Assessing The Resource Base of Japanese and U.S. Auto Producers A Stochastic Frontier

Assessing the Resource Base of Japanese and U.S.
Auto Producers: A Stochastic Frontier Production Function Approach Author(s): Marvin B. Lieberman and Rajeev Dhawan Source: Management Science, Vol. 51, No. 7 (Jul., 2005), pp. 1060-1075 Published by: INFORMS Stable URL: http://www.jstor.org/stable/20110398 . Accessed: 17/06/2013 04:38
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Management Science.
http://www.jstor.org
This content downloaded from 111.68.103.120 on Mon, 17 Jun 2013 04:38:58 AM All use subject to JSTOR Terms and Conditions
SCIENCE MANAGEMENT M3H.

Vol. 51, No. 7, July 2005, pp. 1060-1075
ISSN 0025-19091 EissN 1526-55011051510711060 ? 2005 INFORMS
DOI 10.l287/mnsc.l050.0416
the Resource Base of Japanese and Assessing Frontier U.S. Auto Producers: A Stochastic Function Approach Production
Marvin
Anderson School of Management,
B. Lieber
man
Los Angeles, California 90095-1481,
of California, Los Angeles, University marvin.lieberman@anderson.ucla.edu
Rajeev Dhawan
J.Mack Robinson College of Business, Georgia State University, Atlanta, Georgia 30303, rdhawan@gsu.edu
"resource-based
view
of
the
firm,,
has
become
an
important
conceptual
framework
in
strategic
man
The agement but has been widely criticized for lack of an empirical base. To address this deficit, we utilize a new method for identifying interfirm differences in efficiency within the context of stochastic frontier produc tion and functions. capabilities and superior Using and scale data test on for Japanese linkages at the and with firm U.S. firm and automobile performance. level. plant manufacturers, The We results apply measures of resources develop the influence of manufacturing to account the parameter estimates for show we
proficiency Toyota's Key
economies
efficiency
relative view;
to other stochastic
producers. frontier; production function; automotive
words:
resource-based
History: Accepted
authors 1 year
by Daniel
and 9 months
Levinthal,
for
business
strategy; received
July 26, 2001. This paper was with
the
2 revisions.
1.
Introduction
years strategic management enormous in the interest scholars have resource-based
In recent
expressed view (RBV) of the firm. This perspective regards the firm as a heterogeneous bundle of resources?some superior and others perhaps inferior?plus organiza that may enable the firm to deploy tional capabilities resources more in than rivals. Variation efficiently and capabilities leads to the the quality of resources of economic which rents, may appear as generation in profitability. Such performance differences differ is imitation entials can persist for long periods when impeded. its appeal as a conceptual the framework, Despite RBV has often been criticized for lack of an empirical have been base. Few researchers able to develop measures and capabilities, their of resources identify a in and link context, importance specific industry resource to of dimensions firms' positions perfor mance. In this paper we attempt such an investigation, data on Japanese and U.S. automobile using historical Our aim is to better understand competi companies. in the global automotive tive heterogeneity industry. our study, we utilize a new method To implement literature. Recent work ology from the econometrics and Battese of Coelli (1995) on the estimation by stochastic frontier production functions (SFPF) pro
a framework the sources of inter for identifying in efficiency. We develop firm differences dynamic measures and capabilities and assess of resources their impact on automotive company performance the Battese and Coelli model. We demonstrate using con the potential of this model for adding empirical our tent to the RBV. More generally, links the analysis vides perspective
tion economics.
of the RBV with
the methods
of produc
show differ findings long-lived operational in our sample. the auto producers among are equiva in efficiency such differences Arguably, lent to sustained (Porter 1996). competitive advantage Our ences in the efficient producer Toyota has been the most auto industry in recent decades, and many prior stud ies have addressed the sources of Toyota's superior 1999a, b). Using Toyota as performance (e.g., Fujimoto an example, we show how the SFPF can methodology a firm's over be used to quantify rivals. advantage as follows. Our discussion is organized First, we introduce tion, the concept of a frontier production to the RBV its connection developing context func in the
specific describe
of the automotive industry. We then the Battese and Coelli and the (1995) model our in is used This followed study. specification by estimation results for the 11 firms in our sample over period from the 1960s to 1997. Draw
a three-decade 1060
Lieberman Management
the Resource Base of Japanese and Dhawan: Assessing Science 51(7), pp. 1060-1075, ?2005 INFORMS 1061
and U.S. Auto
Producers
ing from these estimates, we make summarizes isons. A final section

concludes.
interfirm compar and the findings
Figure 1
ProductionFrontier FirmsRelative to Industry
2. The RBV Within

Function Model Performance
a Production
of Firm
Frontier Production Function Models the RBV views the firm as a bundle of Where resources economics and capabilities, neoclassical in which labor and cap considers the firm as a vessel as and ital (and other potential such materials inputs, are to combined form outputs. productive energy) is captured by the concept of a "produc This notion
tion function," e.g.,
XQ
Input x
heterogeneity,
approaches
whereas
assume
more
traditional
economic
it away.
Y = F(K,L), (1)
Y denotes the firm's output, and K and L are where its capital and labor inputs. Given data on a sample meth of firms in an industry, standard econometric the industry production ods can be used to estimate function. The estimated show the trade parameters as as economies offs between well inputs potential data are available, of scale. If time-series the rate of technical progress biases) (and possible input-saving can also be estimated. A vast literature applies such to firm- or plant-level data on a range of methods it is recognized that individual industries. Although func firms can deviate from the industry production as are such deviations taken random tion, normally
error.
1 illustrates the concept of an industry pro the maximum where feasible out frontier, from of is any quantity put, Y*, input given by the = X to the Y* where function, F(X), corresponds on of In firm A lies 1, (1). inputs Equation Figure this efficient frontier, whereas firm B falls below. Both of input, but B has firms consume the same quantity Figure duction lower output. The technical efficiency (TE) of firm B is as the ratio of B's output to that of fully effi defined can be thought cient firm A. Thus, technical efficiency of as the firm's scaling factor relative to the frontier in the range: 0 < TE < 1. The output of any firm i can be written as, Yz = F(Xi)TEi, or in the case of a two-input function, F(K?, L?)TE2. of the SFPF model it possible forms made Early to estimate the industry production function and the data. technical efficiency of firms using cross-section in the determinants of techni Researchers interested a second cal efficiency have often pursued stage of are on a set estimates the where TEZ analysis regressed of explanatory factors (e.g., Caves and Barton 1990, Caves and Posen this 1992, Knott 2005). Although suffers from two-stage procedure conceptual prob it lems (Kumbhakar and Lovell 2000, pp. 262-266), a for method differences provides assessing efficiency at a specific point in time. Given that data are needed on many individual units, studies of this type have focused on industry-level factors rather than the per formance of specific firms. In recent years, panel-data models have been devel production for SFPF. This paper utilizes the panel-data oped of and which allows Battese Coelli (1995), approach a function of as to technical estimated be efficiency a pro factors. Consider firm-specific, time-varying of the form, duction frontier model Yit = F(Kit,Llt)TE(Zlt), (2) firm i in period t, and labor capital
with this traditional is that approach the function embodies the conceptually, production trade-offs faced by an efficient firm that utilizes best most for its industry. However, practice methods firms are not fully efficient in their use of inputs, and thus they fall below the industry frontier. Econo et al. (1977) and Meeusen metric advances by Aigner and van der Broeck (1977) led to the development to identify of SFPF models that can be estimated the production of frontier and the relative positions firm firms.1 Such SFPF models explicitly recognize One problem
1 Kumbhakar of SFPF overview. three and and Lovell (2000) Greene survey the historical a
development technical good literature offers
and models, In addition
techniques
tor productivity
least-squares for analysis but of multioutput tages production vide statistical for estimated inference parameters.
to SFPF, for the analysis of productive total fac efficiency: data envelopment (TFP) indices, (DEA), analysis DEA has advan econometric models. production does The not SFPF pro and
(1997) provides the econometrics
allow for such inference, and the SFPF least-squares methodologies a model that it incorporates has the further advantage of approach can be exam so that interfirm the inefficiency differences effects, ined. For an overview Coelli et al. (1998). and comparison of these methodologies, see
the output of where Yit denotes and Kit and Lit are the firm's
Lieberman
and Dhawan:
the Resource Base of Japanese and U.S. Auto Assessing Science 51(7), pp. 1060-1075, ?2005 1062 Management
Producers INFORMS
is determined of F() by the product inputs.2 Output to and TE(). The first term, F(Kit,Lit), corresponds in the industry's function production "best-practice" t. A firm that meth period fully employs best-practice ods (given its current levels of Kit and Lit) and exe t would cutes perfectly in period lie on the frontier the firm's represented by F(). The TE term represents as a func is parameterized technical efficiency, which tion of firm-specific factors, denoted by the vector, Zit. Given panel data on firms in a given the industry, Battese can be used to and Coelli (1995) approach the parameters estimate of such a production func tion and efficiency model. The approach offers advan both the tages over prior methods by estimating
substitute." coalitional upon Building MacDonald and Ryall (2004) have out game theory, set of conditions lined a more under which precise a firm can be shown to both create and appropri ate economic value. Such studies provide only limited on ways to implement however, guidance, empiri cal tests of the RBV, or even to calibrate the impor tance of resources and capabilities in a given industry
environment.
"difficult
to
function and the determinants of firm a so in a in and stage, efficiency single by doing to vary that allows parameters and efficiencies way an estimate over time. The approach of also yields in of each firm the technical each year, efficiency which allows the dynamic performance of firms to be production directly compared. View of the Firm
on how firm emerged factors should be classified under the RBV. specific are provided Some of the clearest distinctions by as observ Makadok "resources" (2001), who defines able assets that can be individually valued and on the other hand, are organi traded; "capabilities," and thus can be transferred only embedded zationally sale of the firm (or major Such subunits). through are useful but incomplete, as definitions they ignore of producer like those derived categories advantage are important from economies of scale, which in auto and other industries. manufacturing (Ironically, scale are one of the few dimensions of firm differentials Reasonable has economists, recognized heterogeneity by conventional in the literature on the yet they are barely considered et al. (2003) suggest that scale advan RBV.) Hoopes but fall in tages are neither resources nor capabilities,
a separate category of "cost drivers."
consensus
Resource-Based In parallel with
economists' of frontier development business researchers functions, strategy production have elaborated the "resource-based view of the firm" as a conceptual for assessing framework interfirm dif in performance ferences 1986, Rumelt (e.g., Barney
and Cool 1989, Peteraf 1993). The RBV 1987, Dierickx as a perspective must be regarded rather than a as debate over con continues its essential theory, et al. 2003). The central structs idea?that (Hoopes can be traced in performance sustained differences in firms' resources and back to underlying differences some incontrovertible. Indeed, capabilities?seems a tautol have that the RBV is essentially argued have given ogy (Priem and Butler 2001). Researchers structure to the RBV by proposing definitions of resources and capabilities and the conditions under to competitive which Even they contribute advantage. so, the RBV has lacked the clarity required for empir to opera and it has proved difficult ical specification, across firms tionalize the RBV in a consistent manner Empirical work on the RBV has been to mod lacking common approaches and measures, eling, testing. hypothesis RBV researchers have sought to identify conditions resource endowments under which support superior and industries. largely ad hoc, on imperfections inmarkets for Focusing profitability. resources that factors, (1991) proposed input Barney to sustained when contribute competitive advantage are to "difficult and "valuable," "rare," imitate," they
the list of Ideally, data are unavailable fore, we measure 2 raw materials, include but such inputs should firms in our study. There for the automotive (i.e., value added). output net of materials inputs
In this paper we do not attempt to resolve the to with RBV. the debate Rather, conceptual regard we propose can that SFPF models help give struc ture to empirical work in this area of research. Other studies in the strategic management field have begun to provide content. the RBV with greater empirical For example, Henderson and Cockburn (1994) doc on the patenting umented influences firm-specific behavior of pharmaceutical and Helfat companies, to (1997) found that the ability of energy companies into synthetic fuels was determined diversify largely field by the nature of their resource base. Quantitative such as Clark and Fujimoto's studies, (1991) work on product in the automotive development industry, on the evidence in of differences provide magnitude firms' capabilities and their performance implications. In the marketing et al. (1999) Dutta literature, a two-stage introduced for operationalizing approach are used the RBV: SFPF methods in the first stage to obtain estimates to relative of firms' capabilities are the industry estimates frontier; these capability on measures then regressed of firms' financial per formance. necessary formance study we While such multistage may be procedures to incorporate of per financial measures an SFPF framework, within in the present
a more direct based on propose approach the single-equation model frontier production (Equa resources tion (2)). This model that are incorporates to production, essential and labor, notably capital
the Resource Base of Japanese and Dhawan: Assessing INFORMS 1063 Science 51(7), pp. 1060-1075, ?2005
and U.S. Auto
Producers
all other influences the production within function; effi are included in the Zit vector of the technical as we the consider capital and labor ciency term. If
firm's basic resources, and Zit as its vector of capa
formalizes the model bilities, of the RBV. For example, Amit that resources p. 35) state that are owned able factors firm... property, etc. Capabilities, resources... to deploy
some
common
notions
and Shoemaker (1993, are "stocks of avail or controlled by the human and capital, equipment, plant in contrast, refer to a firm's capacity
can abstractly be thought they of 'intermediate by the firm to pro goods' generated its Such resources." of vide enhanced productivity fron of the the RBV resemble of the logic conceptions tier production function model. a common the focus on firm heterogeneity, Despite have emphasized different a frame SFPF models provide performance RBV has the whereas for assessing work efficiency, in to account variation for been invoked traditionally we that Nevertheless, propose efficiency profitability. studies for empirical may represent a superior metric of the RBV Indeed, Peteraf and Barney (2003) argue of that the RBV is "an efficiency-based explanation RBV and SFPF literatures metrics. differences" performance minants of profitability,
collusion.
at the firm level, are com Our methods, applied to "insider econometrics" the approach plementary or data to test hypothe using plant production-line ses on the nature of organizational (e.g., capabilities and Shaw Ichniowski 1999, 2003). We use publicly on others' accounts data and draw heavily accessible to our to give structure of the industry phenomena the into model. methods offer Our insights empirical sources of efficiency but do not yield the differences in more that is possible understanding fine-grained studies. microlevel
3.
A
Drivers
of Productive
the Automotive
such
that excludes as market
other
deter and power
have value only in firm's resources and capabilities of context. One can easily identify generic categories cat such businesses, (for manufacturing capabilities include product production, design, egories would relations, marketing, etc.), but the specifics supplier on the industry For example, environment. depend in automotive skills the industry are product design that support drug clearly distinct from the capabilities in the pharmaceutical and development discovery any empirical study of the RBV must industry. Hence, resources in the industry and capabilities consider
Efficiency Industry
in
in efficiency but differences emphasizes a of firm perfor broader also conveys conception mance than might be immediately apparent. We esti a mate of Equation transformation (2) that allows Our model and scale economies, capital efficiency, in terms of their impact on to be evaluated measure a of per labor productivity, comprehensive in manufac that is particularly meaningful formance gains flow not only to turing industries. Productivity the firm's shareholders (as increased profits), but also consumers to employees (price (wage increases) and of economic from the standpoint Thus, reductions). a more fundamental welfare, represents productivity metric than Furthermore, profitability. performance can be a misleading indicator of performance profits across countries where are drawn when comparisons auto makers, conditions differ. Among competitive as a supe for example, Toyota is commonly regarded over rates the rior competitor, yet Toyota's profit technical investment period Motors fell below sample whose (GM), performance as has often been described poor.3
the ratio of period, on average, for sales was slightly higher, been (4.9%). Toyota has, nevertheless, Toyota In each year since standards: by Japanese the 1983-1997 ing income combined. has exceeded that of all other 3 Over
hold value. context where they potentially automotive Prior studies of the industry offer guid ance on the types of resources and capabilities likely cited to be important in that sector. In a widely on et al. Womack the automotive book industry, recent in that best practice has shifted (1990) suggest to of "mass production" from a paradigm decades mass one of "lean production." While pro aspects of innovations duction still matter, by leading Japanese in factory have led to important changes producers with and coordination management, product design, on recent the automotive studies suppliers. Many these trends and their impli industry have explored and Nobeoka cations 2000; 1996; Dyer (e.g., Dyer and and Sako 1995; Novak 1999a, b; Helper Fujimoto 2001). Eppinger we In the following, the "mass produc adopt to organize dis distinction tion/lean production" held by and capabilities of key resources cussion as possi assemblers. To be as specific automotive measures in we used how the show ble, empirical our study have for our sample of eight evolved from the mid Japanese and three U.S. auto makers 1960s through 1997.4 Our data are entirely from public
all of the major includes firms that produced passen sample States with in Japan and the United their own name ger cars under we omit observations For Honda, the exception of Mitsubishi. prior of motorcy to 1975, when the firm's output consisted primarily varies cles. The starting year for other Japanese slightly, producers depending on data availability. 4 The
of our
those of General in recent decades
income operating GM (6.0%) than
to for
enormously profitable 1983, Toyota's operat auto makers Japanese
Lieberman
and Dhawan:
Producers INFORMS
annual seg reports on business (company motor unless to vehicle production, relating some of otherwise indicated).5 Given data limitations, serve only as weak our measures for types proxies in the literature that are highlighted of capabilities on the automotive industry. We include these imper as our objective is to in our analysis, fect measures a are that set indicators of compre dynamic develop to capture the at least in principle, hensive enough, considered of heterogeneity main dimensions impor tant in the auto industry. At the end of the paper we stem from deficien that may various biases discuss other limitations. cies in our measures, among sources ments Scale Economies that has always been One mass production concept is economies in the automotive industry important incur substantial model-specific of scale. Producers in the automotive value fixed costs at many points are investments chain. Such fixed (and mostly sunk) and for pro development, product design required duction dies and tooling, and in some cases, adver in the mass market, high tising. To be competitive life model volumes cycle or per year, depending (per are required to spread the fixed on the cost element) costs over many units. Given continual improvements in technology and uncertainty sales, regarding model the firm must be large enough to sustain the frequent of new vehicles. development scale is its total Our measure of a firm's overall employment. ees for the year. range Isuzu, 2 plots the count of employ Figure firm and automotive companies by to axis is the used (A logarithmic capture our in firms Daihatsu, among Fuji, sample.) and remained very small; their been less than has always employment GM's (worldwide) employ comparison, These size exceeded 600,000 for decades. While have been remarkably persistent. can sometimes for inter substitute makers Suzuki have striking alliances with
such as
Figure2
Numberof Employees
?-sh?GM a Ford
? ?
a Chrysler ? 72 74 76
Toyota Nissan Honda 78 80
+ Isuzu ? Mazda - Suzuki * Daihatsu < Fuji (Subaru) 82 84 86 88 90 92 94 96
64
66
68
70
savings from the spread of product costs may depend on the annual or lifetime volume of specific vehicle models (or of prod uct families that share components). We have chosen measures of scale in not to incorporate vehicle-specific firm. For example, design and tooling this study, as they are hard to assess and only partially the ability to under the control of the firm. Moreover, to firm be related scale achieve these economies may our in is already incorporated size, which analysis. A related dimension of scale that is both meaning size of the firm's is the average ful and observable studies final assembly plants. Early engineering (e.g., that most scale Pratten 1971) suggested plant-level at a vol in the auto industry are achieved economies of about 200,000 annual vehicles per In the the 1960s, however, Japanese assembly plant. of automotive to modify the configuration began on-site stamping with assembly plants by combining two or more vehicle lines to create a much assembly new organization is If this volume per plant. higher more cost-effective, we should find gains in efficiency as plant size increases to 400,000 units or more. To economies of these incorporate plant-level potential we include each firm's annual scale in our model, threshold output per assembly plant in Zit. annual 3 plots the average per output Figure the firms in our sam for domestic assembly plant maintained the largest ple.6 It shows that Toyota has in the annual scale, with range of output plant three The to 800,000 vehicles 400,000 per plant. average U.S. reflect this range, fall well below producers of historical their capaci limiting plant ing policy ties to about 200,000 units annually. Figure 3 gives that Toyota and other major Japanese clear evidence
6 For each
ume
(Japanese) 20,000. By ment has differentials small auto nal

(e.g., to
scale by
share
other
producers
such
components,
engines),
are imperfect. arrangements to unit-cost In addition savings that may arise with in the overall size of the firm, it is mean increases at sublev?is of the scale economies ingful to consider
5 Financial annual detail and issues employment of the Daiwa are from for Japanese companies Guide, with Analyst's supplementary from Daiwa Secu obtained directly period data
for the 1965-1976
to motor vehi data are limited The Japanese rities Corporation. of outside within cle production operations Japan; all transplant are excluded. annual The U.S. data are from the companies' Japan Nonautomotive financial (such reports and Compustat. operations subsidiaries) financing the data on value firms, ment include international reported on automotive as have added, been mostly excluded. For U.S. invest and capital employment, as sufficient detail was not operations, in the United States. units
total domes firm these values were obtained by dividing of domestic of vehicles tic production assembly by the number and plant data are from company The annual production plants. and the Japanese Auto annual Reports, reports, Wards Automotive Association. Manufacturers mobile
and U.S. Auto
Producers
Figure3
900,000 i, =
Average Vehicle Outputper Plant
maintained much producers higher plant scale than rivals since the early 1980s.7 their American American have operated smaller assem producers reasons. a belief for several One is that bly plants reduce the likelihood of strikes and dispersed plants A second reason is historical: many other disruptions. all facilities U.S. plants date back to a period when a were built to contain line. Once single assembly it is uneconomic to replace investment is in place, such plants unless their deficiencies become substan tial. A third reason is skill based: the Big Three have been slow to develop for mixed-model capabilities a are to that operate assembly required efficiently volume Annual has model output per higher plant. in the United been much higher States than in Japan; auto makers to ded this has allowed the American to assembly icate each of their plants of a specific and model. Such an approach operations simplifies on the production minimizes the need for flexibility line. In the long run, though, the problem-solving skills gained by the Japanese in running mixed-model cost savings have provided may greater assembly 1982, pp. 119-121). (Schonberger in plant operation between U.S. and This difference the interaction between Japanese producers suggests an and economies of scale. capability organizational are Firms that operate assembly plants larger-scale in to have skills mixed-model also likely assembly. A significant effect for plant size in our SFPF analysis would denote the potential influence of both factors. Capital Another Investment resource
in the 1980s identified higher in rates of investment as and the factor respon equipment primary plant to the United sible for Japan's rise relative States in manufacturing and industries (e.g., Norsworthy and Kuroda 1983, 1992). By Malmquist Jorgenson little if any role to dif the RBV assigns comparison, in fixed investment. ferences Stocks of plant and are an to firm automotive imitate; easy equipment cannot gain competitive increas by merely advantage Thus, the firm's stock of fixed capi ing investment. a strategic resource from the tal cannot be considered our study we therefore take In of the RBV. standpoint the firm's stock of fixed capital as a control variable, to compare but we make calculations its impact with more that of other, potentially strategic factors. Over the three decades the auto of our sample, their and companies substantially upgraded plant as human automated equipment machinery replaced areas of vehicle In Japan, effort in many assembly. rose real capital stock per employee (K/L) steadily, roughly tripling from the mid-1960s through the late the highest investment 1980s.8 Toyota maintained per over its Japanese differential worker, with a growing was rivals. The U.S. pattern, by comparison, irregu lar: capital stock per worker remained for stagnant the late 1970s, but a subse the Big Three through enabled the American firms quent rise in investment
to match, or even exceed, the average Japanese capi
tal stock per worker by the mid-1990s. Thus, our data in American investment noted confirm the deficiency in prior economic studies, but after about a decade, the gap with closed. Moreover, the data Japan was as show that Toyota, commonly the leader regarded was in lean production also the leader in methods, it is important in our study to con investment. Hence in capital input when the trol for differences assessing impact of other factors. Lean Manufacturing Capabilities Our aim is to incorporate measures of lean production areas as impor in three broad capabilities regarded tant in the automotive literature: (1) manufacturing,
a real a stock series for each firm using capital ? = Kt (1 d)Kt_x + inventory equation: capital adjustment as where investment is defined investment, gross gross in the firm's undepreciated the change stock since the pre capital we which depreciation, ceding year, and d is the rate of economic constructed investment 8 We
perpetual deflated
input in the "mass production" is aggregate Economists category capital investment. in rais have long emphasized the role of investment studies econometric ing output per worker. Moreover,
7 While differences such U.S. differences between in plant scale could reflect environmental that Toyota's the two countries, it is notable in Georgetown, has of capacity Kentucky, than twice the capacity of a stan year, more
to be equal to 10%. For Japanese firms, we deflated gross the domestic deflator for non gross using expenditure residential investment Agency. Planning reported by the Economic the GDP deflator for nonresidential fixed For U.S. firms, we used assumed investment for the two purchasing per dollar depreciation from the 1998 were Economic countries converted Values Report of the President. a to common currency using rate for capital goods of 235 yen Our assumption of 10% annual a
power parity exchange in 1982 (OECD 1987). rate is consistent with of the economic
flagship plant 500,000 vehicles per dard U.S. plant.
categories and Wykoff (1981). Results capital stock.
over asset average weighted rates by Hulten depreciation reported were measures for alternative of similar
Lieberman
and Dhawan:
Producers INFORMS
(2) supplier relations, and (3) product design. We start on the manufacturing with capabilities shop floor, where we have the best indicator measure. on the Our proxy for lean production capabilities (WIP) factory floor is the level of work-in-process inventory. Plants with frequent production problems must to avoid disrup hold buffers large inventory a summary tions in output. The level of WIP provides of the firm's manufacturing and capabilities, can a serve as in reductions WIP driver furthermore, for process Lieberman and Demeester improvement. measure as an indicator of validated the WIP (1999) skills in the context of the automo manufacturing tive industry. For a sample of 52 Japanese automotive suppliers and assemblers, they found that WIP reduc tions preceded productivity gains, and lower WIP lev In els were associated with higher labor productivity. this study we follow a similar approach, the using ratio as a measure of factory management WIP/sales skills. in the WIP/sales Figure 4 shows large variation our in ratio among the auto makers study.9 More some pattern of inventory over, all firms exhibit its campaign to cut reduction. Toyota, which began inventories the late 1950s, had the lowest during WIP levels in the 1960s and 1970s and is tied with in later years. Several other Japanese assem Honda and Suzuki, blers, Fuji, Mazda, including Daihatsu, in the had large and widely inventories fluctuating late 1960s and early 1970s. By the late 1970s these were as lean fluctuations eliminated manufacturing in Japan. The one methods became widely adopted is Fuji (Subaru), whose WIP inventory ratio in Japan and began rising again in highest the late 1980s. Figure 4 shows that in the 1960s the were WIP in the top range levels of U.S. producers exception remained of the Japanese, orders of magnitude above Toyota, where remained until the 1980s. Over the they early next decade, however, WIP inventories fell substan States as the Big Three began to tially in the United understand
practices.10
Figure4
WlP/SaIes Ratio
statistic
are important in the automotive indus performance rate time and of cost, try, including development new product and degree of product introduction, Recent studies appeal to consumers. (e.g., Clark and and Cusumano 1991, Nobeoka 1997) doc Fujimoto ument cut devel that have organizational changes time and cost, enabling firms to introduce opment a greater number of new products. These improved were developed in Japan in the late 1970s methods later to the United States. 1980s, spreading a is in which Product difficult domain design firms' evolving gauge capabilities. Comprehensive so we are forced historical data are unavailable, and
use a weak proxy measure. We appraise firm's capa
to to
bilities based on the quality of their product designs as assessed a trade by the staff of Car and Driver, In in annual issues journal. January 1983, beginning a set of "10 best cars" from Car and Driver identified the regular production models sold in the United the criteria used by Car and Driver States.11 While final rather than emphasize product design quality a in underlying capabilities, supplementary analy sis we rates found that the Car and Driver award were correlated with firms' adoption of the "rapid transfer strategy" identified by Nobeoka and design Cusumano and Toyota (1997). Honda, Mazda, Nissan, have accounted for nearly half of all winning vehicles since 1983. Honda, in particular, has been an outlier in these ratings, accounting for more than twice as cars as 10" firm. other many any "top in their capa Automotive assemblers also differ bilities for coordination with component suppliers supe (Helper and Sako 1995, Dyer 1996). Firms with can reduce procurement costs and rior capabilities raise product and Such skills are subtle, quality.
comprehensive measures are unavailable. We can,
and embrace
the Japanese manufacturing
Product design is another area where the Japanese have been influential. Multiple dimensions of design
9 The WIP been adjusted inventories and GM have Fuji, Nissan, (Hughes) to the manufacture related of inventory which have much WIP requirements higher are not the American the U.S. com raw materials be overstated of
to remove
aerospace than auto 10 The
products, assembly. data
for quite comparable In their financial producers. reports, a combined and panies single give figure for WIP causes which the U.S. ratios to inventory, inventory inventory and Japanese in our of sample. inventory, ratio of U.S. WIP/sales lower than in Japan. Census data, which separates Using Lieberman and Asaba (1997) show auto assembly plants may
that
two types the average now be slightly these
however,
11 The "design,"
observe
each
firm's
degree
of backward
selection "ride,"
criteria "value,"
multiple and "driveability,"
include
categories "handling."
such
as
and U.S. Auto
Producers
For many decades, into parts production. integration Ford and GM supported substantial in-house parts whereas the operations, Japanese have manufacturing and close collabora been known for subcontracting in the two tion with suppliers. Thus, auto assemblers countries have maintained different very capabilities with respect to the interface with parts production. a superior strategy in the Integration was considered auto of the early years industry, but as the industry the advantage has shifted to Japanese (Womack et al. 1990, Dyer 1996). style subcontracting Indeed, after the final year in our sample GM and Ford moved toward the subcontracting dramatically model their off internal parts-making by spinning has matured,
operations.
of this study, the measures show remarkable some and of Con convergence. persistence degree on is evident for the data investment vergence per vertical and WIP/sales, worker, suggest integration, imitation of industry best practice. ing the gradual the American had closed the gap By 1997, producers in with fixed moved toward the investment, Japan period and streamlined the pro Japanese outsourcing model, cess flow of their factories. On the other hand, the data on firm and plant scale show little if any con vergence. Figure 2 shows continuing large differences in firm size (although several companies, including and Isuzu, later Daihatsu, Chrysler, Nissan, Mazda, fell under the control of other auto makers). Perhaps are those for the most differences persistent plant scale: Over the period of the sample, the American firms remained locked-in investments by historical and lack of skills in mixed-model assembly, while of the small suffered from many Japanese producers to volume fill efficient-scale inadequate plants. are consistent with Such patterns theoretical per on industry evolution and adoption of best spectives under the RBV. Winter (1987) has pointed practice out that the firm's tangible and intangible assets are to state variables in control theory?they analogous are difficult to change over a short time span, but to management evolve over time in response efforts and environmental influences. Sim (control variables) Dierickx and Cool that (1989) argue ilarly, strategic asset stocks can be changed only gradually. Our data on the speed with which automo evidence provide tive firms have adjusted important scale. ities, and operational resources, capabil
Our measure of vertical is the firm's integration value added as a proportion of sales (V/S). This mea sure allows a test of the relative value of integration versus Wfhile the V/S informative, subcontracting. ratio is not an indicator of the assembler's skills in or with internal (whether parts coordinating suppliers are presumably more than external), which important our tests of the degree of integration, per se. Hence, are coordination quite limited. supplier-assembler To avoid
put changes,
spurious
we use a
correlation
lagged,
with
four-year
short-term
moving
out
aver
age of the V/S ratio, shown in Figure 5. In the early the U.S. assemblers maintained years of our sample, ratio of their Japanese about twice the integration Since the late 1980s, a convergence has counterparts. occurred: Some Japanese assemblers such as Toyota have increased their integration, while the Americans have shed most operations. parts-making Following the spin-offs of parts plants by Ford and GM in the late 1990s (not shown in our data), the degree of ver tical integration has become very similar across all
producers.
4.
Frontier Stochastic Function Model
Production
The Dynamics of Organizational Choices Design in The measures section this reflect organi presented zational design choices of the automotive companies over time. Over the three-decade and their evolution Figure5 Value-Added/Sales (VerticalIntegration)
The usual panel data estimation techniques?fixed and random-effects models?are for inappropriate our study because assume or two-sided symmet they ric deviations from the production frontier. We need a that can capture the fact that firms always technique lie on or beneath the "best-practice" func production tion. Hence, we use the SFPF methodology to estimate structure of Equation the production refine (2), with ments introduced Battese and Coelli (1995) that by as a function to be estimated allow technical efficiency of firm-specific, factors. time-varying we first add standard SFPF Following methodology, a or error to Equation term disturbance that repre (2) sents statistical noise in a typical regression. This term captures
or
0.3
"
* * * ? ' * f( <* * >-*. -* i '
* ??
the effects
of occurrences
such as successful
strikes, etc.,
0.1
- ?GM a Ford Chrysler 70
unsuccessful
4 ? 72
Toyota Nissan Honda 74 76
x Mazda + Isuzu - Suzuki # Daihatsu ? Fuji (Subaru) 78 80 82 84 86 88 90 92 94 96
advertising
campaigns,
outcome. that affect the production It is hypothesized that the realized production of a firm is bounded function by the product of the parametric production
Lieberman
and Dhawan:
the Resource Base of Japanese and U.S. Auto Assessing 1068 Management Science 51(7), pp. 1060-1075, ?2005
Producers INFORMS
error term. This is the the symmetric random stochastic production can then be frontier. The model as written = Yu (3) F(Kit,Lii)TEitev<>, and where the V^s are the independent and identically distributed random which have a errors, symmetric, normal zero and unknown with mean distribution
variance a2.
the deviation from constant represents a positive value of y signifies scale, where
returns to scale.
returns increasing
to
The dependent is value added
variable per
As described earlier, TEfi is a scaling factor, 0 < TE < 1, such that the actual outcome is the production below frontier. We statistically tionalize TE?? as a second error term, Uit, such TEit = ? = e-u?, (4)
where always opera that
or, equivalently, Uit lnTEzi, where by definition, unobserv Uit > 0. The Uits are one-sided, nonnegative able random variables with associated the technical such that, for a given tech inefficiency of production, and levels of inputs, the observed nology output falls short of its potential output. A common assumption in the SFPF literature is that U is distributed as a non truncation of the normal with distribution negative unknown variance a2. In this study we use Battese and Coelli's for parameterizing U as a (1995) method function of additional, variables. firm-specific Now, given a sample of N firms for T time peri frontier production function can be ods, the stochastic as written = Yit F(Kit,LH)ev?e-u<>. (5) In this study we assume that F(-) has a Cobb-Douglas functional that occurs form, with technical progress
a constant rate ?i over time, i.e.,
ity.13 Thus, Equation tical assessment of potential of labor determinants The function relates labor productivity. production to we and labor productivity capital inputs; expect that productivity will rise with investment per worker with the size of the firm (K/L) and possibly (L). In addition, labor productivity will be influenced by other resources and capabilities of the firm, as repre sented by the factors in Uit. and Coelli Battese technical ineffi (1995) specify a as effects function of time LZfis ciency firm-specific, as factors varying following: U,t = Zu8 + Wu, (8)
in this transformed model or labor employee, productiv as a statis (7) can be viewed
where such variables, Zit is a vector of explanatory as those collected in our study. Here, 8 is a vector of unknown to be estimated and Wits are parameters random variables.14 unobservable The technical effi
ciency (TE)of the /th firm in the tth year then is

TEU = exp(-Zlt8-Wit). (9)
at
F(Kit,Llt) = e*'l$L%. (6)

The time trend reflects the potential outward move ment or growth in the frontier after for the controlling factors that can be observed in the data.12 The stochas can be written tic frontier specification in per capita terms by combining and (5) (6), taking log Equations arithms, and dividing by labor, as
are predicted The technical efficiencies using the con ditional of the com expectations exp(?Uit), given error term the of stochastic frontier. Following posed the suggested parameterization by Battese and Coelli, = a2 we define and estimate + cr2 and y = a2/a2, of vector S by maximum-likelihood and estima a2, y, ?, tion (MLE) methods.15
the firm's sales during the fiscal year, minus equals of purchased materials and services. This is equivalent to the sum of all payments to labor and capital, indirect taxes. plus For the Japanese we used value-added companies, figures provided the costs puted added used Securities by Daiwa value added was computed Corporation. by summing by For the the U.S. factor companies, we Real com value payments. nominal value added dividing by the index for motor vehicles. For Japan, we 13 Value added
ln(Y/L)?
= txt + e\n{K/L)it + yln(L)i( + Vit- Uit,

value added
domestic
(7)
where Y/L represents is capital stock per
K/L per employee, L is the number of employee, are and and the random variables Vit Uit employees, described above. The coefficient, is equal 0, which to ?lf is the elasticity of output with respect to cap ital. The coefficient, is equal to ?x + ?2 ? 1, y, which
12 We chose the Cobb-Douglas it is a com because specification can be estimated functional form, which using single to give methods consistent estimates parameter (see specification, but the
for transport price deflator equipment Indexes Annual, the Bank of Japan. For the published by United States, we used the Bureau of Labor Statistics price producer cars. Yen and dollar index for passenger were values converted a to a common currency power using purchasing parity exchange rate of 171 yen per dollar on OECD in 1982, based estimates for motor vehicles rates for all other years are (OECD 1987). Exchange from Price defined by the 1982 rate and the price deflators. 14 are assumed which to be W/fs, distributed, independently obtained of the normal with mean distribution by truncation and unknown that Uit is nonnegative such variance, a2, W?>-Zit8). 15 in this study were The estimates obtained by programming Battese and Coelli function the maxlik (1995) likelihood using tine version 3.1.3. in the Gauss econometric package. are zero (i.e.,
producer the domestic
price wholesale
monly
used
equation et al. 1966). We also tried a Zellner trans-log function likelihood failed to converge.
the rou
the Resource Base of Japanese and Dhawan: Assessing INFORMS Science 51(7), pp. 1060-1075, ?2005
and US.
Auto
Producers
1069
Table 1
Variable
Data Summary
Mean Std. dev.
Units millions of 1982 yen count from 1965 millions of 1982 yen
thousands of employees ratio ratio number of citations
Sources See footnotes See footnotes See footnote See footnotes See footnote See footnote See footnote See footnote 5 and 13. 5 and 8. 5. 5, 9, and 10. 5. 6. 6. 6.
Valueadded per employee (Y/L) 17.1 Time(f) Capitalstock per employee (K/L) Numberof employees (L) WIP/sales ratio (W/S) Value-added/salesratio (V/S) Design quality(CD) Numberof vehicles produced (Q) Volume per plant (Q/N) Cumulative output(? Q)
Note. Data sample
8.6 9.0 16.4 144 229 0.033 0.243 0.66 0.85 1.68 168 281 44.97
4.1 7.7 0.029 0.111 1.52 58.85
Car and Driver
millions of vehicles thousandsof vehicles millions of vehicles
includes 336 observations.
within incorporate measures the stochastic of ciency component We Utl = fi0+ S1ln(W/S)i/_1
the technical ineffi frontier as follows: + S3(CD)it
were
sion analysis. All regressions
+ 52ln(V/S){t_1,4
+ 54ln(Q)jf + S5ln(Q/N)I, + S6\nr?Q)u + Wu, (10)

to sales ratio; is the WIP W/Sit inventory is the average of the value V/Sit_14 four-year moving aver added to sales ratio; CDit is a two-year moving the of citations awarded the number of age design firm by Car and Driver;16 Qit is the total number of mar in motor vehicles produced firm its home the by out ket during year t; Q/Nit is the average vehicle in is and the t; J^Qit year put per assembly plant firm's historical cumulative domestic vehicle produc tion through the start of year t.We include the latter measure as a general and index of firm experience in test the We Qit inefficiency learning (Argote 1999). where term to verify that the measured effects of plant scale, and cumulative Q/Nit, output, J2Qw are n?t simply due to correlation with annual vehicle output. A positive value of the 8 coefficient associated with as of these variables that the level of that indicates any variable
goes up
to natural for the regres logarithms Table 2 reports the estimation results. ratio in the inef include the WIP/sales as a proxy for ficiency term, given its prior validation In lean manufacturing capabilities. light of the cor we add other relation among some of the measures, converted
to the inefficiency term in various combina of the results across firms tions.17 To verify robustness in the sample, the last two regressions show the esti are mates when Toyota and the American producers omitted. in Table 2 relate to the The first three parameters is specified as a func frontier. The frontier production variables to be and is assumed time trend, ?jl, is pos itive and significant; its value implies that the fron at an average tier level of efficiency rate increased of about 2.5% per year. This can roughly be inter preted as the rate of growth of total factor productiv in the auto ity associated with best-practice operation The coefficient, 6, identifies capital elasticity industry. a and associa statistically quantitatively significant tion of capital and labor inputs rate. The shifting at a constant tion between and higher greater capital investment labor productivity. The returns to scale parameter, y, is about 0.09, which suggests significant increasing returns to scale in the production function (i.e., a 10% increase in firm size was associated with an increase of 0.9% in output per worker). The estimated param eters of the production frontier change only slightly with different of the inefficiency model. specifications are of in the inefficiency The coefficients model in this study. The WIP/sales coeffi in all and 8lf regressions generally that higher levels of suggesting highly significant, WIP were associated with lower levels of efficiency, as expected. Thus, the results point to the importance skills on the factory floor. of lean manufacturing prime cient, interest is positive
some com likelihood function failed to converge with to include We were, for example, unable of parameters. in the same regression. total vehicle and output per plant output differ only by the count of assembly (These measures plants.) 17 The model
goes up,
and vice
the level of technical

versa. For example, a
inefficiency also
positive coef
ficient for W/S with the level tially, a positive
rises implies that technical inefficiency of WIP We expect, poten inventory. for W/S and negative coef coefficient for V/S is
and J2 Q- The sign ficients for CD, Q/N, not clear from a theoretical standpoint.
5.
Results
Table 1 provides statistics on the measures summary in the study. With the exception of the time trend and the count measure the values of design quality,
16 Given Driver that all CDit values included ratings, we early years, are 0 prior a separate and to the start of variable the Car and (set equal
binations
to 1 for these
dummy 0 otherwise).
Lieberman
and Dhawan:
1070
Table 2 Parameter Estimates of the Stochastic Frontier Model*
Assessing
the Resource Management
Base of Japanese and U.S. Auto Science 51(7), pp. 1060-1075, ?2005
Producers INFORMS
Stochastic Time /i
frontier 0.024 0.027 0.023 0.025 0.028 0.025 0.029 0.029 0.043
(0.003) ratio 0 Capital/labor Employees y model Inefficiency Constant 80 WIP/sales ratio (lagged)^
Value-added/sales 0.364
(0.003)
0.321
(0.003)
0.388
(0.003)
0.317
(0.003)
0.265
(0.003)
0.347
(0.003)
0.260
(0.004)
0.231
(0.004)
0.133
(0.051)
0.090
(0.060)
0.105
(0.062)
0.089
(0.064)
0.108
(0.061)
0.114
(0.055)
0.098
(0.061)
0.113
(0.063)
0.115
(0.074)
0.126
(0.011) 3.029 (0.462)

0.123
(0.013)
0.747
(0.014)
1.216
(0.014)
1.182
(0.014) 3.316 (0.442)

0.062
(0.011)
1.117
(0.014) 3.366 (0.449)

0.078
(0.014) 3.705 (0.522)

0.092
(0.016) 1.925 (0.425)

0.189
(0.214)
0.186
(0.268)
0.192
(0.129)
0.166
(0.109)
0.201
(0.030)
(0.024)
(0.028)
(0.030)
0.105
(0.034)
0.196
(0.027)
(0.036)
0.161
(0.045)
0.172
(0.045)
-0.194
ratio(lagged) 82 Design quality 83

Number of vehicles -0.184 -0.011
(0.072)
(0.071)
0.035
(0.076)
0.035
(0.084)
0.040
(0.022) produced 84 Volume per plant 85 Cumulative production 56

Variance parameters 0.052 0.054 0.055 0.054 0.050 0.055
(0.023)
(0.027)
(0.109) 0.066 (0.025)
(0.021)
-0.197 -0.202 -0.230 -0.096
(0.040)
0.022
(0.039) (0.012)
(0.039)
(0.045)
(0.035)
0.050
0.054
0.043
(0.006)
0.563
(0.006)
0.700
(0.007)
0.599
(0.006)
0.633
(0.005)
0.577
(0.169) function Log-likelihood

Firms in sample No. of observations 45.03
(0.149)
33.28
(0.204)
31.95
(0.184)
32.78
(0.162)
48.62
(0.006) 0.649 (0.164)

32.00
(0.006)
0.589
(0.007)
0.711
(0.006)
0.682
(0.154)
49.75
(0.118)
47.65 Exc. Toyota
(0.156)
58.08 Exc. US Big-3
All 336
All 336
All 336
All 336
All 336
All 336
All 336
304
240
measure. Numbers in *AII and inthe inefficiency the explanatory variables inthe stochastic frontier model are in logarithms, except for the design quality
parentheses are estimated standard errors. Coefficients in bold are significant at the 0.05 level.
to plant scale. The strong result relates for average output per assembly plant, 85, and highly that effi significant, implying more firms that vehi for produced higher cles per plant. As discussed earlier, this finding may and denote the joint influence of scale economies with mixed associated manufacturing capabilities are able model assembly. Firms with such capabilities to operate with lower levels of WIP inventory, which account the reduced for coefficient forWIP/sales may when volume is included. per plant 2 includes the cumulative number of Regression curve the vehicles for proxy produced, learning in the early years effects. Japanese producers began of the sample with low levels of cumulative output sur but experienced rapid growth. Toyota ultimately the cumulative but other of output Chrysler, passed wise the relative experience rankings are fairly stable, with GM remaining by far the most "experienced" firm. The coefficient of the cumulative output vari not the is able, 86, suggesting statistically significant, of any simple connection between cumula absence tive output and efficiency for the firms in our sam is positive, the coefficient the of Moreover, sign ple. Another coefficient is negative ciency was
that more firms signifying "experienced" (typically the U.S. Big Three) were less productive. Experiments that allowed the stock of production to experience over reverse to time failed result. We this depreciate that firm-level conclude cumulative output does not serve as an effective proxy for organizational learning in our sample. among the automotive companies 3 includes the firm's total vehicle pro Regression in the observation duction of the year as a component Q, to con inefficiency model. We test this measure, firm that the results for plant scale, Q/N, and cumu are not to lative output, ? Qf correlation simply due In addition, with the firm's annual vehicle output. at indicator of scale economies Q serves as a potential can as an firm to the alternative the be viewed level; it test for scale economies denoted y by the parameter in the production frontier. In regression 3, the associ ated coefficient, S4, has the expected negative sign but we is insignificant. evidence of find firm-level Thus, economies of scale in the production function (y > 0) but not in the inefficiency term.18
18 If the production function if y is omitted from is specified the model), as constant the coefficient returns to scale
(i.e.,
for vehicles
and U.S.
Auto
Producers
and relating to supplier integration or insignificant results. give weak design product 4, 5, and 7 show that the value-added/ Regressions measure into parts pro of backward sales integration with vol included when duction becomes significant ume per plant. The positive that of 82 implies sign was more into associ parts production integration ated with greater inefficiency. This is consistent with views on the advantages of subcontracting. However, and could the result is not robust across specifications tends reflect the fact that labor productivity simply to be lower in parts production than in assembly. from Car of design The measure collected quality in 6 and Driver is statistically regressions insignificant no is there and 7 and carries the wrong sign. Thus, awards had evidence that firms with more design levels of efficiency.19 higher To summarize the main findings in Table 2, our esti mates of the production function show that greater was investment with labor associated capital higher as of Moderate economies productivity, expected. at the firm level. The best-practice scale are observed as the frontier gradually shifted outward, presumably result of technical progress not captured by factors in our model. estimates of the inefficiency Furthermore, at the of scale economies model show the presence a WIP and connection between inventory plant level, that and efficiency. Less conclusive evidence suggests were firms with more less effi vertical integration cient. We find no indication of a general "learning curve" at the firm level, and no connection between firm efficiency of and our Car and Driver measure design quality. a series of robustness to We performed checks be examine that our results might the possibility in our contaminated by unobserved heterogeneity to A standard control such hetero for way sample. is to include a fixed-effect parameter for each geneity our SFPF model firm. Unfortunately, failed to con a full set of firm dummies was included verge when in either the production function or the inefficiency we term. As an alternative, estimated the model to with from the sample, selected companies omitted assess whether to these firms specific heterogeneity our find distort our estimates. For example, might ings could be dominated by the strong performance of Toyota (so that any measure correlated with Toyota or by to have would value), appear explanatory
produced, economies from 19 We 84, becomes statistically significant. show up in the inefficiency model function. obtained scale Thus, firm-level if they are excluded
Our measures
in Table 2 effects. The final columns country-specific our with for estimates of model observations provide or the three American 8) compa Toyota (regression nies of regres 9) excluded. (regression Comparison sions 7 and 8 reveals that the omission of Toyota has or their coefficients very little impact on the estimated levels. Hence, it is clear that heterogene significance to not is for our find Toyota ity relating responsible the Omission American of three firms leads to ings. in the estimated greater changes parameters, although When the the general pattern is largely unaffected. are American the coefficient of excluded, producers in magnitude and significance, increases WIP/sales measure that the WIP/sales is corre thus confirming in Japan. The coeffi lated with productive efficiency in cient of volume smaller per plant becomes much 9 (although it remains statistically regression signifi that the productivity differential we cant), indicating to plant scale exists mostly have attributed between in Japan and the United States. The coef producers ratio changes ficient of the value-added /sales sign in regression and becomes 9, suggesting insignificant that our findings per integration relating to vertical com tain to differences and between U.S. Japanese a country and thus could be simply effect. panies, The design quality measure achieves statistical signif 9 but shows icance in regression the wrong sign, a result that is likely spurious given the deficiencies of
this measure.
A (avail this approach further, Appendix Carrying at able as an electronic companion http://mansci. reports a set of pubs.informs.org/ecompanion.html) 11 regressions based on Equation (1) of Table 2, where each producer is sequentially excluded from the sam The estimated remain coefficients ple. fairly stable across these regressions. there is no evidence Hence, that any single firm exerts unusual influence on our at http:// B estimates. Moreover, (online Appendix a mansci.pubs.informs.org/ecompanion.html) gives set of OLS regressions, similar to Equations (1) and (7) of Table 2, but with the inclusion firm fixed effects. in Table 2 are main Most of the patterns observed tained in the OLS results when fixed effects are added. are a reduction in the the notable Among changes WIP/sales that most of the coefficient (suggesting measure across value is of this explanatory producers, rather than within firms over time), and an increase in
the production also from annual
measure results for a quality insignificant issues of Consumer Reports, which gives vehi an on reliability and frequency of repair. cles ratings with emphasis the proportion of models from each manufacturer that We recorded obtained received a "recommended" rating in each year.
of volume per plant (which may pick of short-run output fluctuations when the cross-sectional variance is removed). the Also, ratio changes sign, again implying value-added/sales are sensitive that our findings for vertical integration to the model such informative, (While specification. in compari OLS estimates suffer various deficiencies son to the SFPF model, which we highlight in ongoing work.)
the coefficient up the effects
Lieberman
and Dhawan:
the Resource Base of Japanese and U.S. Auto Assessing 1072 Management Science 51(7), pp. 1060-1075, ?2005
Producers INFORMS
6.
Explaining Performance
Differences Among
in Firms
worker.
We now
from Table 2 to draw com apply firms. One is to account among parisons challenge for the substantial in performance differences that have existed between in the two the largest producers the estimates countries, Toyota these companies
can be made.
GM's assembly plants had about one-fourth the average volume of Toyota's. Within its plants, GM held about 10 times more WIP as a frac inventory, tion of sales. GM also maintained more substantially
and GM. We present calculations for to show how interfirm comparisons
is a summary measure of firms' efficiency 6 shows the estimated techni performance. Figure cal efficiency in each year, based on of producers are 5. values the conditional (The plotted regression estimates of TE given by formula (A.10) in Battese and Coelli The of the graph corre 1993.) top margin to the frontier, which was sponds industry's efficiency a at rate of about 2.8% per year, according increasing to the value of fi in regression 5. The TE estimates Technical in Figure 6 suggest that Toyota has operated close to the frontier since the late 1970s, whereas GM has been firms typically falling away from the frontier. Other lie in between. that the TE estimates in Figure 6 (Note exclude the effects of firm-level scale economies and are in which the pro capital investment, incorporated duction function.) the data values and Probing deeper, Table 3 utilizes estimated of the model to draw compar coefficients isons among firms. The calculations a break provide down of the labor productivity between differential on means GM and Toyota, based of the relevant over the 1965-1997 for the two producers The first of the table shows the extent to part period. which GM and Toyota differed along the dimensions in this study. On average, GM's output considered (value added) per worker was only 62% of Toyota's. as GM had more than 13 times as many employees as much investment per Toyota, but with only 79% variables Figure6
?
into parts production: internal integration revenue 46% of final sales for operations represented as with 18% for GM, compared Toyota. the logarithm of these ratios and multi Taking it coefficients, by the applicable plying regression an estimate is possible to make of the contribution of each factor in explaining the overall differential in output per worker. The results of these calcula tions are shown in the final columns of Table 3.20 labor productivity between differential GM and on in terms. ?0.48 Based the coef Toyota equals log ficients 1 of Table from regression this differ 1, ential can be attributed about equally to Toyota's (2.35 x superior positions relating to WIP inventory = x -0.1229 and -0.29) per plant output (-1.27 0.1840 = -0.23), with an additional small effect due to = investment (-0.24 x 0.3655 -0.09). Toyota's higher Our estimates were that these suggest disadvantages of scale at partly offset by GM's greater economies the firm level (2.62 x 0.0897 = 0.24).21 Thus, the four factors in combination for about three may account fourths of the labor differ (=0.37/0.48) productivity ential between GM and Toyota. A similar calculation, the effect of vertical is shown including integration, in the last column of Table 3, based on the coefficients from regression includes all the major vari 5, which ables except design quality. Some estimates, such as the impact of WIP/sales, between change magnitude the columns, to the underlying revealing sensitivity of the model. specification similar comparisons between Figure 7 illustrates and all in other firms the Toyota sample, using the coefficient in regression estimates 5. Over the 1965-1997 period, Toyota enjoyed substantial advan relative to most producers. tages in labor productivity These advantages were based on many factors consid ered
20 Note linear
backward
The
Technical Efficiencyby FirmandYear

A ? GM Ford ?a? x Fuji Chrysler ? Nissan o Honda Toyota * Mazda + Diahatsu + isuzu
in this study:
that
capital
investment,
firm and plant
the regression, of Equations (7) and (9), is composed for the variables of interest. The difference logarithms between GM's and Toyota's output per worker, log Y/L, is equal to in the (logged) differences variables multiplied explanatory by their estimated the in difference coefficients, respective regression plus errors (or the In Table 3, we con prediction portion). unexplained in the data into logarithms, in the values take differences for GM and Toyota, and plug these differences into the regression equation. 21 in firm scale is as GM's to be overestimated, advantage likely the calculation GM's worldwide with the compares employment domestic of Toyota. it is possible that scale Also, employment economies our sample, be may which diminishing would also Otherwise, advantages scale over lead of the range of firm to an overestimate in Table firms the two sizes of in the vert
ii. 66
. 68 70 72 74 76 78 80 82 84 86 88 90 92 94 96
GM-Toyota large but versus
differential.
the estimates
offsetting level. plant
at the
2 imply firm
the Resource Base of Japanese and Dhawan: Assessing INFORMS Science 51(7), pp. 1060-1075, ?2005
and U.S. Auto
Producers
1073
Table 3
ComparisonCalculation GM-Toyota Averagedatavalues (1965-1997)
Dependent
variable
GM 43.8
Toyota
70.7
GM/Toyota
0.62
log(GM/Toyota)
-0.48
Outputperworker (Y/L)*:
to be explained) (=differential of factoron \oq(Y/L)* Impact
Explanatory Resource
factors levels 74.5 94.5 0.79 13.8 -0.24 2.62
Based on reg. 1
Based on reg.5
Capitalperworker (K/L)** Numberof employees (L)

Capability measures
-0.09 0.24
-0.06 0.30
754,327
0.071 0.461
54,846
0.007 0.182
WIP/sales
Value added/sales
10.4 2.5 0.28
2.35 0.93 -1.27
-0.29 -0.23 -0.37
-0.15 -0.18 -0.25 -0.34
Vehicles/plant
Total:
170,687
607,418
= log(GM/Toyota) x regressioncoefficient. of factor *lmpact

**Thousands of 1982 dollars.
The calculations suggest that Toyota's can to our measure linked be greatest advantages at the of plant scale. Toyota lacked scale economies firm level relative to GM and Ford, but enjoyed scale at both the firm and plant level relative to advantages
Japanese rivals.
scale, and WIP.
7.
Conclusions
The contributions of our study are both methodolog ical and substantive. the perspective By combining of the resource-based the methods of pro view with an we duction have outlined economics, approach the RBV operational. the SFPF for making Applying to model of Battese and Coelli (1995) public data on we 11 automotive have firms' identified companies, to relative the fron positions industry best-practice we have shown how the parameter tier. Furthermore, estimates shed light on potential deter of the model in the auto industry. The minants of firm performance over pre Battese and Coelli model offers advantages in that it allows for dynamic per vious SFPF methods in a single formance and is estimated comparisons
stage.
we have pro Based on the coefficient estimates, on sources vided calculations of interfirm the rough in performance. Our estimates differences suggest auto makers that productivity differentials among in orga have been mostly the result of differences and scale. In contrast with in nization earlier work that organiza the economics literature, we conclude tion and scale are much more than capital important in accounting dif investment for labor productivity ferences in the automotive industry. our quantitative While suggestive, findings must be in this study caution. with The estimates interpreted limita may be biased, perhaps given substantially, tions of the data. Important of categories capabilities from the model. Those measures that may be omitted we have included in our analysis are imperfect prox serve as valid indicators, but ies. Ideally, our measures are based on correlations rather the results ultimately
than causality. In areas where our proxies are weak,
a broad set of influences on com By considering our to adds pany performance, analysis perspective on work the have automotive We iden prior industry. on the man tified benefits associated with capabilities as as well economies of scale shop floor, ufacturing at the firm and plant level. Our estimates of the lat ter are likely to incorporate the value of capabilities run to effectively for mixed-model needed assembly our in the larger automotive Weaker evidence plants. that auto producers with higher levels study suggests of vertical integration have been less efficient. Other in our analy factors prove insignificant statistically of the deficiencies of our proxy sis, perhaps because
measures.
low correlation with the desired constructs biases our if our statistical estimates toward zero. Alternatively, are correlated with other important measures factors that we have failed to recognize, the estimated coef ficients may be inflated. Moreover, serial correlation to of statistical may cause our estimates significance
be overstated.
our findings point to the these limitations, Despite as a source of effectiveness importance operational in the of competitive automotive indus advantage most has been efficient the try. Toyota long industry's and has increased its lead over time. Porter producer that operational effectiveness alone is com sustainable for a firm to achieve mar a must firm also have the petitive advantage; it from competitors. While that insulates ket position this is true in industries where improve operational can be easily imitated, the differentials we have ments (1996) argues not sufficient
Lieberman
and Dhawan:
Producers INFORMS
Figure7
Estimated Impact of Factors on Value-Addedper Employee, Relative to Toyota

-40% -30%-20%-10% 0% 10% 20% 30% 40%
in SFPF models may allow a richer set of trade-offs be explored. to this paper is available An online appendix http://mansci.pubs.informs.org/ecompanion.html. Acknowledgments
The authors thank Rebecca Henderson, David
to at
Capital per worker (K/L)
Hoopes,
Chris
Postrel,
Knittel,
the
Dan
Levinthal,
editor, two
Richard
anonymous
Makadok,
referees,
Steve
and
associate
seminar participants
Austin, and Number of Employees (firm-level scale economies)
at CMU,
INSEAD,
and for the
SMU, USC, UT
annual Harvard comments.
University, Washington Stanford conferences strategy
helpful
The UCLA Center for International Business Education Research provided financial support.
and
References
WIP/Sales (shop floor manufacturing) P. Schmidt. 1977. D., C. A. Knox Lovell, Aigner, of stochastic estimation frontier production /. Econometrics Amit, 6 21-37. organizational and Formulation function and models.
assets 1993. Strategic and R., R Schoemaker. rent. Strategic Management J. 14 33-46. L. 1999. Organizational Learning: Creating, Argote, Kluwer, Boston, MA. Transferring Knowledge.
Retaining
VA/Sales (vertical integration)
GM
DFord Chrysler B Nissan Mazda D Honda
factor markets: J. B. 1986. Strategic Barney, Expectations, Sei. 32 1231-1241. business strategy. Management and sustained J. B. 1991. Firm resources competitive Barney, /. Management 17 99-120.
luck, and advan
Vehicles/Plant (plant-level scale economies)
tage. frontier G. E., T. J. Coelli. 1993. A stochastic Battese, produc a model for technical tion function incorporating inefficiency of Econometrics, Uni effects. Working 69, Department paper of New Australia. versity England, G. Battese, effects E., T. J. Coelli. in stochastic 1995. A model frontier production 20 325-332. of technical functions inefficiency for panel New York
Daihatsu D Fuji (Subaru) Isuzu Suzuki
data. Empirical Econom. Car and Driver. 1983-1998. Diamandis (January R. E. Caves, issues). 1992. Industrial
communications.
Note.
Reg. 5 coefficients
and 1965-1997
avg. data values.
Efficiency
in Six Nations.
MIT
Press,
MA. Cambridge, R. E., D. R. Barton. Caves,
industry have been sus tained for long periods. Many years or decades have to achieve been required the imitation of superior scale and organizational skills. Consequently, lagging identified in the automotive to industry best converged only slowly at while like Toyota firms all), (if practice stronger have made continual advances, thereby maintaining or their lead. expanding Such findings raise questions about the relative im versus market of effectiveness portance operational firms as sources of competitive position in the automotive firms industry tioned as broad-line producers, mon advantage. Major are similarly posi and there is com have
Industries. Clark,
MIT
Press,
K., T. Fujimoto.
egy, Organization HBS Press, Boston, Coelli, T.
in U.S. Manufacturing 1990. Efficiency MA. Cambridge, 1991. Product Development Strat Performance: and Management in the World Auto Industry. MA.
to 1998. An Introduction J., D. S. P. Rao, G. E. Battese. Kluwer, Boston, MA. Analysis. Efficiency and Productivity Daiwa Securities Research Institute Guide. (annual issues). Analyst's Tokyo, Dierickx, Japan. 1989. Asset stock I., K. Cool. of competitive advantage. tainability 1504-1511. S. Rajiv. 1999. S., O. Narasimhan, Is marketing capability accumulation Management in high Success critical? Marketing Keiretsu competitive a and Sei. sus 35
Dutta,
technol Sei. 18
ogy markets: 547-568. Dyer,
the best practice. Such uni agreement regarding enhance the role of operational effective may formity ness. Market more to is be positioning impor likely tant in industries in that support diverse competitors our In niches. of this specialized study specification the production frontier has incorporated only limited dimensions of firms' positioning. Future refinements
matter? J. H. 1996. Does governance as sources asset specificity of Japanese Sei. 7 649-666. Organ.
alliances
and
advantage.
Dyer,
2000. Creating and managing J.H., K. Nobeoka. mance The Toyota network: knowledge-sharing J. 21 345-367. Management
high perfor case. Strategic
a T. 1999a. The Evolution Fujimoto, System at Toyota. of Manufacturing Oxford York. New Press, University T. 1999b. Reinterpreting the resource-capability view of Fujimoto, of the the firm: A case of the development-production systems
the Resource Base of Japanese and Dhawan: Assessing INFORMS 1075 Science 51(7), pp. 1060-1075, ?2005
and U.S. Auto
Producers
A. Chandler, P. Hagstrom, O. Solvell, auto-makers. Japanese Role of Technology, Firm?The eds. The Dynamic Strategy, Orga York. and Regions. Oxford Press, New nization, University Greene, P. Schimidt, Microeconomics. Helfat, C. W. H. 1997. eds. production Handbook of Applied MA. Blackwell, Maiden, and asset The Frontier functions. M. H. Econometrics, Pesaran, Vol. II:
MacDonald,
competition Makadok,
2004. G., M. D. Ryall. determine whether
How and do value creation a firm value? appropriates of the resource-based and creation. Strategic Manage estimation composed from error.
Sei. 50 1319-1333. Management a synthesis R. 2001. Toward views of rent dynamic-capability ment J. 22 387-401.
1997.
Know-how
dynamic Management
accumulation: capability J. 18 339-360. 1995. they Supplier converging?
complementarity case of R&D.
and Strategic
Meeusen,
1977. Efficient W., J. Van der Broeck. functions with Cobb-Douglas production Internat. Econom. Rev. 18 435^444. K., M. growth: A. Cusumano. benefits of 1997. Multiproject rapid design
S. R., M. Sako. Helper, United States: Are 36(3) Henderson, 77-84.
relations
in Japan and the Sloan Management Rev.
Nobeoka, sales product
The
and strategy in new transfer
1994. Measuring R., I. Cockburn. competence? Explor in pharmaceutical research. Strategic Manage ing firm effects ment J. 15 63-84.
Norsworthy, and productivity Amer. Novak, Econom.
development. J. R., D. H.
Strategic Management 1983. Malmquist.
J. 18 169-186.
in Japanese growth Rev. 73 947-967.
Input measurement and U.S. manufacturing.
is there a D. G., T. L. Madsen, G. Walker. 2003. Why Hoopes, a hetero view? Toward resource-based theory of competitive 24 889-902. J. Management geneity. Strategic Hulten, C. R., F. C. Wykoff. 1981. The measurement of economic and the C. R. Hulten, ed. Depreciation, Inflation, The Urban Institute Income from Capital. Press, D.C. K. Shaw. 1999. The effects of human resource man
com 2001. Sourcing Product S., S. D. Eppinger. by design: and the supply chain. Management Sei. 47 189-204. plexity of Economics and Statistics. 1987. Purchasing OECD, Department Power Parities and Real Expenditures, 1985. OECD, Paris, France. Peteraf, A Peteraf, M. 1993. The cornerstones of competitive advantage: resource-based M., J. Barney. Decision view. J. 14 179-191. Strategic Management 2003. Unraveling the resource-based tangle. Econom. 24 309-323. Harvard Business Rev. 74(6) is strategy?
depreciation. Taxation of Washington, Ichniowski, agement C,
Managerial Porter, M. E. 61-78.
1996. What
on economic An international systems performance: Sei. 45 of U.S. and Japanese comparison plants. Management 704-721. C, K. Shaw. 2003. Beyond of the value of complementary practices. D. W., M. /. Econom. Kuroda. Perspect. incentive human esti pay: Insiders' resource manage and States, international 1960-1985.
Ichniowski, mates ment Jorgenson,
17 155-180.
C. F. 1971. Economies of Scale in Manufacturing Industry. UK. Press, Cambridge, University Cambridge "view" a useful Priem, R. L., J. E. Butler. 2001. Is the resource-based Acad. Manage for research? perspective strategic management ment Rev. 26 22^0. Pratten, R. P. 1987. Theory, Rumelt, strategy, and entrepreneurship. ed. The Competitive Challenge: Strategies for Industrial and Renewal. MA. Ballinger, Cambridge, D. Teece, Innovation
in Japan and competitiveness Econom. Stud. Quart. 43 313-325. Knott, A. M., H. Posen. ment J. 26 617-641. Kumbhakar, Cambridge Lieberman, M. S. C, 2005.
1992. Productivity the United
Is failure
good? Stochastic UK.
Strategic Frontier
Manage
C. A. K. Lovell. Press,
2000.
Analysis.
R. J. 1982. Japanese Manufacturing Techniques: Nine Hid Schonberger, den Lessons in Simplicity. York. Free Press, New as strategic assets. S. G. 1987. Knowledge and competence Winter, D. Teece, ed. The Competitive Strategies for Industrial Challenge: Innovation and Renewal. MA. Ballinger, Cambridge, D. T. D. Roos. The 1990. Machine P., Womack, J. Jones, the World: The Story of Lean Production. Rawson New York. Zellner, A., J. Kementa, tion of Cobb-Douglas 784-795. J. Dreze. 1966. that Changed Associates,
University
Cambridge,
reduction and produc 1997. Inventory B., S. Asaba. of Japanese and U.S. automotive tivity growth: A comparison sectors. Managerial Decision Econom. 18 73-85. M. B., L. Demeester. 1999. Inventory in the Japanese reduction automotive and pro sector.
Lieberman, ductivity
growth: Linkages Sei. 45 466-485. Management
production
and estima Specification Econometrica 34 functions.

Assessing The Resource Base of Japanese and U.S. Auto Producers A Stochastic Frontier

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Assessing The Resource Base of Japanese and U.S. Auto Producers A Stochastic Frontier

Uploaded by

Copyright:

Available Formats

Assessing the Resource Base of Japanese and U.S.

SCIENCE MANAGEMENT M3H.

ISSN 0025-19091 EissN 1526-55011051510711060 ? 2005 INFORMS

of California, Los Angeles, University marvin.lieberman@anderson.ucla.edu

proficiency Toyota's Key

producers. frontier; production function; automotive

July 26, 2001. This paper was with

of the RBV with

and U.S. Auto

ing from these estimates, we make summarizes isons. A final section

interfirm compar and the findings

ProductionFrontier FirmsRelative to Industry

2. The RBV Within

development technical good literature offers

and models, In addition

(1997) provides the econometrics

Resource-Based In parallel with

and U.S. Auto

that excludes as market

deter and power

those of General in recent decades

income operating GM (6.0%) than

enormously profitable 1983, Toyota's operat auto makers Japanese

Toyota Nissan Honda 78 80

+ Isuzu ? Mazda - Suzuki * Daihatsu < Fuji (Subaru) 82 84 86 88 90 92 94 96

(Japanese) 20,000. By ment has differentials small auto nal

for the 1965-1976

and U.S. Auto

Average Vehicle Outputper Plant

flagship plant 500,000 vehicles per dard U.S. plant.

categories and Wykoff (1981). Results capital stock.

the Japanese manufacturing

aerospace than auto 10 The

products, assembly. data

two types the average now be slightly these

multiple and "driveability,"

and U.S. Auto

Frontier Stochastic Function Model

* * * ? ' * f( <* * >-*. -* i '

- ?GM a Ford Chrysler 70

Toyota Nissan Honda 74 76

x Mazda + Isuzu - Suzuki # Daihatsu ? Fuji (Subaru) 78 80 82 84 86 88 90 92 94 96

The dependent is value added

where always opera that

ciency (TE)of the /th firm in the tth year then is

F(Kit,Llt) = e*'l$L%. (6)

= txt + e\n{K/L)it + yln(L)i( + Vit- Uit,

where Y/L represents is capital stock per

producer the domestic

4.1 7.7 0.029 0.111 1.52 58.85

Car and Driver

millions of vehicles thousandsof vehicles millions of vehicles

includes 336 observations.

the technical ineffi frontier as follows: + S3(CD)it

sion analysis. All regressions

+ 54ln(Q)jf + S5ln(Q/N)I, + S6\nr?Q)u + Wu, (10)

the level of technical

ficient for W/S with the level tially, a positive

the Resource Management

(0.011) 3.029 (0.462)

(0.014) 3.316 (0.442)

(0.014) 3.366 (0.449)

(0.014) 3.705 (0.522)

* * * ? ' * f( <* * >-. - i '