Integrating Theory and Data to Assess Government Policy Options

包括暴力抗议、恐怖活动和叛乱在内的持不同政见者行动要求国家作出反应，以促进国家安全。决策者和军方越来越多地求助于研究人员，以帮助他们更好地制定和评估他们所采取的干预措施。但潜在的国家干预范围很广，从外交或信息接触到经济发展或制裁，再到军事力量或警察行动，我们对这些干预如何发挥作用的理解有限。这个项目将从两个方面增进我们的了解。首先，它将采用尖端的大数据方法来预测不同干预后结果的可能性。这些预测将包括干预措施的平均效果及其方差，使政策制定者能够在成本效益计算中使用它们。其次，它将为这些预测提供一个新的理论基础。这将大大扩大我们预测的范围，因为我们将发现干预措施是如何以及为什么发挥作用的，从而使吸取的经验教训能够用于其他情况。除了其直接的实践和科学效益外，我们的研究还将为理论知情的大数据应用指明前进的道路，其新颖的核心概念具有在广泛领域显著吸收的潜力。目前，我们对国家干预亚州冲突的有效性及其运作方式的理解有限。为了提高这种理解，我们需要开发因果逻辑模型，通过这种模型，旨在在个人或社区层面发挥作用的干预措施会改变持不同政见者的集体动员。为了在这个问题上取得进展，我们将前沿的大数据方法与我们开发的一种新型理论模型结合起来：一种综合的理论-经验贝叶斯模型，该模型评估国家干预的个人层面影响，然后汇总个人决策来预测反国家行为。我们的模型将其输入——个人和派系层面的数据——与：(i)反对国家行动的动机的理论人口分布和（ii）对国家干预的个人层面反应的总体分布联系起来。该模型的输出是在给定状态干预的情况下预测的反状态行为的直接可测试的概率分布。因此，该模型将对国家干预措施的影响进行定量预测，以解决干预措施的平均效果和预测效果的差异。以前的模型没有将个人层面的因果机制、经验预测和直接的政策后果结合起来。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Dissident actions including violent protest, terrorist activity, and insurgency demand state responses in order to foster national security. Policy makers and the military are increasingly turning to researchers to help them better formulate and evaluate the interventions they undertake. But the range of potential state interventions is great, comprising anything from diplomatic or informational engagement to economic development or sanctions to military force or police action, and our understanding of how these interventions work is limited. This project will improve our understanding in two ways. First, it will employ cutting-edge Big Data approaches to make predictions as to the likelihoods of different post-intervention outcomes. These predictions will include both the mean effect of the interventions and their variance, allowing policy-makers to use them in cost-benefit calculations. Second, it will provide a novel theoretical undergirding for these predictions. This will greatly enhance the scope of our predictions, in that we will discover how and why the interventions worked as they did, enabling lessons learned to be used in other contexts. In addition to its direct practical and scientific benefit, our research will also show a way forward for theoretically-informed big data applications, and its novel core concepts have the potential for significant uptake across a wide range of fields.We have a limited understanding at present as to both the efficacy of state interventions into substate conflict and the manner in which they function. To improve this understanding we need to develop models of the causal logic by which an intervention designed to work at the individual or community level alters collective dissident mobilization. To make headway on this problem, we tie cutting-edge Big Data approaches to a new type of theoretical model we develop: an integrated theoretical-empirical Bayesian model that assesses the individual-level effects of state interventions and then aggregates individual decisions to predict anti-state action. Our model ties its inputs---individual- and faction-level data---to: (i) theoretical population distributions over incentives to act against the state and (ii) distributions over ensembles of individual-level responses to state interventions. The output of the model is a directly testable probability distribution over predicted anti-state action given a state intervention. As such, the model will produce quantitative predictions on the impact of state interventions that address both the mean effect of the interventions and the variance in their predicted performance. No previous model merges individual-level causal mechanisms, empirical prediction, and direct policy consequences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.