Infrastructure investments are typically long-term. As a result, observed benefits to households and communities may vary considerably over time as short-term outcomes generate or are subsumed by longer-term impacts. This paper uses a new round of household survey as part of a local government engineering department's rural road improvement project financed by the World Bank in Bangladesh to compare the short-term and long-term effects of rural roads over eight years. A dynamic panel model, estimated by generalized method of moments, is applied to estimate the varying returns to public road investment accounting for time-varying unobserved characteristics. The results show that the substantial effects of roads on such outcomes as per capita expenditure, schooling, and prices as observed in the short run attenuate over time. But the declining returns are not common for all outcomes of interest or all households. Employment in the rural non-farm sector, for example, has risen more rapidly over time, indicating increasing returns to investment. The very poor have failed to sustain the short-term benefits of roads, and yet the gains accrued to the middle-income groups are strengthened over time because of changing sectors of employment, away from agriculture toward non-farm activity. The results also show that initial state dependence-or initial community and household characteristics as well as road quality-matters in estimating the trajectory of road impacts.