冷战结束后，国际军事冲突爆发的次数日益回升,提示了国际安全环境的高度不确定性。由于敌对国家间的争端根源未有明显变化，对于国际军事冲突爆发的理解可着眼于第三国（即赞助国）的参与是否激化矛盾并上升至军事冲突。在赞助国向保护国提供的多种安全产品中，本文选取武器以及防御性同盟义务以研究其对军事冲突开端的影响。武器转让及防御性同盟体现了截然不同的赞助国成本结构和信号机制。本文回顾了过往在武器转让领域以及同盟理论领域的文献，发现就武器转让与同盟关系对军事冲突的影响有诸多争议。争议主要集中于影响的方向，有学者认为这样的安全产品引发了更多的军事冲突，而部分学者认为它们降低了冲突发生的概率。两方观点都有一定的理论及经验支持。本文认为，现有文献观点的分歧可能有以下原因：未划分保护国在军事冲突中的角色，混淆了不同类型的同盟义务，忽略武器转让与同盟间的比较与结合，基于有缺陷的数据集，以及缺少统一的庇护关系视角。本文在过往文献的基础上，试图通过庇护关系视角分析武器转让与防御性同盟如何在赞助国、保护国与敌对国之间传递信号从而影响军事冲突的发生。文章提出假设:由于武器转让是可双向解读的幕后沉没成本信号，其带来的信息不对称使得武器转让的增加提升了保护国发起军事冲突的概率；为了降低信息不对称带来的损失，武器转让的增加也提升了敌对国向保护国发起军事冲突的概率。相反的，由于防御性同盟义务发出界定清晰的台前束手信号，赞助国与保护国之间的防御同盟可以解决部分信息不对称问题，从而降低武器转让带来的冲突风险。本文构建的面板数据包含1950年至2014年之间所有国际系统内国家对及其存在的有向武器转让及防御同盟关系。由于研究问题中的因变量军事冲突与主要自变量武器转让可能存在双向因果关系,文章基于常规面板probit形成三种解决内生性的策略，对提出的假设进行检验。这三种策略分别是使用滞后自变量，使用工具变量，以及采用二阶段条件似然模型。统计结果表明，武器转让使得保护国更可能主动发起军事冲突且更可能被当作军事目标攻击；防御性同盟能够减弱保护国被攻击的可能性；由于内生性的影响，统计结果无法证明两者之间的调节效应。

Since the end of the Cold War, the highly unpredictable global security environment has witnessed a notable rise in the frequency of militarized interstate disputes (MIDs). Overall, rivalries were involved in MIDs for recurring conflictual issues that have not undergone significant changes. Therefore, to understand the onsets of MIDs, it makes sense to ask whether the involvement of third parties (in this case, patron states) escalates the existing conflicts into militarized disputes. Among the security goods patrons provide to client states, this study selects arms transfers and defensive alliance obligations to probe into their effects on MID onsets. Arms transfers and defensive alliances embody distinct cost structures and signaling mechanisms of the patrons. By reviewing past literature in the field of arms transfers and alliances, this thesis shows that their effects on the probability of MID outbreaks, especially the direction of the effects, are highly controversial. Some scholars believe that more arms transfers and/or alliances contribute to more MIDs, whereas others believe that the two decrease the probability of MID initiation. Both sides are supported by a fair amount of theoretical and empirical evidence. This thesis attributes the mixed findings in the literature to the following potential reasons: failure to delineate the roles of clients in MIDs, confusion between different types of alliance obligations, insufficient comparison and combination of arms transfers and alliances, construction of limited datasets, and lack of a unified perspective of patron-client relationships.Based on the existing scholarly works, this article adopts a patron-client perspective to analyze how arms transfers and defensive alliance obligations send signals among patrons, clients, and adversaries and to investigate how such signaling influences the probabilities of clients being initiators and targets. This thesis then proposes the following hypotheses: because arms transfers are off-stage, sunk-cost signals that can be interpreted in both directions, the increases in arms transfers might aggravate the information asymmetries between clients and adversaries and eventually increase clients’ willingness to launch MIDs; on the other hand, in order to reduce the loss from information asymmetries, adversaries are also more likely to initiate MIDs against clients with increased arms transfers. Unlike arms transfers, defensive alliance obligations send well-defined, upstage, tying-hand signals that clear some of the doubts and uncertainties of adversaries, mitigating the MID risk brought by increased arms transfers.This study constructs a panel dataset that considers all country dyads identified in the state system between 1950 and 2014, along with the dyads’ directed arms transfers and defensive obligations. As the research question may suffer from simultaneity underlying MID onsets and arms transfers, three probit-based strategies dealing with endogeneity are developed to test the hypotheses. The first strategy uses lagged arms transfers as the main explanatory variable, the second strategy utilizes the instrumental variable framework, and the third employs two-stage conditional maximum likelihood (2SCML). The statistical results show that arms transfers make client states more likely to initiate MIDs and more likely to be targeted in MIDs; defensive alliance obligations can reduce the likelihood that clients become targets in MIDs. Due to endogeneity, the statistical results do not demonstrate the existence of moderation effects between arms transfers and defensive alliances.