Personalization of Power and Social Welfare Spending under Dictatorships: A Cross-Country Analysis

傳統觀點認為，因為選舉競爭促使民主國家領導人採取社會政策以爭取公民政治支持，所以民主國家比獨裁國家具有更多福利支出。然而，由於獨裁國家沒有選舉、或其選舉並不自由和／或不公平，因此本文的研究問題是：獨裁者與其盟友之間的權力分布如何影響福利支出？本文基於晚近發展的權力分享理論，討論權力分享結構對於威權國家福利支出的影響。當權力個人化統治程度較低，獨裁者面臨被其他精英取代的風險。我們認為，獨裁者藉由福利政策向公民分配資源，換取公民政治支持。當權力更集中於獨裁者，增加權力個人化統治將減少福利支出。我們使用1961年至2006年90個獨裁國家時間序列橫斷面資料檢驗我們論點。我們發現，總福利支出隨著獨裁者權力個人化統治的增加而減少。這種影響在不同模型設定中皆相當穩健。此一發現有助於理解獨裁國家權力個人化統治與福利支出間的聯繫。本文貢獻在於指出福利支出不僅由文獻認為的政權領導人需要公眾支持所決定，在獨裁國家中也是由其統治聯盟內部權力分布所決定。Conventional wisdom posits that democracies have more social welfare expenditure than dictatorships because electoral competition induces democratic leaders to adopt social policies in exchange for ordinary citizens’ political support. However, as elections are absent, unfree, and/or unfair in autocracies, the research question in this paper is: How does the distribution of power between dictators and their allies influence welfare spending? This paper discusses the power-sharing structure’s impact on welfare spending based on the theory of power-sharing in authoritarian regimes. When power is less personalized, dictators face the risk of being replaced by other elites. We argue that dictators allocate resources to citizens through welfare policies in exchange for their political support. Increasing personalism rules will reduce welfare spending when power is concentrated on dictators. We test our arguments using time-series cross-sectional data for 90 dictatorships from 1961 to 2006. We find that total welfare spending decreases as the personalization of power of the dictator increases. This effect is robust to different model specifications. This finding helps us to understand the connection between the personalization of power and welfare spending in dictatorships. This paper contributes to the literature by showing that welfare spending is determined by regime leaders’ need for public support as the conventional wisdom suggests, and by the power dynamics within their ruling coalitions in dictatorships.