Ajay Agrawal, Avi Goldfarb, Florenta Teodoridis. 2013. Does knowledge accumulation increase the returns to collaboration?. Working Paper 19694. National Bureau of Economic Research
http://www.nber.org/papers/w19694
Abstract:We conduct the first empirical test of the knowledge burden hypothesis, one of several theories advancedto explain increasing team sizes in science. For identification, we exploit the collapse of the USSR asan exogenous shock to the knowledge frontier causing a sudden release of previously hidden research.We report evidence that team size increased disproportionately in Soviet-rich relative to -poor subfieldsof theoretical mathematics after 1990. Furthermore, consistent with the hypothesized mechanism,scholars in Soviet-rich subfields disproportionately increased citations to Soviet prior art and becameincreasingly specialized.
摘要：我们首次进行了对“知识负担”假说的实证研究。这一假说是几个先进的解释科学研究团队规模扩大现象的理论之一。为了证明这一假说，我们将苏联解体——这一事件导致了先前被隐藏的研究的突然释放——作为一个对知识前沿的外生冲击。我们报告了有证据表明：在1990年以后，在理论数学研究中苏联历史上占优势的子领域相对于其历史上不占优势的子领域，团队规模有所增加。此外，在苏联历史上占优势的子领域的学者也不成比例地增加了对苏联之前研究的引用，并且在研究上变得越来越专门化。
数据来源：本文研究了1970-2010这41年间理论数学领域的期刊发表数据。数据库为美国数学社会数据库（American Mathematical Society）。
研究方法：工具变量法（以“苏联解体”这一事件为工具变量，与知识前沿扩张有关，同时与合作率变化无关）；双重差分模型(difference-in-differences model)以“团队规模”为因变量。以“苏联历史上在该领域是否占优势”为0-1虚拟变量，代表了知识前沿受冲击的程度。以“铁幕是否结束”为0-1虚拟变量，代表了苏联解体前后的两个时期。
Discussion and conclusion:We report evidence that an outward shift in the knowledge frontier is associated witha subsequent increase in research team size and researcher specialization. Importantly, thisevidence is consistent with the knowledge frontier explanation but not the other explanationsfor the widely documented increase in team size. In other words, although this evidence isnot intended to (and does not) rule out the possibility that the other explanations also playa role, it suggests that the knowledge frontier hypothesis is a plausible explanation for atleast some of the observed increase in team size in science.
In our setting, a back-of-the-envelope calculation indicates that the knowledge frontiereffect accounts for 24% of the increase in team size in Soviet-rich fields in theoretical mathematics. We calculate this as follows: team size in Soviet-rich fields increased by 33%, from1.34 to 1.78, in the before versus after period. We estimate that the Soviet-rich fields experienced an 8% disproportionate increase (relative to Soviet-poor) during this period (Table 3,Column 1). This represents 24% of the overall percentage increase. While this rough calculation can be seen as a lower bound because it assumes none of the increase in Soviet-poorsubfields was due to an outward shift in the knowledge frontier, we resist this interpretationbecause of the numerous other assumptions underlying the 24% value.
More broadly, it is important to clarify the limitations of our test of the knowledge burdenhypothesis. First, we test a particular implication of the knowledge burden hypothesis: theimpact of a sudden outward shift in the knowledge frontier on collaboration and specialization. An underlying assumption of this interpretation of our estimates is that the team sizeresponse to a shock is similar to that for a gradual outward shift in the knowledge frontier.However, that may not be the case. Researchers may be able to absorb gradual increases inthe knowledge frontier in a manner that does not generate as high returns to collaboration asthose resulting from a sudden shock that may be more costly for researchers to internalize.Thus, our empirical results may not measure the impact of a gradual shift in the knowledgefrontier.
Second, there may have been other impacts of the collapse of the Soviet Union on the field of mathematics. Borjas and Doran (2012, 2013a, 2013b) emphasize the labor marketimpact of increased competition from Soviet scholars. This increased competition also mayhave driven an increase in collaboration if, for example, returns to collaboration increaseddue to reasons such as risk mitigation (diversification of research projects). While we viewour results on Japanese publications, citations to Soviet prior art, and specialization as moreconsistent with the knowledge burden hypothesis, we cannot definitively reject the possibilitythat changing labor markets also played a role.
Third, we focus on one particular field: mathematics. Adams et al (2005) show thatmathematics is somewhat of an outlier in team size relative to other disciplines in havingrelatively small teams. In the first year of their study, 1981, mathematics publicationshad the fewest number of authors (of 12 fields). Furthermore, mathematics had the lowestannual growth rate in team size from 1981 to 1990 and the second lowest from 1990 to 1999.In contrast, physics and astronomy had the highest growth rates, which likely was at leastpartly driven by the increasing role of capital-intensive equipment (e.g., particle accelerators)in those fields. Therefore, even if 24% is a reasonable lower-bound estimate of the fractionof the percentage increase in team size caused by an outward shift in the knowledge frontierin mathematics, it may be an overestimate in fields where capital equipment plays a morecentral role.
Overall, we document that the knowledge shock caused by the exogenous collapse ofthe Soviet Union led to a disproportionate increase in collaboration among non-Soviet researchers in those subfields in which Soviet mathematicians were strongest relative to othersubfields of theoretical mathematics. Our examination of citations to Soviet prior art, specialization, and team sizes in Japan provides further evidence consistent with the burdenof knowledge hypothesis: a knowledge shock leading to increased specialization and collaboration. In a series of papers (2009, 2010, 2011), Jones presents a variety of interventionsthat are potentially welfare-enhancing in the presence of a knowledge frontier effect. Theseinclude subsidies and rewards to incentivize entry into research careers, team-based evaluation of grant applications, and national or regional subsidies and specialization to preventpoverty traps due to underinvestment in human capital from coordination failures arisingfrom thin markets for complementary skills. Although our study offers no means by which tocomment on the suitability of these interventions to particular policy settings, the evidencewe present here does suggest that the knowledge frontier effect is worthy of further researchand possibly policy attention.
讨论与结论：我们发现有证据表明知识前沿的向外扩张和随后的研究团队规模扩大、研究者专门化的现象有联系。尤为重要的是，这一证据和知识前沿解释相一致，但和其他对于被文献广泛提及的团队规模增长的解释无关。换句话说，虽然这一证据并不排除其他方面的解释发挥作用的可能性，但它表明知识前沿假说至少对于某些观察到的科学研究团队规模增长是可信的解释。
在我们的研究中，一项并不复杂的计算表明知识前沿效应占据了理论数学苏联历史上占优势的子领域的研究团队规模增长的24%。我们的计算步骤如下：在苏联解体前后的两个时期对比中，苏联历史上占优势的子领域的研究团队规模增长了33%，从1.34人到1.78人。我们估计在此期间这些领域经历了8%的不成比例的增长（相对于苏联历史上不占优势的子领域）。这代表了总百分比上有24%的增长。这项粗略的计算可以被视为约束比较少的情形，因为它假设在苏联历史上不占优势的子领域，没有任何团队规模的扩展被归因于知识前沿的向外扩张。我们接受这一演绎因为无数其他的假设要以24%的值为基础。
更概括来讲，阐明我们对于“知识负担”假说检验的局限性非常重要。首先，我们检验的是“知识负担”假说的一项特别应用：突然的知识前沿向外扩张对于合作和研究专门化的影响。但是，这也许并非事实的全部。研究者或许可以通过不同于成本高昂的突然冲击这一方式，而是逐渐缓慢地吸收知识的增长以实现国际化。因此，实证结果也许没有测量知识前沿缓慢移动的影响。
第二，苏联解体这一事件对数学领域可能有其他影响。Borjas和Doran强调了苏联学者带来更多的竞争对劳动力市场的冲击。如果科研合作因为例如风险降低（研究项目多样化）等原因而增加，那么竞争的增加也可能导致合作的增加。当我们用日本的出版物来检验我们的结果时，对苏联先前研究的引用、研究专门化的现象都和“知识负担高度相关”，我们不能最终拒绝劳动力市场变化也发挥作用的可能性。
第三，我们聚焦于一个特别的领域：数学。Adams et al的研究展示了相对于其他限制而言，相对小的团队规模对于数学而言并没有显著影响。在他们研究的第一年，1981年，数学刊物在12个研究领域中作者最少。另外，从1981年到1990年，数学研究团队规模的年增长率最低；从1990年到1999年则第二低。相反，物理学和天文学的研究团队规模有着最高的增长率，这一现象可能至少受到了资金密集型的设备的部分影响（如粒子加速器）。因此，尽管24%是对知识前沿向外扩张所带来的影响的粗略计算，在一些资金设备发挥主导作用的领域，它也可能被高估了。
总的来说，我们证明了苏联解体这一外生事件导致的知识冲击引发了非苏联籍的研究者在苏联理论数学研究史上占优势的子领域的合作不成比例的增加。我们对于引用苏联先前研究、研究专门化、日本研究团队的规模的检验进一步证明，现实和“知识负担”假说相符：知识冲击导致研究专门化和合作现象的增加。在一系列研究中，Jones展示了多种在知识前沿效应下潜在使福利增强的干涉变量。这包括了使人们进入研究生涯的补贴、奖金等物质激励，以团队为基础的拨款申请评估，国家或地区的补贴和专业化以防止补充技能的狭小市场协调失灵下的人力资本投资不足引发的贫困陷阱。虽然我们的研究没有提供评价特殊政策干预因素合理性的渠道，但我们提供的证据表明知识前沿效应值得更深入的研究和更大的政策关注度。
（By 苏小舟） |
