Not Your Stepping Stone: Collaboration and the Dynamics of Industry Evolution in Biotechnology
Download 154.14 Kb.
|
- Navigate this page:
- FINDINGS
RESEARCH METHODSTo test the arguments developed above, we utilize within firm statistics, employing a panel regression model. We are not comparing large pharmaceutical corporations to small biotech companies, nor are we measuring the relative mix of activities performed internally or externally. Both assessments would, of course, be interesting but would entail exceptional data access. Rather we focus on the more tractable question of whether and how biotech companies change the amount, variety, and scope of collaborative arrangements they engage in over the period 1988-99. To eliminate any spurious effects due to differences between firms, we incorporated a fixed effects, or dummy variable, model. Consequently, the estimated coefficients will capture only the amount by which a dependent variable shifts within a firm in response to a preceding change in that firm’s predictor variable. This fixed-effects approach is preferable to the alternative random-effects, because the predictor variables are likely to be correlated with the firm effects and we have the population of biotech firms over our observed time period and not a random sample. We are interested in estimating a dynamic model, in which the independent variables are lagged one year. Some firms may be “imprinted” or otherwise start on developmental trajectories before the start of our observation period. The trajectories may have naturally evolving patterns that change over time in coherent ways, but ones that we cannot foresee or measure. The omission of an important factor that changes over time within firms will result in autocorrelated errors and may bias estimates of the parameters in which we are most interested. One way of breaking the correlation over time, so as not to overestimate the effects of our hypothesized independent variables, is to sample on a lagged dependent variable, y. That is, our arguments involve whether firms that have grown, matured, or proven successful by using collaboration then retreat from or expand their network of involvement. We are not, here, concerned with the antecedents of a firm’s initial collaboration. For example, in testing our hypotheses predicting number and depth of ties at t based on size, age, or success at t-1, we use only those firms who had ties at t-1. A second theoretical consideration is that the dynamics we are studying involve the co-evolution of firms and networks. This process leads to an additional source of statistical nonindependence across our observations. Hence, we need to control for effects that vary over time but are constant across firms, such as the overall number of outside partners, the density of the industry’s network, government budgets for medical research, or the economic circumstances of pharmaceutical companies. To do so, we included fixed year effects: a dummy variable for each year. FINDINGSIn table 3 we present the effects of firm growth on depth of collaboration, to test hypothesis 1. Recall that we wish to assess whether firms that use collaboration to overcome liabilities of smallness subsequently choose to go it alone, or whether they deepen their collaborations. Consequently, we include in the analysis only firms that have at least one prior tie. We use three measures of depth of collaborations: the overall number of ties, the number of types of activities pursued with those ties, and the number of complex ties. The first row of the table contains GLS regression estimates of the effect of size at time t-1 on three measures of the depth of collaboration at time t, each in a separate column. The standard errors for the estimates are presented just below the estimates in parentheses. Also presented for each model are the within-firm and full r-squares, the number of records, and the number of firms involved in those records. The within-firm r-square indicates, for each measure of depth, the amount of variance over time from a firm’s average on each measure that is explained by size. The full r-square explains the amount of overall variation in depth, both between and within firms, that is explained by size and by the fixed firm and year effects that are included as controls. The results are clear and consistent across the three measures of depth. In each case, as a firm grows, it subsequently deepens its portfolio of collaborations. For the overall number of ties and the number of complex ties, the effects are significant beyond the .0001 level. For each of these measures, prior size explains about 10% of the within-firm variation in subsequent collaboration. Prior size explains only about 3% of the within-firm variation in the subsequent scope of production activities; however, the effect is significant beyond the .001 level. Given the restricted range for the number of types of tie activity, ranging from 1 to 6, there is doubtless some attenuation. In Tables 4 and 5, we turn to the effects of age on the scope and reach of collaboration. In Table 4, we test whether firms that use collaboration to overcome liabilities of newness subsequently choose to restrict or expand the types of activities they collaborate on. To do so, we separately model the initial use of each type of collaborative activity, including in each analysis firms that only have ties of other types. Recall that the six types of activity, presented in Table 1, are: R&D, Finance, Licensing, Evaluation, Commercialization (Manufacturing, Marketing, etc), and Complex. The GLS estimates (and their standard errors) of the effect of prior age on subsequent initial use of each type of collaborative activity are found in the first row of Table 4. There are six columns, each presenting a separate model for the type of activity listed. The table also contains the within-firm and full r-squares, as described above, along with the number of records and number of firms that meet the inclusion criteria for each model. Aging consistently increases the scope of activities for which firms use collaboration. For all six types of activity, among firms that only have other types of ties, the effect of prior age on subsequent number of ties of that type is positive. The results are significant beyond the .0001 level for all types, except evaluation, where the effect is significant beyond the .05 level. In Table 5, we ascertain whether firms that utilize alliances to overcome liabilities of newness subsequently rely on the same partners for each type of activity, or whether they broaden the reach of their affiliations. To do so, we separately model the number of kinds of partners used for each type of collaborative activity, including in each analysis only firms that previously have at least one kind of partner for that type. The identity of partners is defined by their industry or sector, including government, university, nonprofit, pharmaceutical, venture capital, biotech, etc., as suggested in column two of Table 1. The GLS estimates (and their standard errors) of the effect of prior age on subsequent number of kinds of partners for each type of collaborative activity are found in the first row of Table 5. There are six columns, each presenting a separate model for the type of activity listed. The table also contains the within-firm and full r-squares, as described above, along with the number of records and number of firms that meet the inclusion criteria for reach model. Aging increases the reach of collaboration for all types of ties save complex, where the effect is positive, but not statistically significant. For each of the remaining types of activity, among firms that have prior ties for that type of activity, the effect of prior age on the subsequent number of kinds of partners used for that particular activity is positive. The results are significant beyond the .0001 level for finance and licensing, significant beyond the .001 level for commercialization, and beyond .01 and .05 for R&D and evaluation, respectively. In table 6 we present the effects of firm success on involvement in collaborative R&D, in order to test hypothesis 3. Here we test whether firms that initially get involved in collaborative R&D to fund expensive research subsequently choose to emphasize exploitation by taking R&D inside, or whether they reinvest in exploration collaboratively. We include in the analysis only firms that have at least one prior R&D alliance. Success is indicated by a firm having reported positive earnings. Our measure of involvement in collaborative R&D is simply the number of R&D alliances. The GLS estimate (and its standard error) for the effect of positive earnings at time t-1 on number of R&D alliances at time t is found in the first row of Table 6. The estimate demonstrates that when firms that are engaged in collaborative R&D cross the earnings threshold, they subsequent add, on average, two more R&D alliances to their portfolio of collaborations. This positive effect is significant beyond the .0001 level. Table 2 showed the weighted-average number of R&D ties for a firm to be just under 2. This reinvestment, therefore, is not only statistically significant, but substantial from a practical standpoint as well. Turning to hypothesis four, concerning the division of labor within the industry, our aim is to discern whether BFs play a specialized, boutique role or are active in multiple stages of production. We also assess whether there is an elite cadre of either BFs or partners that dominate all types of activity. Finally, we look to see if there is evidence as to whether BFs play an orchestrating role in knitting together the industry. With respect to hypothesis 4A, whether BFs are involved extensively rather than narrowly, 27 biotech firms have collaborations for all six types of activities, while another 128 are involved in five types of collaborations. Thus 32% of the biotech firms have interorganizational ties for at least five different business activities. The extensive involvement of BFs in a wide array of ties suggest that BFs are not just specialized participants. On the other hand, among the partners there are 25 organizations that have ties to BFs for all six activities. Recognize that this is not a symmetrical comparison because while our data cover biotech firms’ ties to all partners, we have only the partners’ ties to biotech firms and not to one another. Still, the composition of the 25 most active organizations is interesting. First, four of the larger biotech firms (Chiron, Genentech, Genzyme, and Immunex) are among this group. Key branches of government – the National Institutes of Health, the U.S. Army, and the National Cancer Institute (a division of NIH) – are included too. Chemical and healthcare companies, such as Mitsubishi, DuPont, BASF, Hoechst, Kodak and Johnson and Johnson, Proctor and Gamble, and Baxter Travenol are deeply involved. Finally, the large pharmaceuticals – Novartis, Pfizer, Merck, Hoffman La Roche – are included, but by no means are all of the largest pharmaceuticals present. In short, both the biotech participants and the partner group show considerable variety rather than specialization. Turning to hypothesis 4B, regarding whether there is a heterogeneous group of players for each stage in the development process, the answers are a bit more complex. On the biotech side, just thirteen companies have an average of two or more ties for each activity in all twelve years. This suggests there may be a small group with extensive involvement in all phases of development. But there are another thirty companies with at least one type of tie every year. Recall, also, that there are but 180 firms in existence for the entire period 1988-1999. So if there is an inner circle, it is a fairly large one. On the partner side, because the competencies are diverse in purpose, we would not necessarily expect a tight inner core. Investment activities are dominated by venture capital firms and financial institutions. Universities play a significant role in licensing. Government institutes are clearly dominant in research collaborations. Several smaller biotechs have become specialists in running clinical trials. And large pharmaceuticals loom large in importance on the manufacturing side. Thus neither among biotechs nor among partners is there an identifiable tight inner core. The linking pin role, hypothesis 4C, is harder to assess with descriptive statistics. But consider that when we count which organizations are among the most active in terms of ties to biotech firms, that is, in the top 10% for 3 or more business functions, 20 organizations emerge. Among these 20 are the ever-present NIH and eight large pharmaceutical corporations, including Novartis, Lilly, SmithKline, Merck, Bristol Meyers, Schering Plough, Hoffman La Roche, and Boehringer. (But note the absence of such giants as Warner Lambert, Pfizer and Glaxo Wellcome.) There are also eight biotechs among this group. The largest biotechs – Genzyme, Genentech, and Chiron are there, but so is Genetics Institute, which was acquired by American Home Products. And present as well are several new startup firms, including ArQule, founded in 1993, and Metra, founded in 1990. Thus with regard to hypothesis 4C, concerning whether biotechs play a subsidiary role or a linking role, we find biotechs – both large and small – playing active and diverse roles alongside government institutes and pharmaceutical companies. Download 154.14 Kb. Share with your friends:
|