This dissertation addresses a consistent empirical challenge to the predictions of democratic theory: the poor, despite being enfranchised and numerous, often fail to secure public provision of services they demand. Traditional explanations suggest that either the poor do not actually demand such services, or more likely, they find themselves unable to hold elected politicians accountable for service provision. In this dissertation, I propose a new mechanism that could underlie constrained accountability. I show that even when the poor demand services, turn out to vote, and vote their interests, they may still be disadvantaged in service provision in democracy under certain circumstances. In particular, they will be disadvantaged if their service interests are diffuse across various service categories while wealthier individuals have more concentrated service demands. This is most likely when the demands of the poor differ greatly from the demands of the wealthy, which I expect to be most common in highly unequal societies. In Chapter 2, I present a mathematical model that demonstrates the logic of my argument. Typical democratic accountability arguments suggest that groups' policy representation should be weighted proportionally to the size of the group, assuming all groups turn out to vote at equal rates and possess the same information. My model confirms this logic, but highlights a mediating factor: group weight in policy representation is also governed by the concentration of service demands for the group, compared to the concentration of other groups' demands. Chapters 3 and 4 attempt to address various implications of the argument in a real setting. I focus on the provision of primary care and hospital services in a single decentralized democracy, Brazil, to hold democratic institutions constant. In Chapter 3, I use original survey data to show that the poor in Brazil prefer more municipal spending on public primary care, while the better-off prefer spending on hospitals. I also show that a conditional cash transfer program (Bolsa Familia) that decreases the dependence of poor recipients on public primary care provision serves to shift the preferences of this group toward hospital care and to increase recipients' responsiveness to service spending by concentrating service demands on their preferred service. In Chapter 4, I turn to the macro-implications of the model, showing that Bolsa Familia is, indeed, associated with decreases in public spending on primary care services at the municipal level. The chapter shows that municipal funds follow the shifting preferences of transfer-recipients, toward hospital care. I show that the shift seems to be the result not just of a growth in group size for voters that prefer hospital spending, but also results from increased policy influence for the recipient group whose preferences become more concentrated on hospital care than they previously were on primary care. Finally, Chapter 5 addresses the welfare implications of this research. It shows that there are consequences for the infant rates of the non-recipient poor following the decrease in spending on public primary care clinics shown to be associated with Bolsa Familia.