This dissertation studies the design of social policy and social protection, specifically programs intended to improve maternal and child health outcomes and human capital accumulation in low- and middle-income countries. The first substantive chapter studies the design of reproductive health policy. We use data from a randomized controlled trial that provides pregnant and immediate postpartum women with improved access to family planning to study the impacts of these services on child growth and development outcomes. We focus our analysis on 1,034 children born to participating women directly before intervention rollout and find that those born to mothers assigned to the intervention arm were 0.28-0.34 standard deviations taller for their age and 10.7-12.0 percentage points (p.p.) less likely to be stunted within a year of exposure to the intervention. They also scored 0.19-0.23 standard deviations higher on a caregiver-reported measure of cognitive development after two years of intervention exposure. Non-measurement of children is a challenge in this chapter; however, our estimates are robust to multiple methods of correcting for potential attrition bias. Our results are consistent with models of fertility that link couples' fertility decisions to child health and human capital and suggest that improved access to family planning may have positive policy-relevant impacts beyond first-line outcomes. One crucial aspect of program design addressed in the next chapter is program targeting. In a world of scarce resources, governments must be able to target their social programs efficiently. Over the past 20 years, governments in contexts without high-quality administrative data have increasingly used proxy means testing (PMT) to conduct this targeting due to its ease and relatively small data requirements. In this chapter, we explore the potential of neural networks and satellite imagery to reduce the error rate of traditional PMT and test whether those reductions translate to improved effectiveness of programs that rely on the method. We use administrative data to conduct an ex-post targeting exercise of the Progresa cash transfer experiment in Mexico. In our context, the overall gains associated with utilizing neural networks and satellite imagery are relatively small -- around an additional three percentage points of variation explained. However, we decompose these gains into different error types and find that our model that overweights households towards the endpoints of the consumption distribution decreases exclusion error with a limited tradeoff in terms of inclusion error. We compare the program effectiveness of Progresa on nutrition and poverty outcomes for the full sample, those targeted by traditional PMT, and those targeted by our preferred model. We find little evidence of significant differences in program effectiveness from the gains in targeting accuracy. This suggests that spatial data and advanced statistical methods may be of limited value when high-quality survey data is available. In the final substantive chapter, we study the impacts of a multi-faceted cash-transfer program, Desnutritión Cero, intended to incentivize maternal health-seeking and improve birth outcomes using cash transfers conditional on ante and postnatal care and lump-sum grants before delivery. We take a difference-in-discontinuities approach and exploit a discontinuity created by the program's targeting method to identify the causal impact of the program and find evidence of small improvements in antenatal care uptake and facility-based delivery. Our results also suggest a marginal decrease in birth weights, which we show is likely the result of selection into measurement. We can dismiss the possibility of large effects and show that our results are robust to several specification and sensitivity analyses. Our results contrast with evidence from other maternity benefit schemes, which often show large improvements in facility-based delivery care and child vaccination and highlight the need for careful program targeting and design.