R&D alliances and the effect of experience on innovation: a focus on the semiconductor industry.

Past success and failure shapes the future decisions of an organization. Recent research shows that experience with previous strategic alliances is an important determinant for new alliances. However, the benefits of past experience depreciate rapidly, and the total number of long-dated experiences does not appear to be a major source of success in dynamic industries. The authors extend Sampson's work to the semiconductor industry to determine the effect of experience on these firms. This article offers an interesting extension to previous studies on experience effects and shows that experience does matter in amount and recency.

Keywords: alliance; experience; innovation; R&D

**********

Firms can learn from experience. New processes are improved continuously as the organization learns new and better ways to deliver a product or service to its clients. Past success and failure shapes the future decisions of an organization (Levitt & March, 1988), thus, the greater the experience of the firm, the more likely it is to adapt to new conditions. The learning curve literature has shown that a firm can achieve important gains in productive efficiency as its cumulative output increases and that this can imply an important competitive advantage (Lieberman, 1987). Extending this line of thought to other organizational areas, recent research has shown that experience with previous strategic alliances is an important determinant for new alliances (Sampson, 2005). However, the benefits of past experience depreciate rapidly over time, and the total number of long-dated experiences does not appear to be a major source of success in dynamic industries. In a recent article, Sampson (2005) looked at past R&D alliances as primary determinants of future alliances success, and even though she found support for this hypothesis, she also reported, "Further work is necessary to link experience, differences in alliance management practices between firms and performance" (p. 1028), and the reasons for the rapid depreciation of this experience are still unclear.

The primary purpose of this article is to extend Sampson's (2005) study. First, we replicate this study to contribute to the validation of research conducted on the learning effects of past alliances (Anand & Tarun, 2000). Second, we will also extend these findings by looking at a different industry, namely, semiconductors. By analyzing a different environment in a different period of time, a stronger case can be built for generalizing Sampson's results.

The primary research questions that this study tries to answer are "How can firms capitalize on past experience with strategic alliances?" and "Why are only recent experiences, when compared to experiences that occurred long ago, the most important determinants of future success?" Answering these questions has important implications from both a researcher's and a practitioner's perspective. Corporate strategy research can benefit from new insights provided by research on organizational learning, particularly by enhancing the ability to learn from past alliances (e.g., by selecting few relevant issues to which devote attention). This is a process with interesting questions but blurred answers.

On the other hand, from a practitioner's perspective, if only recent experiences are more important when subscribing to new alliances, managers can disregard distant history and look more carefully to the recent past. Furthermore, by understanding the specific reason for the depreciation of past experiences, managers can choose among competing governance structures, or they can even select the number of new alliances that are advisable given the organization's history.

Sampson's (2005) results are expected to be validated with this study. The underlying theory and her results support the hypotheses presented; however, the present study can contribute to the generalizability of these findings. Our study looks at a different industry and at a different period of time. (1)

Learning From Experience

There is a fair amount of literature that discusses learning by doing and the experience curve. This learning process has been mainly studied in manufacturing settings, and the variable of interest has been production cost. As a firm's cumulative output increases, it learns how productive processes impact the cost of final output, and this experience allows for the development of new and improved processes. Thus, a firm's history of past success and failure will shape its ability to learn (Levitt & March, 1988). Consequently, a firm that has been exposed to more experiences will probably be better prepared to face new challenges in a more effective way, for example, by exploiting scale economies for repetitive processes.

Levitt and March's (1988) arguments for cost reductions due to cumulative output and experience are widely accepted, and there are enough subsequent research findings that support their views. For instance, Lieberman (1984) studied the learning curve in the chemical industry and found that cost reductions cannot be attributed exclusively to time. This conclusion provides supporting evidence of the importance of experience effects, as is also the case with similar studies in the shipbuilding industry (Argote, Beckman, & Epple, 1990). In summary, there is sufficient evidence to support the benefits of experience in the manufacturing environment. However, there is also compelling evidence showing the impact of experience effects on other industries. Service industries have been studied by Darr, Argote, and Epple (1995); Dutton and Thomas (1984); and Dutton, Thomas, and Butler (1984), and their research constitutes irrefutable evidence that experience is major determinant of cost reductions in different environments.

Experience has been also tied to success in innovation and new product development because firms that have developed an extensive knowledge base from their technological and market-related experiences are more likely to succeed in new product introductions (Barnett & Hansen, 1996; Nerkar & Roberts, 2004). These studies however do not explicitly address recurrent managerial processes through which a firm gains experience. These processes usually involve complex managerial practices, and the experience a firm gains over time might not only be associated with the number of successfully introduced new products but also with how a firm handles the learning from experience process itself.

Thus, experience effects need to be extended to more general activities such as management in organizations. Managers observe the outcomes of their decisions in activities such as development of new divisions, changing product lines, and research and development of new products or processes (Sampson, 2005). Furthermore, Baum and Ingram (1988) showed in their analysis of the hotel industry that experience is an important factor when it comes to anticipating consumer preferences in a dynamic environment. In a similar fashion, other studies have analyzed managerial practices and learning in acquisitions (Haspeslagh & Jemison, 1991), cross-border entry (Chang, 1995), and alliances (Harbison & Pekar, 1998). Even though these studies do not explicitly address experience effects, the findings are suggestive of this phenomenon.

R&D Alliances

Alliances are recurrent activities that organizations undertake and they are particularly difficult to manage because of the very nature of this managerial practice. It involves two parties with very different characteristics, cultures, and goals trying to coordinate activities that are complicated even within a hierarchical structure. A firm that is involved in an alliance must align its internal processes and routines with those of a partner that may have a completely different approach to dealing with similar issues. Other complications arise, for example, in the communication and coordination when one firm has difficulties observing the partner's activity. If this is added to conflicting interests, potential problems are evident between the allies.

Previous research suggests that many allies report low levels of satisfaction with the alliance. As much as 40% of incumbents expressed their dissatisfaction with the collaboration (Bleeke & Ernst, 1993). Moreover, another body of literature provides evidence of the difficulties in managing alliances. In this line, empirical studies report a significant termination rate among joint ventures where precarious governance structures were the origin of the early termination (Harrigan, 1985). We do not have enough evidence for these two affecting issues about alliances, but it has been suggested that inherent activities in managing the alliance have at least some responsibility. Thus, learning from past experience can provide an organization with significant insight on how to conduct new alliances.

If a firm has engaged in many different alliances, the experience drawn from the processes that led to successful collaboration should provide a foundation for facing a new alliance. Similarly, a firm that has experienced failures in the past will try to avoid those activities that were at least in part responsible for the negative outcomes. For instance, the selection of the appropriate governance structure or contractual relationship, which has been linked to performance, can be improved by experience.

Moreover, the greater the repertoire in terms of experiences from which a firm can draw to make decisions, the more accurate the judgment of an environment characterized by high levels of uncertainty. Alliances are managerial practices that usually involve judgmental and idiosyncratic calls, where previous processes will shape many of the future decisions, particularly when a firm is stepping into new territories. Generally, managerial experience has the greatest potential to affect performance in situations that are characterized by greater complexity and/or where outcomes are highly idiosyncratic or uncertain (Sampson, 2005). Thus, greater complexity calls for greater managerial attention and skills. For instance, a firm may find it difficult to assess the benefits of an alliance if these are directly unobservable (improved R&D processes) and parties' contributions have not been agreed on in advance. This is the perfect environment for experience effect becoming an important source of competitive advantage.

The preceding discussion suggests that the greater a firm's experience, the larger its dynamic capabilities are in alliance management. However, there are also many arguments that limit the potential benefit a firm can obtain by learning from experience. Learning can potentially lead to the adoption of specific processes more frequently. These processes in turn will be seen as more reliable, and this perception can hinder a firm from exploring new alternatives or from adopting potentially beneficial new processes. This has been labeled as organizational inertia (Hannan& Freeman, 1984), which may reduce collaborative effects and thus transform the potential benefits of learning into a core rigidity (Leonard-Barton, 1992). Therefore, learning and imitation of prior experiences may inhibit experimentation that could, in turn, improve collaborative benefits (March, 1991).

More recently, empirical evidence finds that the benefit of learning from past experience will decay over time; namely, more dated experiences will contribute little to overcoming new challenges. In the selected setting, R&D alliances, this poses a significant threat because organizations can forfeit exploring new alliance-related practices and stay with those that have provided relative success. Thus, it is expected that the hypothesized relationship will show a decreasing marginal returns to alliance experience.

Hypothesis 1: There are decreasing marginal returns to prior alliance experience. Firm collaborative benefits initially increase with the firm's prior alliance experience, but this rate of increase diminishes at higher levels of experience.

Empirical Analysis

Empirical Design

The empirical test looks at the relationship between prior alliance experience and alliance outcomes. The dependent variable is firm innovation, measured after the announcement of the alliance, as a function of past experience with R&D collaborations. As discussed in the previous section, we expect to find supporting evidence of this relationship after relevant control variables have been accounted for. Firm innovation is selected as the left-hand variable because R&D alliances are the primary focus of this study and innovation is more closely linked to R&D than financial performance measures.

Collaborative activities are usually a central point in any R&D strategy, and benefits of this practice will primarily impact the firm's innovativeness. Thus, innovation is probably the most meaningful measure of R&D alliance successes. This is of course an indirect measure of alliance effectiveness, but we can rely on firms' innovativeness, operationalized here as postalliance patenting activity as a close proxy.

It is empirically difficult to capture the contribution of past alliances to the overall performance, but it can be accomplished, provided strong control variables that will capture the rest of the variability in the model are included.

Data and Method

The data set used to test the hypothesis was constructed with information about patents and alliances in the semiconductor industry. (2) This industry was selected because of its dynamic characteristics and the fact that many firms in this industry rely on alliances to overcome high R&D costs and to face a changing environment. Alliances in this industry are also important because they serve as a mechanism for gaining access to new capabilities and speeding up new technology adoptions. On the other hand, this industry has significant patenting activity associated with R&D collaborations, and patents have been related to firm performance (Macher, 2004). In this industry, patents are closely related to the ability of firm to appropriate the returns of innovation.

The source of alliance data is the Securities Database Corporation (SDC) Database on Joint Ventures and Alliances. This is a comprehensive database that contains information on all kind of alliances since 1988 and is compiled from publicly available sources, including SEC filings, industry and trade journals, and news reports. Clearly this database offers the desired characteristics for conducting empirical studies on alliances, (3) and it has been extensively used for similar and unrelated research studies (Anand & Khanna, 2000; Sampson, 2004).

The sample for this study contains R&D alliances for firms in the semiconductor industry for the years 1997 and 1998. This period of time is appropriate given the important number of alliances in these 2 years and because it is possible to track firms' patents after the alliance was announced. Each record in this data set corresponds to an R&D alliance exclusively or in addition to marketing, production, and/or supply activities and funding activities. The final sample includes 86 R&D alliances involving 137 firms. Some characteristics of this sample are presented in Table 1.

International alliances are the number of alliances where the involved parties' headquarters are in different nations. Clearly, the majority of alliances involve one (or more) American firms.

For each firm mentioned in the previous sample, information about its patenting activity was collected directly from the U.S. Patent and Trademark Office (USPTO) Web site. (4) This governmental portal contains all the information about patents issued in the United States for the last 215 years and more. Detailed information is only available however since 1976, and it includes assignee name, patent technological classification, inventor name, issue date, and announcement date. Patent information was collected for every firm involved in an alliance as well as the other firms in its corporate structure because patents are often assigned to the ultimate parent firm and not the single subsidiary where the innovation took place. Sampson (2004) found that 73% of patents are assigned to the ultimate parent firm; thus, corporate-level patents were considered to avoid noisy measures of firm innovation. For each firm in the alliance data set, the relevant affiliations were obtained from the Directory of Corporate Affiliations. (5) The patent portfolio therefore includes patents assigned to firms in the alliance sample as well as parent firms.

Measures

Dependent Variable

Firm innovative performance (PATENT) is measured via citation-weighted, firm patenting in a 4-year, postalliance window. For example, if an alliance commences in 1997, PATENT is the sum of the citation-weighted patents applied for between 1998 and 2001, inclusive. This I-year lag is consistent with Hausman, Hall, and Griliches (1984), who showed that there is an almost contemporaneous relationship between alliance commencement and patent assignation.

Each time a new patent is assigned by the USPTO, the record of that patent includes information about the patents on which it was based. Thus, the number of citation-weighted patents after the alliance has been formalized provides us with a better indicator of R&D alliances outcomes than R&D spending, for example. Even though this is an approximation, patents are closely related to new product development (Comanor & Scherer, 1969), literature-based invention counts (Basberg, 1982), and nonpatentable innovations (Patel & Pavitt, 1997). Furthermore, because each patent cites previous patented inventions, this lineage can be used to determine the relative importance of the original patent, namely, the patents produced as a result of an alliance can be weighted to capture its relative importance. Empirical evidence shows a strong correlation between the ex post citations of the patent and the estimated value of the underlying invention (Trajtenberg, 1990). As such, citation weighting provides a less noisy measure of innovation than simple patent counts (Griliches, 1990). As Sampson (2005) suggested, we use the application date because this date is the earliest point at which we can identify new firm technology.

Independent Variable

Previous alliance experience (EXPERIENCE) is the focal independent variable and is measured as the number of alliances that a firm has been part of from 1990 up to but not including the year of the focal alliance. No restriction on the type of alliance was imposed on this data, and alliances for this time period include marketing, manufacturing, supply, or funding alliances. Any kind of alliance was selected here because firms learn to manage the coordination difficulties inherent in R&D alliances with any type of prior alliance experience rather than just prior R&D alliance experience. As noted by Sampson (2005),

With any type of alliance, firms learn how to coordinate

across organizational boundaries, select appropriate

contract structures, evaluate performance and

manage differences in corporate cultures. Coordination

across firm boundaries is always challenging and,

therefore, skills gained in improving this coordination

likely are gained from any type of alliance. (p. 1012)

This concern was, in any case, tested by considering only previous R&D alliances. Results do not exhibit any significant differences.

Given the nonlinear relationship of EXPERIENCE and the independent variable, the natural log is used as the independent variable.

Control Variables

There might be other factors influencing the patenting activity of a firm other than alliance managerial competence. For instance, it is possible that firms with greater experience are actually more capable firms in R&D. To control for several factors that might be influencing the results, several variables are included here. These variables will allow us to isolate other effects on the independent variable, thus allowing us to capture the hypothesized relationship. Control variables selected here are basically the same as those used by Sampson (2005), but they include some differences that are noted when appropriate.

Prealliance firm and partner patents (FIRM PATENT and PARTNER PATENT). These variables allow us to rule out other factors that may influence firm innovation. Particularly, the number of (firm's and partner's) prealliance patents has been used to capture independent technological capabilities (Patel & Pavitt, 1997), technological acquisitions, R&D spending, and a firm's propensity to patent (Trajtenberg, 1990). Thus, the inclusion of prealliance patents will allow us to control for firm individual R&D and technological efforts. This of course means that both firms' prealliance patent must be accounted for. Hence, two variables, FIRM PATENT and PARTNER PATENT, are included. Prealliance patent count was obtained from USPTO Web site for a 4-year period (1993 to 1996). These variables are not weighted because Trajtenberg (1990) found that "unweighted patent counts are more highly correlated with R&D spending and other R&D inputs than citation-weighted patent counts and, as such, are better measures of R&D inputs." For these variables once again, the total patenting activity of the focal and the parent firm were considered.

Alliance scope (SCOPE). This variable is intended to capture the breadth of R&D collaborative projects. R&D alliances have a wide range of purposes; they can have a very specific and narrow objective (e.g., a very specific software or application) or be broader and more general (e.g., to develop next-generation semiconductors). Thus, an alliance will exhibit different characteristics based on the breadth of the agreement. With this in mind, three dummy variables are selected using information in the SDC database. We look at the alliance description field in this database and assigned each alliance to one of three categories as described in Table 2. Because alliances with intermediate breadth are the most common, we omit SCOPE (Intermediate) from the empirical analyses.

Multilateral alliances (MULTILATERAL). This is also a dummy variable that discriminates alliances between two (bilateral) and alliances between three or more partners. Multilateral alliances might be an indication of larger objectives, and we expect them to be associated with greater returns. Thus, this possible effect is controlled, too. The number of total multilateral alliances for the sample period is 7, as represented in Table 1.

International alliances (INTERNATIONAL). When firms from different nations engage in an R&D alliance, there are some difficulties that they have to overcome (e.g., coordination efforts, cultural differences, etc.). These difficulties have an important impact on the alliance effectiveness, and they are here accounted for by the dummy variable INTERNATIONAL that equals 1 if the firms are from different countries and 0 otherwise.

Other concurrent alliance(s) (OTHER ALLIANCE). If a firm is involved in one or more alliances other than the focal alliance, that firm has to distribute its efforts to successfully coordinate activities with more than one partner. We expect this to have a negative impact on R&D collaboration. Once again, a dummy variable is constructed to control for this possible effect. OTHER ALLIANCE is a dichotomous variable that has a value equal to 1 if the firm is involved in one or more other alliances for the sample period (1997 to 1998), 0 otherwise.

Diversity of technological capabilities (TECH DIVERSITY). One of the main characteristics of collaborative projects is the fact that partners learn new capabilities from the ally, and thus, collaborative benefits increase as partners' capabilities differ because partners with dissimilar capabilities have much more to learn from each other. This must be controlled for to isolate the hypothesized relationship. Following Jaffe (1986) and Sampson (2005), we construct a measure of technological diversity based on the distribution of each firm's patents across different USPTO classifications. For each patent of a certain firm, we use the codifications contained in the class field in the USPTO Web site. This allows for the creation of a multidimensional vector of patents assigned to firms, [F.sup.s.sub.i] = ([F.sup.s.sub.i] ... [F.sup.s.sub.i]), where [F.sup.s.sub.i] represents the number of patents assigned to firm i in class s. With this information a composite index can be calculated as:

TECHDIVERSITY = 1 - ([F.sub.i][F.sub.j]')/[square root of ([F.sub.i][F.sub.i]')([F.sub.j][F.sub.j]')], i [not equal to] j

A value of 1 indicates the highest diversity possible between partners; 0 indicates high similarity.

It has been suggested that by using absorptive capacity, firms must have some capabilities in common to benefit from newly acquired knowledge. Thus, high technological diversity might actually hinder collaborative efforts. Following this line of arguments, the square of TECH DIVERSITY is used in the model.

Statistical Method

Because the dependent variable is a count of citation-weighted patents, this is undoubtedly a nonnegative, integer value. These types of variables also have high frequency of 0 and small integer values. Therefore, to take these characteristics of the sample into consideration, a negative binomial model is selected because 0 and small values are "naturally incorporated into the model" (Hausman et al., 1984).

The negative binomial is an extension of the Poisson model, but it relaxes the assumption of the equality of mean and variance. Therefore, it is a preferred method when Poisson estimation is inappropriate because of overdispersion. Overdispersion is said to be present when the variance is greater than the mean. This is the case in our study.

As such, the following model is used:

Pr[PATENT=p] = [e.sup.[lambda]][[lambda].sup.p]/p! (1)

where [lambda] is [e.sup.[beta]]X + z], [beta] is a vector of independent variables, and X is a vector of parameters. (6)

This particular model has independent errors, and parameter estimates do not require correction because maximum likelihood estimates are unbiased and consistent even if the assumption of independence is violated (Greene, 1993). Nevertheless, observation exhibits a lack of independence because for each firm there is the possibility of multiple observations if the firm participated in multiple alliances. This calls for a correction of the standard errors. A technique by Huber (1967) has been used in the past to correct for this lack of independence. This technique involves the creation of "super observation" calculated as the sum of the likelihood scores for each firm involved in more than one alliance. This value is then used to calculate the standard error.

Results

As reported in Table 1, 137 firms are included in the sample, 86 are involved in only one alliance, 7 are involved in two alliances, and 7 are involved in three or more alliances. Following Sampson (2005), we created a separate observation for each alliance a firm is involved in, leading to 182 observations. Table 3 shows descriptive statistics of the variables.

The estimation of the general model is set out in Model 1 of Table 4. This is the original model, and it considers the logarithm of alliance experience, EXPERIENCE (LOG), as the focal independent variable. From this table we can see that the count of alliance experience has a positive and significant relationship with firm innovation (p < .01). Using the median values of the independent and control variables, we evaluated the model E[PATENT] = [lambda] = [e.sup.[beta]'X], and then we increased the number of prior alliances in one, keeping the rest of the variables unaltered. This exercise produced a change in the expected number of patents equal to 2.56; however, this value was not significant at the .05 level.

We also considered the possibility of a linear relation between EXPERIENCE (COUNT) and PATENT, namely, a regression model was obtained (Model 2) using the original alliance experience variable. Table 4 shows again that the hypothesized relation is statistically significant at the .01 level.

Table 4 shows the model (Model 3) that considers the dummy variable EXPERIENCE > 0 as the independent variable. This variable has value 1 if the firm has prior experience, regardless of the exact number, and 0 otherwise. Once again, the relationship is significant at the .09 level, showing that indeed prior alliance experience per se has a positive impact on postalliance patenting activities. This is an important result because firms without experience should expect lower returns (in terms of innovation) than firms that have at least one alliance in their history. Following a previously described method to predict the change of patent output when prior experience goes from 0 to 1, a significant result is obtained. The predicted number of patents (E[PATENT]) increases in 53 patents, a number that is substantial if we take into account that the median value of PATENT is 82. This shows exact correspondence with Sampson's (2005) results and reinforces the notion that it is the existence rather than the amount of alliance experience that determines whether a firm will benefit from past alliance activity.

This of course calls for a more detailed analysis of the relationship between experience effects and patent activity. To achieve this objective, three dummy variables are used for a specific amount of previous alliances: EXPERIENCE: 1-5 has value 1 if the firm has 1 to 5 prior alliances, inclusive; EXPERIENCE: 6-10 is 1 if the firm has somewhere between 6 and 10 alliances; EXPERIENCE > 10 is 1 if the firm has more than 10 prior experiences. The results of this model are shown in Model 4. This model shows some interesting and unexpected results; interesting because they are in opposition to Sampson' (2005) results. For instance, where Sampson reported coefficients that are significant at the .05 level, the fitted model for the semiconductor industry shows that only firms with more than 10 prior alliances should expect a larger patent activity. This result suggests that the extent matters as much as the existence of alliance experience, at least for the chosen industry. Clearly, this contradicts Hypothesis 1 because more experience appears to be more important than the mere fact of having some experience.

Now we turn our attention to the point in time when prior alliance experiences took place. Here, the main concern is whether the recency of past experience has a role to play in the studied relation. In other words, we look at the possibility of recent experience's having a larger impact on new alliance formation (Sampson, 2005). With this in mind, the EXPERIENCE variable is decomposed into six variables, including EXPERIENCE: 1 YEAR to EXPERIENCE: 6 YEARS, where the first variable includes only the firm's alliances announced 1 year before the focal period of time (1997 to 1998), the second represents the number of alliances in the year 1995, and so on. The model is shown in Table 5, and the results suggest that neither very recent nor old alliance experience has much impact on the weighted number of patents. Only two variables have significant coefficients, EXPERIENCE: 2 YEARS and EXPERIENCE: 3 YEARS, thus indicating that only the number of prior alliances conducted 2 and 3 years previous to the time period selected for this study are significant at p < .05. EXPERIENCE: 4 YEARS and EXPERIENCE: 5 YEARS have negative coefficients even though these are not significant. Negative coefficients might indicate the presence of multicollinearity, thus we conducted a test to determine if this was the case in our sample. Indeed, a small multicollinearity problem is present for EXPERIENCE: 4 YEARS and above. This could be verified by dividing the sample in two groups and comparing the coefficients of each regression. Even though the differences are not significant, tolerance ([VIF.sup.-1]) in one case was low (< .1). Therefore, as a check for robustness, we grouped the experience in groups of 2-year periods and verified that multicollinearity was no longer an issue. The results for this regression are shown in Table 5, Model 6. In this case, instead of the original six variables, we used three dummy variables, each one accounting for 2 years (EXPERIENCE: YEARS 1 AND 2, EXPERIENCE: YEARS 3 AND 4, EXPERIENCE: YEARS 5 AND 6). The results in this case support the original claim and verify that recent experience is not significantly associated with higher innovative output, whereas experience that is a couple of years old appears to have an impact on the dependent variable. EXPERIENCE: YEARS 5 AND 6 does not show a statistically significant coefficient.

This hints at a possible inverted-U shaped relationship between the number of alliances at a specific point in time and the benefits a firm can gain in terms of new alliance success. This confirms, as Sampson (2005) predicted, that experience depreciates with time, but it does not support the claim that the more recent the alliance experience the better. Indeed, too recent an experience appears to have no effect on patenting activity in this industry.

Conclusion and Discussion

It has been shown that firms can indeed reduce production costs by means of experience, but it is still veiled if this premise applies to management practices and if so, how. In particular, R&D alliances are a managerial practice that research suggests is amenable to experience effects. This article builds on existing research that indicates that a firm's experience with alliances can increase its probability of conducting new ventures successfully. In a very recent study, Sampson (2005) looked at alliances conducted by firms in the telecomm equipment industry, and she reported that the experience effects, even when this refers to a single alliance, are significant and substantive. This claim is confirmed here as expected, but the nature of the relationship appears to be in contradiction to what was stated in the aforementioned analysis. Namely, prior alliance experiences are important only when these alliances took place 2 to 3 years prior to the focal alliance. For the semiconductor industry, very recent R&D alliance experiences seem to have no impact on the firm's innovativeness, measured as its postalliance patenting activity.

As for the depreciation of experience, there are some explanations that can be offered. First, there is the possibility of a rapid change in managerial practices contributing to the loss of benefits to be gained from lessons learned in the past. Technological industries such as telecommunication equipment and semiconductors are particularly susceptible to this kind of effect. An alternative explanation could be found in the rate of managerial turnover. Managers departing after conducting alliances might take with them the knowledge obtained in the alliance formation process, thus forcing the organization to face a new venture with little or no practical experience. The implications of either explanation are clear for a firm that is looking for a way to enhance its innovativeness through a collaborative project. These firms need to ensure the assimilation of prior alliance experiences to benefit from them. One way to do so is the creation of "alliance management offices" within the organization. Of course, the existence of this organizational function is still relatively new, and the real impact on alliance management practices are still to be established as the next step for the line of research proposed in this article. Moreover, and in relation to the lack of impact of recent experience discussed in the next section, these offices could not only reduce the pace of depreciation of experience, but they might also contribute to reduce the period of time for recent experience to be a contribution to new ventures.

More elusive is the explanation for the lack of significance of the most recent experience on patenting activity. One possible explanation is again, the process of assimilation and diffusion of past experience within the organization, namely, firms might need more time to assimilate the lessons obtained from past experiences, particularly if they are conducting several alliances at the same time. The semiconductor industry shows a large number of alliances both before and during the focal period that might help to explain the delayed impact of alliance experiences when compared to Sampson's (2005) results. On average, the firms in this sample have a mean experience of 10 previous alliances and a median value of 11. Moreover, during the focal period of time (1997 to 1998), 137 firms were involved in 182 alliances, thus suggesting that a single firm initiated more than one alliance at the same time. A firm conducting parallel alliances has to devote its resources to more than one task, and this could be translated into a larger period for the consolidation of the alliance. Naturally, this conclusion calls for a more detailed analysis at the organizational level to understand the processes that contribute and those that hinder the organization from using more recent experience.

There are many limitations to this study. Besides the fact that the sample size must be increased by looking at a longer time period, one major limitation is that this study assumes that each prior alliance has the same weight on the firm's ability to conduct new collaborative ventures successfully. Given the large number of alliances in the selected industry, this is of particular importance because a firm will not deem each alliance as equally important, and consequently, it will not deploy the same amount of resources. Second, this study focuses exclusively on R&D alliances. Future research should focus on other kinds of alliances, such as manufacturing and marketing alliances.

Despite the limitations, this article offers an interesting extension to previous studies on experience effects and shows that prior experience does matter, but only when it is relatively recent. Generalizability concerns are clearly prominent, but when we consider this study and Sampson's (2005) together, a stronger case can be made in favor of experience effects and their depreciation over time. Nevertheless, one remarkable difference, one that calls for further research, is the fact that the recency of the experience might be moderated by the type of industry. One possible way to answer this question is to conduct a cross-sectional analysis that looks at different industries simultaneously. Finally, an important managerial implication is the evidence that suggests that firms might have something to gain from the institutionalization of alliance management offices.

References

Anand, B. N., & Tarun, K. (2000). Do firms learn to create value? The case of alliances. Strategic Management Journal, 21, 295--316.

Argote, L., Beckman, S. L., & Epple, D. (1990). The persistence and transfer of learning in industrial settings. Management Science, 36, 140-154.

Basberg, B. (1982). Technological change in the Norwegian whaling industry: A case study in the use of patent statistics as a technology indicator. Research Policy, 11, 163-171.

Barnett, W. P., & Hansen, M. T. (1996). The Red Queen in organizational evolution. Strategic Management Journal, 17, 139-157.

Baum, J. A. C., & Ingram, P. (1998). Survival-enhancing learning in the Manhattan hotel industry, 1898-1980. Management Science, 44, 996-1016.

Bleeke, J., & Ernst, D. (1993). Collaborating to compete. New York: John Wiley.

Chang, S. J. (1995). International expansion strategy of Japanese firms: Capability building through sequential entry. Academy of Management Journal, 38, 383-407.

Comanor, W., & Scherer, F. M. (1969). Patent statistics as a measure of technical change. Journal of Political Economy, 77, 392-399.

Darr, E. D., Argote, L., & Epple, D. (1995). The acquisition, transfer, and depreciation of knowledge in service organizations: Productivity in franchises. Management Science, 41, 1750-1762.

Dutton, J. M., & Thomas, A. (1984). Treating progress functions as a managerial opportunity. Academy of Management Review, 9, 235-247.

Dutton, J. M., Thomas, A., & Butler, J. E. (1984). The history of progress functions as a managerial technology. Business History Review, 58, 204-233.

Greene, W. H. (1993). Econometric analysis (2nd ed.). Upper Saddle River, NJ: Prentice Hall.

Griliches, Z. (1990). Patent statistics as economic indicators: A survey. Journal of Economic Literature, 28, 1551-1707.

Hannan, M. T., & Freeman, J. (1984). Structural inertia and organizational change. American Sociological Review, 49, 149-164.

Harbison, J. R., & Pekar, P., Jr. (1998). Smart alliances: A guide to repeatable success. San Francisco: Jossey-Bass.

Harrigan, K. R. (1985). Strategies f or joint ventures. Lexington, MA: Lexington Books.

Haspeslagh, P., & Jemison, D. (1991). Managing acquisitions: Creating value through corporate renewal. New York: Free Press.

Hausman, J., Hall, B. H., & Griliches, Z. (1984). Econometric models for count data with an application to the patents-R&D relationship. Econometrica, 52, 909-938.

Huber, P. J. (1967). The behavior of maximum likelihood estimates under non-standard conditions. Proceedings of the Berkeley Symposium on Mathematical Statistics and Probability, 1, 221-233.

Jaffe, A. B. (1986). Technological opportunity and spillovers of R&D: Evidence from firms' patents, profits, and market values. American Economic Review, 76, 984-1001.

Leonard-Barton, D. (1992). Core capabilities and core rigidities: A paradox in managing new product development. Strategic Management Journal, 13, 111-125.

Levitt, B., & March, J. G. (1988). Organizational learning. Annual Review of Sociology, 14, 319-340.

Lieberman, M. B. (1984). The learning curve and pricing in the chemical processing industries. Rand Journal of Economics, 15, 213-228.

Lieberman, M. B. (1987). The learning curve, diffusion, and competitive strategy. Strategic Management Journal, 8, 441-452.

Macher, J. T. (2004). Technological development and the boundaries of the firm: A knowledge-based examination in semiconductor manufacturing. Management Science, 52, 826-843.

March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science: A Journal of the Institute of Management Sciences, 2, 71-87.

Nerkar, A., & Roberts, P. (2004). Technological and product-market experience and the success of new product introductions in the pharmaceutical industry. Strategic Management Journal, 25, 779-799.

Oxley, J. E., & Sampson, R. C. (2004). The scope and governance of international R&D alliances. Strategic Management Journal, 25, 723-749.

Patel, P., & Pavitt, K. (1997). The technological competencies of the world's largest firms: Complex and path-dependent, but not much variety. Research Policy, 26, 141-156.

Sampson, R. C. (2004). Organizational choice in R&D alliances: Knowledge-based and transaction cost perspectives. Managerial and Decision Economics, 25, 421-436.

Sampson, R. C. (2005). Experience effects and collaborative returns in R&D alliances. Strategic Management Journal, 26, 1009-1031.

Trajtenberg, M. (1990). A penny for your quotes: Patent citations and the value of innovations. RAND Journal of Economics, 21, 172-187.

Notes

(1.) Sampson's (2005) study considered patents between 1988 and 1996.

(2.) SIC classification code 3674.

(3.) Securities Database Corporation database also contains information on mergers and acquisitions.

(4.) Web site is http://www.uspto.gov

(5.) Web site is http://www.corporateaffiliations.com/

(6.) This model was estimated using STATA and the nbreg procedure.

Jaime C. Rubin de Celis is a PhD candidate at the University of Pittsburgh. Prior to returning to the University of Pittsburgh he served as a professor of project management and strategic management at Santa Maria University at both the Santiago, Chile and Guayauuil, Ecuador campuses.

John Lipinski is a PhD candidate at the University of Pittsburgh. He currently serves as an assistant professor at Robert Morris University in Moon Township, PA.

Jaime C. Rubin de Celis

University of Pittsburgh, Pennsylvania

John Lipinski

Robert Morris University, Moon Township, Pennsylvania Table 1 Characteristic Values of the Sample Number of alliances 86 International alliances 34 (40%) Multilateral alliances 7 (9%) Number of firms 137 (65% United States, 15% Japan,

9% Europe, 10% Central Asia) Number of nations 18 Table 2 Alliance Scope Categories Category Description Narrow Alliances focused on development of new

products based closely on existing technology Intermediate Alliances that do not fall into either of the other

categories Broad Alliances that are intended to produce next-

generation products Table 3 Descriptive Statistics

1 2 3 1. PATENT 1.000 2. EXPERIENCE (LOG) 0.308 *** 1.000 3. PATENT (FIRM) 0.966 *** 0.276 *** 1.000 4. PATENT (PARTNER) 0.134 ** 0.014 0.084 5. TECH DIVERSITY -0.086 -0.081 -0.055 6. TECH DIVERSITY2 -0.104 -0.085 -0.075 7. INTERNATIONAL 0.129 * 0.169 ** 0.150 ** 8. MULTILATERAL 0.121 0.062 0.105 9. OTHER ALLIANCE 0.261 *** 0.103 0.254 *** 10. YEAR98 0.106 0.029 0.104 11. SCOPE NARROW -0.015 0.067 -0.012 12 SCOPE BROAD -0.012 -0.034 -0.018 Mean 1,683.63 9.92 1,331.98 Median 82.00 11.00 81.00 Minimum 0.00 0.00 0.00 Maximum 8,002.00 26.00 7,523.00 Standard deviation 2,496.82 7.34 2,029.87

4 5 6 1. PATENT 2. EXPERIENCE (LOG) 3. PATENT (FIRM) 4. PATENT (PARTNER) 1.000 5. TECH DIVERSITY 0.046 1.000 6. TECH DIVERSITY2 0.037 0.996 *** 1.000 7. INTERNATIONAL 0.084 0.080 0.072 8. MULTILATERAL -0.054 0.079 0.079 9. OTHER ALLIANCE 0.083 -0.210 *** -0.218 *** 10. YEAR98 0.090 -0.059 -0.060 11. SCOPE NARROW -0.024 0.079 0.071 12 SCOPE BROAD 0.164 ** 0.074 0.075 Mean 1,165.74 0.90 0.83 Median 42.50 0.95 0.91 Minimum 0.00 0.43 0.19 Maximum 11,367.00 1.00 1.00 Standard deviation 2,205.44 0.13 0.21

7 8 9 1. PATENT 2. EXPERIENCE (LOG) 3. PATENT (FIRM) 4. PATENT (PARTNER) 5. TECH DIVERSITY 6. TECH DIVERSITY2 7. INTERNATIONAL 1.000 8. MULTILATERAL -0.204 *** 1.000 9. OTHER ALLIANCE -0.069 -0.073 1.000 10. YEAR98 -0.052 0.072 -0.015 11. SCOPE NARROW 0.032 -0.031 -0.055 12 SCOPE BROAD -0.114 -0.014 0.117 Mean 0.38 0.13 0.86 Median 0.00 0.00 1.00 Minimum 0.00 0.00 0.00 Maximum 1.00 1.00 1.00 Standard deviation 0.49 0.34 0.35

10 11 12 1. PATENT 2. EXPERIENCE (LOG) 3. PATENT (FIRM) 4. PATENT (PARTNER) 5. TECH DIVERSITY 6. TECH DIVERSITY2 7. INTERNATIONAL 8. MULTILATERAL 9. OTHER ALLIANCE 10. YEAR98 1.000 11. SCOPE NARROW -0.004 1.000 12 SCOPE BROAD -0.115 -0.217 *** 1.000 Mean 0.29 0.29 0.14 Median 0.00 0.00 0.00 Minimum 0.00 0.00 0.00 Maximum 1.00 1.00 1.00 Standard deviation 0.46 0.45 0.35 Significant at * 10%, ** 5%, and *** 1% level for one-tailed tests. Table 4 Previous Alliance Experience and Innovative Performance

1 2 INTERCEPT -3.967 -3.647 EXPERIENCE (LOG) 0.365 *** EXPERIENCE (COUNT) 0.058 *** EXPERIENCE > 0 EXPERIENCE = 1-5 EXPERIENCE = 6-10 EXPERIENCE > 10 PATENT (FIRM) 0.001 *** 0.001 *** PATENT (PARTNER) 0.000 0.000 TECH DIVERSITY 25.381 *** 24.924 *** TECH DIVERSITY2 -18.839 *** -18.630 *** INTERNATIONAL 0.203 0.185 MULTILATERAL 0.741 ** 0.798 ** OTHER ALLIANCE 0.314 0.321 YEAR98 -0.012 -0.037 SCOPE NARROW 0.037 0.009 SCOPE BROAD 0.673 0.637 n 182 182 Wald chi-square 201.54 207.02

3 4 INTERCEPT -2.901 -3.649 EXPERIENCE (LOG) EXPERIENCE (COUNT) EXPERIENCE > 0 0.910 * EXPERIENCE = 1-5 0.776 EXPERIENCE = 6-10 0.441 EXPERIENCE > 10 1.489 ** PATENT (FIRM) 0.001 *** 0.001 *** PATENT (PARTNER) 0.000 0.000 TECH DIVERSITY 21.313 ** 23.099 *** TECH DIVERSITY2 -17.050 ** -17.130 *** INTERNATIONAL 0.247 0.189 MULTILATERAL 0.823 ** 0.670 ** OTHER ALLIANCE 0.514 0.220 YEAR98 -0.068 -0.039 SCOPE NARROW 0.202 0.084 SCOPE BROAD 0.681 0.583 n 182 182 Wald chi-square 174.43 178.34 Note: Negative binomial estimation. Dependent variable is citation-weighted patents issued to each firm in a postalliance period. Positive coefficients indicate increased patent output. Significant at * 10%, ** 5%, and *** 1% level for one-tailed tests. Table 5 Previous Alliance Experience and Innovative Performance

5 6 INTERCEPT -1.986 -1.635 EXPERIENCE: 1 YEAR 0.067 EXPERIENCE: 2 YEARS 0.162 ** EXPERIENCE: 3 YEARS 0.196 *** EXPERIENCE: 4 YEARS -0.112 EXPERIENCE: 5 YEARS -0.011 EXPERIENCE: 6 YEARS 0.063 EXPERIENCE: YEARS 1 AND 2 0.122 EXPERIENCE: YEARS 3 AND 4 0.096 ** EXPERIENCE: YEARS 5 AND 6 0.013 PATENT (FIRM) 0.001 *** 0.001 *** PATENT (PARTNER) 0.000 0.000 TECH DIVERSITY 22.589 ** 22.086 ** TECH DIVERSITY2 -17.257 *** -17.421 *** INTERNATIONAL 0.068 0.069 MULTILATERAL 0.805 ** 0.791 OTHER ALLIANCE 0.091 0.086 YEAR98 0.039 * 0.049 * SCOPE NARROW -0.128 -0.118 SCOPE BROAD 0.592 0.799 n 182 182 Wald chi-square 221.22 221.54 Note: Negative binomial estimation. Dependent variable is citation-weighted patents issued to each firm in a postalliance period. Positive coefficients indicate increased patent output. Significant at * 10%, ** 5%, and *** 1% level for one-tailed tests.