Patterned interactions in complex systems: Implications for exploration
Patterned interactions in complex systems: Implications for exploration
复制标题
DOI：
10.1287/mnsc.1060.0626
复制
发表时间：
2007-07-01
影响因子：
5.4
通讯作者：
Siggelkow, Nicolaj
中科院分区：
文献类型：
--
作者：
Rivkin, Jan W.;Siggelkow, Nicolaj
将组织、社会和技术系统视为一组相互依赖的决策的学者越来越多地使用生物和物理科学的模拟模型来检查系统行为。这些模型揭示了一个持久的管理问题：需要进行多少探索才能发现良好的决策配置？这些模型表明，随着决策之间的相互作用加剧和局部最优扩散，需要更广泛的探索。然而，模型通常假设决策之间的相互作用是随机分布的。与这一假设相反，最近对真实组织、社会和技术系统的实证研究表明，决策之间的相互作用是高度模式化的。集中化、小世界联系、幂律分布、等级制度和优先依恋等模式很常见。我们将这些模式嵌入到 NK 模拟模型中，并获得了引人注目的结果：在决策之间的交互总数保持固定的情况下，交互模式的变化可以使局部最优的数量改变一个数量级以上。因此，面对某些交互模式比面对其他模式时更广泛探索的长期价值要大得多。我们制定了简单、直观的经验法则，使决策者能够检查两种交互模式，并确定哪种模式值得在广泛的探索中进行更多投资。我们还发现，在保持交互模式固定的情况下，无论模式如何，交互数量的增加都会增加局部最优的数量。这验证了先前关于交互数量的比较静态结果，但强调了早期工作中一个重要的隐含假设——随着交互数量的增加，潜在的交互模式保持不变。
Scholars who view organizational, social, and technological systems as sets of interdependent decisions have increasingly used simulation models from the biological and physical sciences to examine system behavior. These models shed light on an enduring managerial question: How much exploration is necessary to discover a good configuration of decisions? The models suggest that, as interactions across decisions intensify and local optima proliferate, broader exploration is required. The models typically assume, however, that the interactions among decisions are distributed randomly. Contrary to this assumption, recent empirical studies of real organizational, social, and technological systems show that interactions among decisions are highly patterned. Patterns such as centralization, small-world connections, power-law distributions, hierarchy, and preferential attachment are common. We embed such patterns into an NK simulation model and obtain dramatic results: Holding fixed the total number of interactions among decisions, a shift in the pattern of interaction can alter the number of local optima by more than an order of magnitude. Thus, the long-ran value of broader exploration is significantly greater in the face of some interaction patterns than in the face of others. We develop simple, intuitive rules of thumb that allow a decision maker to examine two interaction patterns and determine which warrants greater investment in broad exploration. We also find that, holding fixed the interaction pattern, an increase in the number of interactions raises the number of local optima regardless of the pattern. This validates prior comparative static results with respect to the number of interactions, but highlights an important implicit assumption in earlier work-that the underlying interaction pattern remains constant as interactions become more numerous.