This dissertation conceptualizes a new type of clientelistic party, which despite being widespread has not been properly theorized. I refer to it as modular. Since clientelism — the exchange of goods for votes — requires substantial organization, scholars often assume that only dominant parties or solid political machines can engage in clientelistic mobilization. I show that is not the case. Rather, modular parties are makeshift organizations whose integrity from one election to the next is uncertain, but whose politicians are nevertheless able to mobilize voters through patron-client relationships.

Modular parties do not own, but "hire" or outsource the networks of clients they use. Well before the advent of elections, community chiefs, community organization leaders, ethnic leaders, landowners, local officials, and other types of local notables already established considerable political capital through their private and relatively small clienteles. Instead of ignoring or dismantling these networks, politicians running elections above the local level created modular parties to connect these networks. In this sense, modular par- ties are made up of two tiers: one on top, responsible for acquiring state resources and acting at the level of subnational or national politics; and one at the bottom composed of multiple modules, each with a local notable running local politics and acting as broker in favor of the upper stratum. Throughout time, these local leaders may take new roles, such as union leaders, bureaucrats, and local politicians, but they remain responsible for the construction and maintenance of patron–client networks. Today in Brazil, this local no- table is usually a mayoral candidate diligently brokering votes for the party offering most state resources or direct payments.

Such outsourcing of the organization of patron-client linkages to local authorities may facilitate the rapid mobilization of voters for politicians in modular parties, but it also prevents these politicians from building a reliable support base. Brokers in modular parties act as free–agents, and switch parties as they see appropriate. As long as there are other parties outsourcing clientelistic mobilization, brokers may switch whenever they receive a more lucrative proposal.

Using a research design that detects when parties receive resources they can use to hire brokers as-if randomly, I am able to show that variations in resources cause parties to expand or contract the number of modules working for them. Moreover, taking advantage of an unexpected institutional reform that made party switching potentially costly, and employing regression discontinuities to separate the brokers who were subject to this new ruling, I was able to evaluate how party switching drastically damages the electoral prospects of modular parties.

Substantively, the fluidity of modules sheds light on why clientelism can be resilient and widespread on many parts of the developing world, at the same time that strong clientelistic machines are relatively rare. It is durable because brokers offer their services to the highest bidding party, thus maximizing their ability to nurture their networks. However, by relying on these autonomous brokers parties will not create direct linkages of their own, frustrating any hope of making parties organizationally strong.