Does knowledge about antipoverty programs spread quickly within poor communities or are there significant frictions, such as due to social exclusion? We combine longitudinal and intra-household observations in estimating the direct knowledge gain from watching an information movie in rural India, while randomized village assignment identifies knowledge sharing with those in treatment villages who did not watch the movie. Knowledge is found to be shared within villages, but less so among illiterate and lower caste individuals, especially when also poor; these groups relied more on actually seeing the movie. Sizable biases are evident in impact estimators that ignore knowledge spillovers.

JEL: D83,I38,O12