comparison portfolio selection criteria:
o Industry match (2-digit SIC)
o Size match (+/- 50% of total asset size)
• Median z-score o Formulation
o Tests for industry and size effect control effectiveness
4.3.0 Lean Manufacturing Practice Variables - Objectives Constructs included:
Internally oriented lean practices
• Just in time production methods (JIT)
• SPC tools to monitor quality (SPC)
• Employee involvement (EMP)
• Group technology to enhance the flow of products (GT)
• Total productive maintenance (TPM) Externally oriented lean practices
• Communication with suppliers (SCM)
• Customer involvement (CUS)
[continued]
Table 4.1 [continued]
4.3.1 Survey Instrument
• 36 items covering 8 practice areas
• 5-point Likert scales 4.3.2 Data Collection
• Target high-level operations managers
• Web-based survey
• Contacts by mail and phone
• “Performance ranking” provided as incentive 4.3.3 Survey Sample Validation
Frame and sample comparison for:
• Representativeness
• Response bias 4.3.4 Practice Construct Formulation
• Factor analysis
• Construct validity and reliability
• Factor score measurement 4.3.5 Lean Archetype Formation
• Cluster analysis to form lean and non-lean clusters
• Validation through:
o Rationalizing practice levels within clusters o Discriminant analysis
o Predictive validity with respect to operations financial performance measures
4.4.0 Relationship Analysis – Objective:
Describe how the relationships between lean practice variables and financial performance were analyzed.
4.4.1 Lean Archetypes and Operations Financial Performance
• Wilcoxon rank-sum test comparing lean and non-lean archetypes with respect to performance.
• Logistic regression of lean and non-lean archetype classification on a set of performance variables 4.4.2 Individual Lean Practices and Operations Financial Performance
• Logistic regressions for each performance variables as above or below industry median performance on all practice measures
4.4.3 Lean Archetypes and Business Financial Performance
• Wilcoxon rank-sum test comparing lean and non-lean archetypes with respect to performance.
• Logistic regression of lean and non-lean archetype classification on a subset of performance variables 4.4.4 Individual Lean Practices and Business Financial Performance
• Two sample t-test comparing above and below industry median performers on their level of practice implementation
• Logistic regressions for each performance variables as above or below industry median performance on all practice measures
[continued]
Table 4.1 [continued]
4.4.5 Lean Operations and Business Financial Performance
• Logit models developed in the survey sample are used to predict lean and non-lean classification in non-respondents in the sample frame based on operations financial
performance measures.
• Business financial performance of classified companies is compared using Wilcoxon rank-sum tests.
4.5 Summary
The study’s research method is summarized.
4.1 Sample Frame 4.1.0 Objectives
Criteria for including companies in the sample frame were developed by imposing standards for the following:
• Availability of archival financial data
• Number of business product categories (4-digit SIC codes) in which the company is reported to participate
• Size of the company in terms of the number of employees
• Industry group
• Number of years of available data.
The specific criteria and rationale are described below. The net result of imposing these criteria resulted in a sample frame of 316 companies. Table 4.2 provides the sample frame companies’ descriptive statistics by industry (SIC code) and key characteristics (i.e. total asset size, number of employees, total inventory, and net sales) and shows data for fiscal year 2001 because this marks the last full year of data available in the employed research databases upon data collection. A complete listing of this data for the sample
frame companies is included as Appendix A. Company identities are coded to maintain survey response confidentiality.
Variable SIC N Mean Median Min Max
Assets - Total (MM$) 28 81 410.7 136.2 8.40 2268.1
35 56 167.5 52.3 5.70 1871.8
36 85 336.6 98.9 4.20 2865.8
38 94 225.8 42.6 5.20 1798.2
Overall 316 292.7 72.1 4.20 2865.8
Employees - Number (M) 28 0.8 0.2 0.05 8.2
35 0.7 0.3 0.05 7.0
36 1.1 0.5 0.06 7.3
38 0.9 0.2 0.05 7.7
Overall 0.9 0.3 0.05 8.2
Inventory - Total (MM$) 28 36.9 4.3 0.00 468.9
35 25.6 11.5 0.04 321.7
36 33.4 13.8 0.16 329.1
38 32.9 6.8 0.04 335.4
Overall 32.7 9.3 0.00 468.9
Sales - Net (MM$) 28 216.9 42.2 0.00 1921.0
35 148.0 63.6 6.70 1519.8
36 229.2 91.1 4.80 2206.0
38 186.6 46.1 0.60 1619.5
Overall 199.0 56.2 0.00 2206.0
Table 4.2: Summary of descriptive characteristics for firms included in the sample frame. Data is for fiscal year 2001.13
4.1.1 Availability of Archival Financial Data
Since financial performance is a key variable in the study and companies often begrudge sharing this type of data with researchers, the study logically began by identifying a large pool of manufacturing companies for which financial account data is readily available.
Compustat and CRSP databases suited this purpose. Compustat contains multi-year data for over 24,000 publicly held companies (Compustat, 2003). Of these, 7542 indicate participation in manufacturing standard industry classification codes 20 to 3914. In
13 M = 1000, MM = 1,000,000
14 Count based on 2002 fiscal year data.
addition to supplying a wealth of archival accounting data for individual companies, Compustat has the advantage of being well-documented and used in many research studies.
4.1.2 Business Scope
Another key concern stresses that the financial data apply to the specific operations under survey. A problem in operations management research is that the effects of operational practices have attenuated by the time they reach the business level—a natural result of business diversity and multidivisional organizational structures. Management practices in strategic business units often lack uniformity across an entire company. This problem limits a majority of previous survey-perceptual studies in operations management to targeting for analysis the plant or, at most, the strategic business unit. In order to examine business level performance and minimize the effects of diversity, this study limits the sample frame to companies that report participation in only one 4-digit SIC code in the Compustat database. At the time of compilation, there were 2044
manufacturing firms that reported participation in only a single 4-digit SIC code15. The fiscal years 2001 and 2000 were used because Compustat archives ending fiscal year results as they become available. At the time of sample frame definition, 2001 stood as the most recent year with a full set of company data available. By selecting only companies reporting a relatively narrow range of product offerings, confidence was increased that the business-level results actually related to the manufacturing
15 Active companies in fiscal years 2000 and 2001 were searched to maximize the starting list of potential companies.
management practices surveyed. Also, companies were eliminated from consideration if a portfolio of at least five comparison companies with at least the same 2-digit SIC code and a size match within +/- 50% of the company’s total asset size was not available in the database.
4.1.3 Size Restriction
Size restrictions placed on sample frame companies, a minimum of fifty and a maximum of 9,000 employees in 2001, enhanced the confidence that business results were being attributed to the relevant management practices. Arguably, below the 9,000-employee level, a single high-level operations management respondent should be sufficiently aware of the practices used across the company to make valid survey responses. This assertion seems especially true when applied in conjunction with the limitation on the number of product lines imposed by the single 4-digit SIC code restriction because the size
restriction makes uniformity of practices across the operation more likely. Regarding the minimum requirement of fifty employees, small companies are not likely to have the team-based infrastructure associated with lean operations. The companies in the sample frame ranged from fifty-one to 8,175 employees. The final sample of companies
responding to the survey ranged from fifty-two to 7,500 employees.
4.1.4 Industry Group
The sample frame includes only four major industry groups as defined by 2-digit SIC codes. They are:
28: Chemicals and Allied Products
35: Industrial, Commercial Machinery and Computer Equipment
36: Electronic, Electrical Equipment and Components, except Computer
38: Measuring, Analyzing, and Controlling Instruments; Photographic, Medical, and Optical Goods
These four major industry groups are the most highly represented in the Compustat data after the imposition of the preceding criteria. By having only four industry codes, comparisons can be made between industries without overly taxing the sample size requirements of multivariate statistical analysis. Research can be performed more efficiently because the range of the industries included limits the scope of alternative explanations for performance. Alternatively, the generalizability of research results is enhanced by representation from a broad spectrum of industry and process types. Both discrete and process industries are represented in the sample. SIC 28, the chemical and allied products industry, is more likely a process-based industry than the other three. SIC codes 35, 36, and 38 are more likely discreet product manufacturers. Lastly, the selection of these SIC codes overlap those studied by previous researchers who did not include the entire range of manufacturing SIC codes. (Shah, 2002) included 35, 36, and 38 and (Koufteros et al., 1998) included 35 and 36. Broad categories of the electronics and machinery industries were covered by Callen et al. (2000), Cua et al. (2001), Flynn et al.
(1995), and Sakakibara et al. (1997). The distinguishing attribute of the current study is the limiting of potential respondents to companies that report participation in only one SIC code.
4.1.5 Years of Data Available
This study sought to improve temporal validity by examining measures of sustained performance over five years. Of all the recent survey empirical studies included in the literature review, only Claycomb et al. (1999) looked beyond current year performance data16. Archival data studies examined less than one year (Hendricks & Singhal, 1996;
Howton et al., 2000) to twelve years (Billesbach & Hayden, 1994a) worth of performance data per company. During preliminary exploration of the data available for companies in the sample frame, it was noted that some companies with data reported for only the previous five years (1997-2001) exhibited what seemed to be start-up phenomena, in that returns would start at very low, erratic levels and then seem to stabilize in years three through five. Start-up companies were excluded from the sample frame by including only companies that had reported data for at least eleven years (1992-2002). Data from 1998 to 2002 was used to calculate sustained financial performance for the sample frame companies.
4.2 Financial Performance Variables 4.2.0 Objectives
This study examines the effect of lean manufacturing practices on two levels of financial performance: operations and business. Variables at the operations financial level tend to be influenced directly by the operation’s function. Operational variables used in this
16 Claycomb, et al asked managers to estimate their ROI, ROS, and profit performance over 3 years with respect to competitors.
study include asset productivity, employee productivity, gross margin ratio, and cycle time. Business financial variables are not directly attributable to any specific function.
Return on equity, sales growth, and stock return are the business financial variables used.
Table 4.3 provides a listing of the study’s financial performance measures.
One criterion for selecting financial variables maintains that the variables must measure sustained performance and not be unduly influenced by abnormal changes in a particular year’s performance. To address this criterion the five-year median performance for each financial variable is used. 17 The use of a five-year performance window (1998-2002) and its median, assures less bias by outlier data than the five-year mean performance.
This increased confidence that the performance being explained is sustainable.
Several general objectives guide the selection of financial variables. The variables must be readily available. Public sources of information, such as those supplied by Compustat and CRSP databases were used exclusively (Compustat, 2003; CRSP, 2003). The use of public data sources minimized the problem of missing data that is usually encountered when one requests sensitive information from companies. Since the data is public, the financial elements of the study can be replicated and verified. The Compustat and CRSP databases, although not perfect, have been thoroughly researched and tested as tools for
17 The value of using the median as the primary indicator of central tendency in financial data analysis is illustrated by the following story: “A local bar is populated with a dozen democrats and a dozen
republicans. Bill Gates, Chairman of Microsoft, walks in the door and the democrats think that in general they are no better off (median net worth has changed only slightly). The republicans think that on average they are all multi-millionaires (mean net worth has changed dramatically).” – Attribution unknown.
conducting financial research on companies. A drawback of using an existing financial database is that the availability of component data constrains variable formulation.
Another standard in the selection of financial variables holds that they be common and understandable to operations managers. This research does not aspire to extend the art of financial analysis; rather, value to practitioners is a key consideration. This study also could serve as a baseline for future research looking to improve financial measures. That financial variables will be more or less effective in predicting or reflecting the effects of certain lean manufacturing management practices is to be expected. Neither does this study propose an optimum set of measures. An effort was made to be comprehensive yet parsimonious in the selection of financial variables. The archival studies in the literature review chapter cover over twenty-eight distinct financial measures. This study focuses on eight measures at the operational level, including a set of five distinct, yet related, cycle time measures, and three business level measures for a total of eleven financial measures.
A final objective in creating the set of financial variables is comparability. Comparability between and among companies and industries is a critical requirement. This issue is addressed at two levels. First, ratios of absolute performance were created by applying a scaling factor. For example, equity book value, number of employees, and cost of goods sold are used as scaling factors for income before extraordinary expenses (ROE),
operating income (employee productivity), and total inventory (inventory cycle time), respectively. Secondly, individual company performance is compared to that of a
matched portfolio of similar companies. Matching is based on a minimum of a match on 2-digit SIC code and on total asset size in 2001 to with +/- 50 %. Several researchers assert that industry and size are two of the most influential determinants of firm financial performance (Barber et al., 1996; Dess, Ireland, & Hitt, 1990; Mauri & Michaels, 1998;
Moch, 1976). A “median z-score” is created for each company financial measure providing a measure of company performance with respect to the median value of the comparison portfolio18.
Return on Equity ROE
Sales Growth SG
Stock Return SR
Return on Cash Adjusted Assets ROCA
Employee Productivity EP
Gross Margin Ratio GMR
Cash to Cash Cycle time CTC
Total Cycle time CTT
Inventory Cycle time CTI
Payables Cycle time CTP
Receivables Cycle time CTR Business
Operational
Table 4.3: Financial performance measures.
4.2.1 Business Financial Performance Variables
Three overall measures of business financial performance are utilized in this study: return on equity (ROE), sales growth (SG), and stock return (SR). To ensure sustained
18 The term “median z-score” is used to indicate a measure that is calculated based on a value’s difference from the median and the sum of squared defenses from the median. Normal z-score calculations use the mean. See Sec. 4.3.3 for details.
performance, five years of data from 1998 to 2002 was utilized. Two different measures of central tendency are applied to the business financial performance measures (the median is used for ROE and SG and the mean for SR). Applicable sections explain the reasons for selecting different measures. Compustat and CRSP databases supplied the raw financial data.19
4.2.1.1 Return on Equity (ROE)
Return on equity is a well-accepted measure of overall business performance. ROE is given a high level of credibility by value investors such as Warren Buffet because of its comprehensive nature and companies’ inability to manipulate its value (Vick, 2001).
ROE measures a firm’s ability to generate profits in comparison to the historical book value of its assets and is computed as income before extraordinary items divided by the book value of common equity. The average equity for the year is assumed the most relevant with respect to evaluating current year returns. Common equity is essentially assets minus liabilities. In equation form:
( )
60 2 60
18
1 t
t D
D D Equity
Common of
Value Book
Items ary Extraordin Before
Income
= +
−
Variables beginning with “D” stand for data items available in Compustat. A complete listing of data items and their descriptions is included in Table 4.4.
19 Individual company years with missing or zero values in the financial ratios numerator or denominator were excluded from the analysis in all cases.
D1 = Cash and Short-Term Investments (MM$) D3 = Inventories, Total (MM$)
D6 = Assets Total (MM$) D12 = Sales, Net (MM$)
D13 = Operating Income Before Depreciation (MM$) D18 = Income Before Extraordinary Items (MM$) D29 = Employees (M)
D41 = Cost of Goods Sold (MM$) D60 = Common Equity, Total (MM$) D70 = Accounts Payable (MM$) D151 = Receivables, Trade (MM$)
Table 4.4: Reference list of Compustat data descriptions (Compustat, 2003)20
4.2.1.2 Sales Growth (SG)
Sales growth measures a firm’s performance with respect to increasing its revenues over the previous year’s. Because a key criterion in valuing stock is a company’s ability to generate ever-increasing returns for its investors, growth is an imperative for most traded businesses. By looking at sales growth in relation to a matched set of comparison
companies (e.g. “median z-score”), the sales growth measure provides a sense of how well the company is taking market share from its competitors. If its growth rate is greater than the median value for the comparison companies then a firm is ostensibly gaining
20 M=1000; MM=1,000,000.
market share. Sales growth is computed as the change in sales from the previous year divided by sales in the previous year. In equation form:
1 1
1 12
12 12
−
−
−
= −
t t t
t D
D D
Sales Sales in Change
4.2.1.3 Stock Return (SR)
Stock return is a market-based measure indicating the premium that the market places on the potential earnings stream for a company. The average annual return for a given company is calculated as the arithmetic sum of the monthly returns adjusted for dividends and splits for the five-year period (60 months) divided by five. The CRSP database provides the adjusted monthly return value. RET is the “Holding Period Total Return.”
“Returns are holding period returns from month-end to month-end, not compounded from daily returns, and ordinary dividends are reinvested at month-end” (CRSP, 2003). Since the return is reset and calculated monthly, indexing to a specific time point is not
necessary and the arithmetic mean is an appropriate alternative to the geometric mean that is often used in stock return calculations.
Since SR is a market-based measure, it is assumed that trend and variability, or risk, is included. In equation form:
5 5
60 1 60
1 ∑
∑m= = m=
RET Return
Stock Monthly
A complete explanation of the algorithm CRSP uses for calculating the monthly return values (RET) is provided on the database website)21.
With respect to the stock return data, two methodological issues are noteworthy. The first is that SR stands as the only financial measure in this study to use the mean, rather than the median, as its measure of central tendency. Since 60 stock return data points were available for each firm, as opposed to only five annual data points, it may be assumed that a stock return average value is less susceptible to outliers. In addition, researchers have found that stock return data distributions generally do not suffer as much from heavy tails (outliers) as other financial measures (Barber et al., 1996). The second issue is that SR uses calendar year data as opposed to the other financial measures that use fiscal year data. Over 50% of the companies in Compustat have December close
21 “Holding Period Return Variable Name = RET
A return is the change in the total value of an investment in a common stock over some period of time per dollar of initial investment. RET(I) is the return for a sale on day I. It is based on a purchase on the most recent time previous to I when the security had a valid price. Usually, this time is I - 1. Returns are calculated as follows:
For time t (a holding period), let:
t’ = time of last available price < t r(t) = return on purchase at t’, sale at t
p(t) = last sale price or closing bid/ask average at time t d(t) = cash adjustment for t
f(t) = price adjustment factor for t
p(t’) = last sale price or closing bid/ask average at time of last available price < t.
then r(t) = [(p(t)f(t)+d(t))/p(t')]-1
t’ is usually one period before t, but t’ can be up to ten periods before t if there are no valid prices2 in the interval. A series of special return codes specify the reason a return is missing.” (CRSP, 2003)