This thesis investigates the design, implementation and impacts of the market-smart input subsidy (NAIVS) in Tanzania’s Ruvuma Region. The research uses a mixed-methods approach, where quantitative data analysis is complemented by qualitative research. Using four waves of household panel data, I found that voucher receipt had no statistically significant impact on maize yields, income poverty or the household assets owned by recipient households. The qualitative research finds that this was due to flaws in NAIVS’s design and in its implementation. Weak institutional capacity was found in voucher management, especially at the lower level of government: a substantial number of vouchers went missing; inputs and vouchers were delivered late most years; and vouchers were resold by farmers. Due to an increase in real input prices, the ‘top-up’ payment required for voucher use was increased, which made it difficult for poor farmers to access the subsidy. In practice, the input vouchers were obtained by elites: households with elected positions in the villages; wealthier households; and those households who were already using improved inputs prior to NAIVS. It contributed to national food security; however, because of the spill-over effects which brought a higher increase ratio in input use among non-recipient than recipient households, the observed impact on maize yields cannot be attributed to NAIVS. Because of the leakage to wealthier farmers and fraud, it did not ensure household food security for poor farmers. The thesis reveals that studies of input subsidy programmes require not only economic analysis but also social and political analysis. Such studies would require the use of a new theory of change, which uses economic analysis but places social and political analysis at the forefront, and in which a mixed-methods approach must be used.