Early in the emergence of a high-technology industry, will firms achieve higher innovative performance by appropriating technological knowledge for themselves or by sharing that knowledge with other firms in the global innovation system? 

Conventional wisdom holds that innovating firms that appropriate scientific and technological knowledge for the sole use of their own industrial research staff will achieve higher innovative performance than firms that share their technological knowledge with competitors. This conclusion finds basis in reasonable applications of current management theories. The resource-based view of the firm holds that a firm can sustain a competitive advantage only if the foundation for that advantage lies in resources that are valuable, rare, imperfectly imitable, and not substitutable (Barney, 1991; Collis and Montgomery, 1995). When knowledge is viewed as a resource that accelerates a firm's technical progress, a decision to share it openly flies in the face of logic. An innovating firm that successfully protects its technological knowledge will improve its chances of maintaining competitive advantage over rivals. A firm that shares knowledge with outside organizations will erode that competitive advantage by transforming valuable resources into public goods that contribute to the research efforts of all innovating firms. 

In this paper, I develop and test the argument that under some industry conditions, firms that share technological knowledge with the innovation system will achieve higher innovative performance than firms that do not share knowledge. Firms can share knowledge with the innovation system by publishing technological advances in scientific journals, presenting papers at technical conferences, or providing knowledge to other researchers directly through conversations. Firms can attain high innovative performance by developing and holding intellectual property rights over new-to-the-world products that are desired by large commercial markets. 

Isolating one predictor of high innovative performance will assist firms with a critical step in the commercialization process. Firms that achieve high innovative performance generally earn significant rents over their lifetimes (Anderson and Tushman, 1990). 

In the next section of the paper, I will construct a logical argument and offer two propositions. In the following section, I will describe the operational definitions of each of the constructs and data sources for the empirical analysis. The empirical analysis explores firms' knowledge sharing activities in the global flat panel display (FPD) industry. FPDs are thin information display monitors that replace cathode ray tubes (CRTs) in a variety of applications such as laptop computer screens, cockpit display screens, and handheld TVs. Several distinct technologies have competed for market share in FPD applications. In the following section, I will report the results of a regression analysis designed to test the theoretical arguments. Finally I will put the empirical findings into context and summarize the importance of the results for managers and researchers. 

Preparadigmatic and Commercial Competition 

Management scholars have long recognized that the nature of competition changes over the course of innovation and commercialization. Early in the emergence of a new industry, conditions of preparadigmatic competition prevail (Teece, 1987). In this preparadigmatic phase, rivalry centers on differences in the fundamental design of firms' products. 

The path dependent nature of technological progress ensures that preparadigmatic competition is particularly intense. Early in the development of a technology, firms hold divergent beliefs about the commercial and technical feasibility of different technological approaches (Nelson and Winter, 1982). Once a firm begins to devote resources to one approach, its competencies become specialized and earlier technological choices constrain future options (Arthur, 1988). Thus, in many industries firms are dispersed among several path dependent technological trajectories and each has an enormous interest in seeing its own trajectory "win" the preparadigmatic competition. Even within one technological trajectory, firms' product designs may compete on more minor attributes. 

The commercial phase of innovation begins when the marketplace chooses one product design as the technological paradigm for any application. In many industries, this amounts to a selection of a single "dominant design" for most or all end applications (Abernathy and Utterback, 1978). In other industries, a small number of competing technologies may survive when the criteria for evaluating the product design vary considerably across end product markets. Often times, one technology will win out for large commercial markets while a competing product design pursues a small niche market. 

The technological paradigm defines the single product architecture that has won out for any particular application. This design sets the pattern for subsequent technological progress in the industry (Dosi, 1982; Sahal, 1981). The establishment of the paradigm comprises a critical step in product standardization, which is necessary to attract complementary products, encourage investment in infrastructure, and achieve production economies (Utterback and Suarez, 1993). Firms continue to improve their products throughout the commercial phase of competition. However, the focus of these efforts shifts towards incremental refinements within one trajectory (Abernathy and Clark, 1985). Inter-firm rivalry shifts away from rivalry between product designs and toward competition based primarily on price and superficial differences in features (Teece, 1987). 

During the commercial phase, one or two product designs usually dominate most large applications. Alternative technologies disappear or are relegated to smaller, niche markets. However, the technological paradigm cannot simply rest on its laurels. Although paradigms, like most institutional arrangements, generally persist over extended periods of time, they are not immutable. For instance, Anderson and Tushman (1991) found that a new dominant design arose three times over 25 years in the minicomputer industry. 

I focus on firms' strategies only under conditions of preparadigmatic competition. The arguments in this paper are not applicable to commercial phase of innovation. These arguments are limited to high technology industries (in which scientific and technological knowledge makes a real contribution to the development of a technology), and to industries in which real competition among different product designs drives preparadigmatic competition. It is outside the scope of this project to test a contingency model of circumstances in which knowledge sharing becomes an appropriate strategy. Instead, I pursue the more modest goal of testing whether circumstances arise in which firms can achieve high innovative performance by sharing technological knowledge with competitors. 

Firms' Strategies of Knowledge Contribution: Strength through Sharing 

Management researchers have noted that many innovations that achieve tremendous commercial success fall short of the technical performance offered by alternative product designs (Arthur, 1988; David, 1985). Scholars have explained this finding by arguing that a firm's innovative performance depends not only on its product's technical viability, but also on the institutional environment in which it operates (Bijker, Hughes and Pinch, 1987; Constant, 1980; Usher, 1954). The institutional environment consists of the technological and evaluation standards that make up the architecture of the innovation system. Technological standards dictate the set of technical interfaces through which a new product interacts with existing and future complementary products. Evaluation standards specify the set of criteria that are used to judge the merits of the innovation. This institutional environment co-evolves with the technological advances of innovating firms via repeated interactions among innovators, suppliers, end-users, and regulators (Garud and Rappa, 1994; Ruttan and Hayami, 1984; Van de Ven and Garud, 1993; 1989; Van de Ven, 1993). 

Firms' Strategies to Shape the Institutional Environment 

Because the technological and evaluation standards present in the institutional environment emerge endogenously, each innovating firm can wield some influence over their character. Indeed, firms can pursue strategies to shape their institutional environment (Das, 1994) in what Latour and Woolgar (1979: 243) called a "fierce fight to construct reality." I conclude that a firm can improve its innovative performance by pursuing strategies to actively shape the institutional environment in its own favor. 

One important resource that a firm can use to shape its institutional environment is its ability to direct the conversation that takes place among industrial researchers concerning technological and scientific advances in the industry. I argue that a firm can best influence this scientific conversation by sharing its technological knowledge with other organizations in the innovation system. By sharing proprietary technological knowledge with the innovation system, a firm can influence the institutional environment in two ways. First, it can attract firms to its own technological trajectory and, thus, form a critical mass of firms with a vested interest in the success of the technology. Second, it can influence the emergence of technical and evaluation standards in favor of its own technology. 

Attracting New Entrants 

Any one technological trajectory is more likely to win the preparadigmatic competition if a critical mass of well-respected firms is pursuing research in the area. By sharing its technological knowledge with the innovation system, a firm's researchers can attract others to the firm's own trajectory. Since a firm's innovative performance depends on its ability to win the preparadigmatic competition, firms that share knowledge will have higher innovative performance than firms that do not share knowledge with the innovation system. 

Let's consider this argument in more detail. First, I argued that the likelihood that a trajectory will win out as the technological paradigm depends in part on the number and reputation of firms pursuing that technological approach. This is true for several reasons. First, as the aggregate amount of resources devoted to a trajectory increases, the speed of progress toward a commercially viable product will accelerate (Podolny and Stuart, 1995). Industry observers often justify the dominance of one product design over another by citing the speed of advance of the dominant technology. Some FPD makers, for instance, point out that a number of technical advances occurred in LCD technologies in the 1980s that paved the way for their selection as a technology paradigm. The speed of a technology's advance may depend partially on characteristics inherent in the technology, itself. However, it is equally plausible that dominant technologies enjoy rapid advancement due to the sheer volume of effort devoted to their progression. 

Additionally, a product design that is close to market will attract investments by makers of complementary assets and guide the expectations of end users in the market. This is particularly important for high-technology products that require deep infrastructures of equipment and materials suppliers for efficient production. Custom tools and materials are prohibitively expensive and their design can delay a firm's time to market for months and years. 

Beyond this, Van de Ven (1993) proposed that innovating firms that "run in packs" will be more successful than those that do not. He reasoned that by cooperating as well as competing, innovating firms take turns performing various functions much in the way that bicyclists use each other to shield the wind in long distance racing. Podolny and Stuart (1995) found empirical evidence showing that entry of well-respected firms into an existing technological "niche" leads to an increased probability that the niche will be important to the development of the industry. Finally, by having a large chorus of well-respected researchers with a vested interest in this technology, firms on the trajectory may speak with a louder voice in influencing the institutional environment. 

Next, I argued that a firm can attract other researchers onto its own trajectory by sharing technological knowledge about its own progress with others in the innovation system. By disseminating technological knowledge, the firm divulges general approaches for developing the technology, indicates the state of current research, and offers clues about methods to overcome specific technical obstacles. Each of these contributions can help reduce barriers to entry for new FPD firms. 

Additionally, knowledge dissemination can help firms attract well-respected innovators to their trajectory. Rappa and Debackere (1992) argued that firms and scientists can improve their reputations by making their technological advances public, for instance by presenting their scientific advances at technical conferences or publishing results in scientific journals. Podolny and Stuart (1995) found empirical evidence that the status of the firms in any technological trajectory significantly affects the probability that high-status innovators will enter the trajectory. This suggests that knowledge sharing can help a firm attract high status innovators to its own technological trajectory. 

Strategies to increase the number of direct competitors a firm faces would be unthinkable under conditions of commercial competition. I argue, however, that such tactics will improve a firm's innovative performance in the pre-commercial phase of the innovation process by increasing the likelihood that its technological trajectory will win the preparadigmatic competition. If the marketplace selects the firm's technology as the paradigm in an application with large end product demand, then all firms pursuing that technological path will achieve higher innovative performance than if they had been relegated to the "losing" path. 

Shaping Evaluation Standards 

A firm can also shape the emergence of its institutional environment by influencing the technological priorities of researchers pursuing all technological trajectories. A firm's own technology is more likely to win the preparadigmatic competition if the firm's researchers are able to influence all scientists' perceptions concerning the most critical technical obstacles to overcome before commercialization. Firms can influence other researchers' perceptions about technical priorities by publicizing their own research activities and making the limitations of the technology known to all members of the innovation system. Therefore, a firm will achieve higher innovative performance if it shares knowledge with other researchers in the innovation system. 

Once again, let's consider this argument in greater detail. First, I argued that a firm is more likely to see its own technology emerge as the dominant design if it can influence all researchers' technical priorities concerning the advancement of the technology. By focusing the attention of the scientific community on a certain set of problems, the firm influences all researchers' perceptions concerning which technical attributes are most important for the commercial product. In effect, by influencing the technical priorities of researchers, the firm helps to set the standards that will be used by all parties who judge each firm's product. The firm's own technological advances necessarily represent its own technical priorities, as well as the problems that the firm has made progress in solving. Therefore evaluation standards based on these advances should reflect well on the firm's own technology. 

For example, FPD product designs reflect a series of tradeoffs between different technical attributes such as breadth of viewing angle, weight, resolution, and power consumption. By sharing knowledge concerning one method of improving a display's resolution, a firm may focus the technological community's attention away from making improvements on the dimensions of weight and power consumption and toward developing high-resolution screens. Since the firm's researchers perceive resolution as a priority and have devoted resources toward the issue, a firm's interests lie in ensuring that the industry acknowledges the importance of that dimension. A firm is more likely to win the preparadigmatic competition if the evaluation criteria that become standardized in the institutional environment reflect the firm's own vision of a high resolution design. In sum, by influencing all researchers' priorities, the firm can persuade innovators along all technological trajectories to compete on the firm's own terms. 

Next, I argued that a firm can influence the priorities of all researchers in the industry by sharing its technological knowledge with the innovation system. Rappa and Debackere (1992a) pointed out that a scientist's choice of problems to investigate is heavily influenced by the opinions of other scientists in related fields. Zuckerman (1978) concluded that two criteria drive scientists' selection of what technical problem to devote effort toward: the assessed scientific importance of a problem, and the feasibility of arriving at a solution. She further noted that social processes in the scientific community bias researchers' choices of problems. These social processes include "reactions to the inferred critical attitudes or actual criticism of other scientists and ¼ an adjustment of behavior in accordance with these attitudes" (Merton, 1938: 219). Additionally, several authors have suggested that researchers' choice of what specific problem to address can be influenced by what is "fashionable" in the scientific community (Barber, 1990; Crane, 1969). These authors conclude that there is a clear potential for one researcher to influence other researchers' choices of problem to address. 

Proposition 1: Under conditions of preparadigmatic competition, firms that share technological knowledge with their innovation system will have higher innovative performance than firms that do not share knowledge with their innovation system. 

Sharing Knowledge with the Global Innovation Systems 

So far, I have sidestepped the issue of the geographic boundaries of a firm's institutional environment and innovation system. Firms could share knowledge within either a national innovation system (NIS) or a global innovation system (GIS). In the context of technological innovation, the notion that institutional environments vary cross-nationally holds important implications. If distinctions in countries' national innovation systems cause evaluation standards to vary cross-nationally, then a different technology may well win out as the paradigm in different countries. If countries' dominant designs emerge independently of one another, then a firm will find its best interests served by tailoring its knowledge-sharing strategy to only one country. 

There are reasons to believe that in many high technology industries, an innovating firm must participate in institution-building activities within its GIS as well as in its own NIS. Kobrin (1991, 1994) has argued that in a growing number of high-technology industries, national markets no longer encompass sufficient geographic space to serve as minimally efficient markets. Kobrin (1991) showed empirically that innovations that require particularly high up-front R&D expenditures tend to associate with globally integrated industries. 

Proposition 2: Under conditions of preparadigmatic competition, firms that share technological knowledge with the global innovation system will achieve higher innovative performance than firms that share knowledge merely with their national innovation system. 

I tested these two propositions using data contained in the scientific papers of industrial researchers in the global flat panel display (FPD) industry. 

Innovative Performance 

Innovative performance consists of a firm's ability, in the early phase of competition, to develop and obtain intellectual property protection for a product demanded by large commercial markets. The act of creating a technological advance that is unique, and therefore patentable, is not enough. The technological advance must have some value in the marketplace once the phase dedicated to commercial production begins. Innovative performance reflects the value of the innovation, itself, rather than the firm's skills at manufacturing and marketing the resulting commercial product. 

The value of a firm's product innovations parallels the value of its patent portfolio. Because patents are marketable through either licensing or sale, the value of the patent portfolio is not dependent on a firm's skill at manufacturing and selling the new product. 

Patent Renewal Method as an Estimate of Patent Value 

A simple count of the patents awarded to a firm is a very poor measure of the value of that firm's patent portfolio or the value of its innovations (Putnam, 1996; Podolny and Stuart, 1995; Lanjouw, 1993; Griliches, 1990; Schankerman and Pakes, 1986, Pakes and Schankerman, 1984). In fact, research has shown that simple patent counts are very highly associated with the amount of resources that a firm puts in to the innovation process, but relative poor predictors of the firm's innovative performance (Griliches, 1990). 

I followed Schankerman and Pakes (1986), Pakes and Schankerman (1984), and Lanjouw (1993) and estimated the value of each firm's patent portfolio by tracking innovators' choices to pay fees to renew their patent over time. In many European countries, innovators must pay an annual fee to maintain intellectual property protection for their patented technology. These fee schedules demand relatively small payments in the early years of a patent and increasingly expensive fees as the patent ages. For instance, in Germany these annual fees begin at DM 100 in the third year of patent protection, and reach DM3300 by year 18. The patent renewal model assumes that patent holders choose to pay renewal fees only when: 
 

The expected revenue from intellectual property protection for the next year + The value of the option for further renewal

>

The cost of renewing the patent + The expected value of the court and legal fees required to defend their patent against competitors' infringement

 

Those patents that hold value and retain that value over time will tend to be renewed for many years. Patents that hold little value or lose value over time will be allowed to expire early on. 

In order for their innovation to enjoy intellectual property protection in all markets, firms based in Japan, the U.S., and all European countries must be awarded a patent in Germany. Therefore, it is reasonable to use German patent data to assess the performance of FPD firms from all over the world. 

I calculated relative values for all FPD firms' German patents using Lanjouw's findings concerning relative differences in the value of German computer patents renewed in any given year. Lanjouw (1993) estimated parameters for the probability of obsolescence of the patent and the real discount rate of a patent using renewal fee schedules and data on 15,000 German patents in four technology areas. 

Table 1 displays the relative patent values for computer technologies renewed each year compared to the value of patents dropped at age 3. 

In order to avoid a systematic bias, the four FPD firms that were headquartered in Germany were dropped from the analysis. In addition, by including a European dummy variable, I controlled for the possibility that those firms were more likely than U.S. or Japanese firms to renew their German patents. After estimating the value of each FPD patent, I aggregated those values to the level of the firm to arrive at a value for each firm's patent portfolio in FPD technologies. 

According to Griliches (1990), the patent renewal method is one of the most appropriate methodologies available to measure the private value of a patented innovation. Additionally, because it is a longitudinal measure, the estimation reflects how the value of the patent unfolds over time, into the commercial phase of innovation. 

Knowledge Sharing 

A firm shares knowledge with the innovation system by making its scientific and technological knowledge public to other researchers pursuing similar topics. For the purposes of this empirical analysis, knowledge trading, or giving knowledge to specific other firms as part of a formal or informal relationship does not constitute sharing knowledge with the innovation system. Knowledge sharing must be done in a public forum such as a scientific journal or technical conference that is open to scientists from a number of different organizations. 

In the empirical analysis, I measured the geographic scope of firms' knowledge sharing activities. A national innovation system (NIS) is the set of resources and institutions, built through the interactions between a country's major research organizations, that domestic firms harness to successfully commercialize innovations. A global innovation system (GIS) is the set of resources and institutions built through the interactions between major research organizations from around the world, that firms from all countries harness to successfully commercialize innovations. The concepts of NIS and GIS are domain-specific. In the empirical analysis, I considered only firms' activities to share and appropriate knowledge with their own technological field. 

The amount of knowledge that a firm shared with its NIS or GIS was, in large part, reflected in the proportion of its audience that resided within its own country and abroad. A firm could share knowledge with its NIS by attending local and regional technical conferences, publishing papers in provincial outlets, and distributing papers to domestic colleagues. It could share knowledge with its GIS by attending foreign conferences, publishing papers in foreign scientific journals, and publishing papers in foreign languages. It could also share knowledge in domestic outlets that were widely read internationally. For instance, some Japanese firms published their own company digest of technical papers in two languages to increase the geographical breadth of their knowledge sharing. 

My measure of knowledge sharing consisted of three dimensions. Tables 1 and 2 list the three dimensions of sharing knowledge with the NIS and with the GIS. 

The first dimension for each measure represented the volume of knowledge that the firm made an effort to share. By publishing its technical advances in scientific journals and presenting papers at scientific conferences, the firm exerted effort to make its knowledge available to the scientific community. The second dimension represented the quality and usefulness of the knowledge it shared. Firms that published only scientific knowledge that was well established in the industrial community, that scrubbed their articles so clean that the publications conveyed little real information, or that published on relatively obscure topics, should rarely be cited. Finally, the third dimension represented the breadth of the community with which the firm shared knowledge. 

Because every country had different numbers of firms and researchers in the FPD industry, the size of each country's national and global innovation systems varied. I calculated the size of countries' NIS and GIS based on the estimated number of domestic and foreign industrial researchers studying FPDs. Table 4 reports the size of the NIS and GIS for firms in Japan, Europe, and the U.S. 

Because the database included scientific journals located in each of the countries that housed FPD firms, I treated a firm's choice concerning where to publish its material as strategic, and not based on inherent limitations due to the size of its NIS. Therefore, I used the size of each country's NIS and GIS to standardize dimensions two and three of the sharing variables (number of domestic and foreign citations to the firm's research; domestic and foreign number of firms that cited the firm's research). I did not standardize the first dimension, the number of articles published in the NIS and the number of articles published in the GIS. 

These three dimensions of knowledge-sharing proved too highly correlated to be interpreted meaningfully in a single regression equation. Because the two, broad knowledge-sharing variables and not the individual dimensions were of greatest interest, I followed Berry and Feldman (1985) and combined the dimensions into two variables: sharing with the NIS and sharing with the GIS. Therefore, the complete operational measures of each of the sharing variables was the following: 
 

Knowledge Sharing =  in the NIS

Zscore (Volume in NIS) + Zscore (Quality in NIS) +  Zscore (Breadth in NIS)

 

These dimensions appeared to fall together quite well. Cronbach's Alpha is .8207 for the sharing with the NIS variable and .9282 for the sharing with the GIS variable. 

Learning from the Innovation System 

Firms that draw knowledge from outside organizations in their innovation system should achieve higher innovative performance than firms that do not scan for advances in their technological environment. I included variables to measure the relative differences in the amount that firms learned from their innovation systems. I operationalized learning from the innovation system as the number of different scientific articles from university and government researchers that the firm referenced in its own publications. Learning data was also partitioned to measure learning from the NIS and learning from the GIS. Similar to the sharing variables, I standardized the learning variables for the size of each firm's NIS and GIS. Table 5 provides operational definitions of both learning variables. 

Several other factors should predict high innovative performance and must, then, be included in the regression equation. First, the size of the firm's research effort in the FPD area has strong implications concerning its ability to develop a technically viable, patented product. Because of the difficulty of measuring this variable accurately, I used a composite measure of three separate proxies of a firm's research efforts. The first proxy consisted of the number of researchers in each firm who held membership in the global industry association, the Society for Information Display (SID). Because SID offered researchers benefits such as trade magazines, scholarly journals and regular industry conferences at a nominal cost, industrial researchers received incentives to join the association. I reviewed SID membership directories from 6 years to identify the number of SID members each firm employed during their peak membership year. 

The second measure consisted of the number of active researchers who authored or co-authored a paper on any FPD technology. In order to estimate this number, I identified the single year in which each firm published the greatest number of FPD articles. I then calculated the number of distinct authors listed on any of the firm's articles in that peak year, as well as the number of different authors publishing in one prior and one subsequent year. Finally, I assumed that the size of the firm's research staff increased monotonically from zero the year before their first article was published to the peak year, and decreased monotonically until the year after the last article was published or patent awarded (if exit from the industry occurred). This calculation of the number of person-years dedicated to FPDs formed the basis for the second variable in the composite. 

The final measure of this control variable was a very conservative one. The literature on patent valuation has consistently concluded that the number of patents that a firm is awarded is a better measure of research inputs than of research outputs. The number of patent awards indicates the number of successful projects a firm has pursued that meet some minimal threshold level of economic value (Griliches, 1990), but may or may not enjoy commercial success. Pakes and Griliches (1982) found a very strong, direct relationship between a firm's R&D expenditures and the quantity of its patents. Their analysis produced an R-squared on the order of .9 for cross-sectional data and remained significant when observing changes in firms' activities over time. That is, when firms increased or decreased R&D expenditures, parallel changes occurred in levels of patenting (Griliches, 1990). 

I used the number of U.S. patents that each firm was awarded as the third dimension of the research effort variable. In many ways, U.S. patent counts provided a far too conservative control variable. Nearly all non-German firms that apply for German patents file a duplicate patent in the U.S. Further, while the patent valuation method used to operationalize the dependent variable provided important information beyond a simple patent count, its valuations were not entirely removed from volume of patent applications. Indeed, the patent valuation method involved adding the patent values across the total number of patent awards. 

I believe, however, that the inclusion of the U.S. patent data as a control was critical. First, the other two measures relied to some extent on firms' knowledge-sharing activities, themselves. U.S. patent data provided a broader measure of resources devoted to FPD technologies and was not dependent on firms' participation in the innovation system. In order to be confident of regression results, I had to ensure that the knowledge-sharing variable was not simply capturing the volume of resources that a firm devoted to FPD research. This control variable, more than any other, warranted a conservative measure. 

In effect, by including U.S. patent counts as one proxy for resource inputs, the regression tested whether a firm's knowledge-sharing strategy was a significant predictor of the value of its patent portfolio, beyond simply the number of patents it was awarded. This simply raised the bar for testing the significance of the knowledge-sharing variables. 

The full research effort variable consisted of the sum of the Zscore of each of the three dimensions described here. 
 

Zscore (Number of FPD employees who held membership in SID.)

+

Zscore (Number of different authors affiliated with the firm, estimated over all years of research activity.)

+

Zscore (Number of U.S. patents awarded to the firm for FPDs.)

 

Size and Nationality 

The statistical analysis required two additional variables. It was possible that large multinational firms may have had slack resources to renew patents of questionable commercial value or have had a greater ability to apply for patents in foreign countries. Size and multinationality proved to be highly correlated. Therefore, I chose to include only the number of total employees in all divisions of the firm to operationalize the size variable in this analysis. Second, previous research has shown that in many high-technology industries, firms located in some regions systematically outperform firms located in other regions. I included dummy variables for Japanese, North American, and European firms. 

As a way of summarizing, I have listed two testable hypotheses. 

H1: The amount of technological knowledge that a firm shares with its innovation system will be a significant and positive predictor of innovative performance. 

H2: The amount of technological knowledge that the firm shares with the global innovation system will be a significant and positive predictor of innovative performance beyond the effects described in H1. 

Data Collection 

I collected data on German patent awards for all FPD firms from microfilm and computer files at the German patent office in Munich, Germany. 

I collected data on firms' publications and conference papers from the INSPEC database, a computer-based database comprised of scientific and technical journals and conference proceedings in physics, electrical engineering and electronics, computing and control, and information technology. 

I downloaded articles that were published on FPD technologies between 1954 and 1989 using Boolean searches of keywords derived from interviews with engineers and managers of FPD companies. Each individual article or conference presentation became one entry in a computer database that contained information on the authors, title, year of publication or presentation, affiliations of all authors, and location of the conference or publication of the written volume. The information associated with each article was given a unique record number and stored in a relational database. Finally, in order to increase the comprehensiveness of the data collection effort, I manually entered information about articles that were cited by retrieved articles, but that had not been listed in the INSPEC database. 

Citations by one industrial researcher of another were processed manually from the physical article, itself, and entered by hand into the database. All citation relationships were aggregated to the firm level for the analysis. 

The final database consisted of 24,997 scientific articles and conference records. I downloaded 17,264 relevant articles from the INSPEC database and manually entered 7,733 more that were listed as citations in relevant articles, but that were either in sources not indexed by INSPEC or preceded the earliest date indexed by INSPEC. The articles in the broad database, which included articles written by researchers in industry and academia, were published in 39 different countries and 19 different languages. 

Of these, 3,448 articles (2,799 from INSPEC; 649 from manual entry) were written by researchers in an FPD firm. These 3,448 articles constituted the full database used to operationalize the knowledge-sharing variables. To operationalize the learning variables, I used a sample of 3,500 articles published by researchers from universities and government agencies from 1954 to 1989. From these 3,448 articles by industrial researchers and 3,500 articles by university and government researchers, I entered 34,802 individual citation relationships (each reflecting one occasion in which Researcher A cited Researcher B) that formed the backbone of the database. 

I gathered data on the size of a firm's FPD R&D team from membership directories of the Society for Information Display (SID) and authorship records in the database, itself. I collected data on firms' U.S. patent awards from the Questel Orbit UPAT database. I calculated the total number of employees in all divisions in a firm from Compact D (1990), Worldscope/Global (1992) and American Business Disc (1994) databases. When a range of employees was given, I assumed the total number to be the midpoint of that range. 

Hypothesis Tests 

This paper has contended that firms that share knowledge with their innovation system will achieve higher innovative performance than firms that keep their technological knowledge secret. I tested the hypotheses using an OLS regression. 

Curve estimates of several of the predictor variables against innovative performance showed quadratic tendencies for the model as a whole. Therefore, I used the eight independent and control variables to predict the square root of innovative performance. The regression reported below showed no problems resulting from multicollinearity. 

Table 8 estimates the model for the dependent variable, square root (innovative performance). 

Y=b₀ + b₁FS + b₂RE + b₃JF + b₄EF + b₅LG + b₆LN + b₇SG + b₈SN + e 

Where FS= Firm Size 

RE= Research Effort 

JF= Japanese Firms 

EF= European Firms 

LG= Learning from the GIS 

LN= Learning from the NIS 

SG= Sharing with the GIS 

SN= Sharing with the NIS 

Model Summary 

R Square .670 

Adjusted R Square .639 

Std. Error 318.6157 

F 21.096 

Sig .000 

Dependent Variable: Square Root (Innovative Performance) 
 

Model	Beta	B	Std. Error	T	Sig.
(Constant)		258.033	128.268	2.012	.047*
Log (Firm Size)	.103	40.579	33.034	1.228	.223
Research Effort	.261	54.631	24.521	2.228	.029*
Japanese Firms	.068	78.029	87.834	.888	.337
European Firms	.092	150.101	116.997	1.283	.203
Learning from GIS	.100	55.385	65.661	.844	.401
Learning from NIS	.003	11.111	225.332	.049	.961
Sharing with GIS	.457	86.376	30.742	2.810	.006**
Sharing with NIS	-.068	-16.069	20.367	-.789	.432

* Significant at .05 level (two tailed test) 

** Significant at .01 level (two tailed test) 

The significance of this model is very high, with an F of 21.096 (significant beyond the level of .000). This model reported an adjusted R squared of .639. The control variable for firm size was positive, but not significant. Importantly, the control variable for research effort showed a high level of significance. Therefore, it is appropriate to interpret the results concerning levels of knowledge sharing and learning, after controlling for the level of research effort exerted by each firm. Firms that shared their knowledge with the GIS enjoyed significantly higher innovative performance than firms that did not share their knowledge globally. However, the same cannot be said for sharing knowledge within the firm's own national innovation system. Sharing knowledge with the NIS actually had a negative sign, though the variable was highly insignificant. 

Neither of the learning variables showed significance in this model. This suggests that firms that demonstrated having learned from the universities and governmental institutions in their national and global innovation systems performed no better than firms that did not demonstrate such learning. This result may derive from a weak measure of the learning variables. In particular, in order for a firm to have demonstrated any level of learning in this data, it must have first published articles, itself. Therefore, this data may have missed learning that took place without acknowledgement. Unlike sharing, a true measure of the more private construct of learning from the innovation system would require asking industrial researchers about their learning habits more directly. 

Neither country dummy variable showed significance in this model. I performed separate regressions to identify the importance of any country-based interaction effects. In order to identify the existence of cross-national differences attributable to both the intercept and the slope of the regression equation, I partitioned the sample into the three regions (North America, Asia, and Europe), and performed separate OLS regressions. These results showed no significant differences among the three models. Additionally, most of the variables, while holding the same direction, became insignificant. The most probable explanation for this is the very small sample size that resulted from the partitioning. Each equation included six variables, and employed sample sizes of only 26 Japanese firms, 14 European firms, and 52 U.S. firms. A simple means test, however, did demonstrate that Japanese firms achieved significantly higher innovative performance than US firms. 

This means test shows that Japanese firms outperformed U.S. firms at a .001 level. It shows no significant difference concerning innovative performance between Japanese and European firms, or European and U.S. firms. 

The means test showed significant cross-national differences in firms' innovative performance. Combined with the insignificance of the country dummy variables, it appears that Japan's higher innovative performance is primarily attributable to firms' learning and knowledge-sharing strategies. 

Direction of Causality 

The study presented in this paper involved a cross-section of data from firms in the FPD industry in nine countries. As with any cross-sectional study, the results provided here do not demonstrate causality between the independent and dependent variables. The data collection was, however, structured to reduce the possibility that reverse-causality would drive the results. 

In order to support a contention of causality, a study must meet three criteria. First, it must offer a sound theoretical argument supporting one direction of causality over the other. Second, the empirical results must show a statistical association between variables. This study met these first two objectives. Finally, the study must demonstrate that the independent variable temporally precedes the dependent variable. Although the study is cross-sectional in nature, data for the independent and dependent variables were lagged in such a way that the causal direction was more likely to be direct than to be reversed. 

Scientists at Philips published the first article that is contained in the database in 1954. Data on the independent variables, sharing with the GIS and sharing with the NIS, spans the time period from 1954 to 1989. The first FPD patent identified was awarded to Philips in 1966, and data on the dependent variable, firms' innovative performance, spans the time period from 1966 to 1995. For 58 percent of FPD firms, the firm's first paper in the publication database preceded the year of its first patent, with an average lag between publication and patenting of 6.4 years. In 13 percent of firms, the firm's first publication and first patent took place during the same year, and in 29 percent of firms, the first patent was listed before the firm's first publication. 

It remains impossible to prove that causality lies in the direction proposed in the theory. However, the design of the study supports the conclusion that a firm's knowledge-sharing strategies led to high innovative performance. The regression equation included a very conservative control for the amount of research effort that each firm dedicated to FPD technologies. Therefore, it is unlikely that the results simply captured the spurious phenomenon that firms that do more research are able to both share more knowledge and enjoy higher innovative performance. Similarly, the lag between the independent and dependent variables makes it improbable that the direction of causality is reversed. Firms did not publish articles because they had high innovative performance. Therefore, it is likely the case that firms enjoyed higher innovative performance because they shared more knowledge with their global innovation system. 

The empirical analysis suggested three interesting findings. First, some firms actively designed strategies to share knowledge with their innovation systems. Eighty firms shared at least one scientific paper or presentation with the innovation system from 1954 to 1989 and these industrial researchers shared 2,818 papers over this time period. Supplementary interview data suggested that these papers, at times, provided important pieces of the technological puzzle that facilitated competing firms' efforts to advance new technologies. 

Second, sharing knowledge with the GIS was a significant predictor of high innovative performance. Statistical evidence supports the hypothesis that firms that shared knowledge with their global innovation system attained higher innovative performance than firms that did not share knowledge. The regression model yielded an R-Squared of .639. This was significantly higher than a model that excluded firms' knowledge-sharing activities. 

The most notable conclusion about firms' knowledge-sharing strategies that emerges from this paper is the critical importance of the global innovation system. For knowledge-sharing strategies to be effective in global industries, knowledge sharing must also be global. Firms that shared knowledge with the GIS enjoyed high innovative performance. However, the relationship between sharing knowledge within the NIS and innovative performance was not significant. This suggests that firms that took a very provincial approach to knowledge sharing by interacting solely with other firms in their home country did not reap the benefits of a knowledge-sharing strategy. 

Third, although Japanese firms did achieve higher innovative performance than U.S. firms, the differences appeared to be attributable to firms' knowledge-sharing strategies. Data did not suggest that the relationship between knowledge sharing and innovative performance varied across countries. No country's firms earned significantly higher innovative performance than was predicted by the firm's size, research effort, learning, and knowledge-sharing strategies. Additionally, there was no statistically significant evidence that the relationship between a firm's knowledge-sharing strategies or learning behaviors and innovative performance varied across countries. 

Many managers and researchers have contended that firms' best interest lies in exploiting proprietary technological knowledge without attracting imitators to their technological trajectory (Teece, 1997; Scherer, 1992; 1980) . Hamel (1991) framed firms' interactions in terms of strategic races to out-learn one another. In interviews, two high technology managers agreed that, whenever possible, they try to give their firms a technological edge by limiting the amount of knowledge that escapes their research labs (Levien, 1996; Holmberg, 1997). 

This empirical study supports the opposite perspective. Under certain conditions, knowledge conveys more value when it is shared than when it is kept secret. 

Current literature on firms' standard setting activities acknowledges that in networked industries, firms may achieve higher performance by attracting firms to their technology through licensing activities and introducing open architectures. Analysts acknowledge that IBM's open architecture strategy facilitated its success in competition with Apple's closed architecture design. The VHS design to video recorders may well have beat out its competitor, Betamax, because several firms licensed and produced VHS systems. Software developers commonly give away versions of their product for free in order to speed market acceptance of their design. 

This paper suggests that the importance of having a critical mass of competitors on the same technological trajectory extends well beyond firms operating in networked industries. The primary argument in this paper rests not on the importance of the common interface standards that are critical for success in networks, but in developing common evaluation standards that favor the firm's own product design. The dominance of LCD screens for laptop computer applications did not depend on preferential product interfaces with other computer components. Unlike the cases of IBM/Apple and VHS/Beta, it was the evaluation criteria and need to build an industry infrastructure, rather than interface standards, that drove preparadigmatic competition in the FPD industry. Once one product design emerged as dominant, the ideal FPD that was envisioned by producers and consumers became limited by the traits and evaluation criteria of the established product. 

This study suggests that a firm's strategy to influence the selection of one technological paradigm must begin long before the beginning of the commercial phase of competition. This raises a distinction between a firm's decision to build a critical mass of competitors by licensing its own technology for production by other firms, and its strategy to actually share technological knowledge and advances before the product is fully developed. Hamel (1991: 84) points out, 

This paper provided evidence that one important way that some firms can shape the technological paradigm and achieve high innovative performance is by sharing their technological knowledge with the global innovation system. Managers should view their firm's scientists' desire to publish their findings in the scientific literature not as a waste of time or a potential threat to the firms' competitive advantage, but often as an opportunity to influence the activities of competing firms. In the early phases of innovation in some industries, knowledge sharing is a way to increase the firm's innovative performance in the long run.