Abstract. We examine how participation in a microfinance program diffuses through social networks, using detailed demographic, social network, and participation data from 43 villages in South India. We exploit exogenous variation in the importance (in a network sense) of the people who were first informed about the program, the "injection points." Microfinance participation is significantly higher when the injection points have higher eigenvector centrality. We also estimate structural models of diffusion that allow us to (i) determine the relative roles of basic information transmission versus other forms of peer influence, and (ii) distinguish information passing by participants and nonparticipants. We find that participants are significantly more likely to pass information on to friends and acquaintances than informed non-participants. However, information passing by non-participants is still substantial and significant, accounting for roughly one-third of informedness and participation. We also find that, once we have properly conditioned on an individual being informed, her decision to participate is not significantly affected by the participation of her acquaintances.JEL Classification Codes: D85, D13, G21, L14, O12, O16, Z13

provided excellent research assistance. We thank Abhijit Banerjee for comments. We are particularly grateful to Betsy Levy Paluck, our discussant, for her detailed and thoughtful review of an earlier draft. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research. NBER working papers are circulated for discussion and comment purposes. They have not been peerreviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.

Does completing a household survey change the later behavior of those surveyed? In three field studies of health and two of microlending, we randomly assigned subjects to be surveyed about health and/or household finances and then measured subsequent use of a related product with data that does not rely on subjects' selfreports. In the three health experiments, we find that being surveyed increases use of water treatment products and take-up of medical insurance. Frequent surveys on reported diarrhea also led to biased estimates of the impact of improved source water quality. In two microlending studies, we do not find an effect of being surveyed on borrowing behavior. The results suggest that limited attention could play an important but context-dependent role in consumer choice, with the implication that researchers should reconsider whether, how, and how much to survey their subjects. measurement effects | question-behavior effects | Hawthorne effects | survey methodology | models of attention M any data collection efforts in the social and clinical sciences rely on surveys. Psychologists and marketing and survey experts have long held that surveying a subject can draw attention to risks or choices with ordinarily little salience and thereby induce changes in subsequent behavior (1, 2), but conclusive evidence on this question from which a causal relationship can be clearly established has been limited. We provide evidence from a variety of settings that the act of being surveyed can affect behavior and confound estimates of parameters that initially motivated the data collection."Survey" or "interview" effects may occur even when the survey does not ask specifically about intent to engage in the behavior of interest or provide new information, and even when subjects do not know that their later behavior is being observed by researchers. Hence survey effects are conceptually distinct from, but closely related to, "question-behavior" (i.e., mere measurement or self-prophecy) and Hawthorne effects. Question-behavior effects arise when behavior changes as a result of asking subjects for intentions or predictions regarding future behavior effects (3-6). Hawthorne effects occur when behavior changes as a result of a subject responding to being treated and observed, as part of an experiment (7). These effects are also related to "push polling" and other efforts to manipulate subject behavior by posing hypothetical questions (8).We describe results from five different field experiments, in four developing countries, on whether being surveyed affects subsequent behavior. The first experiment randomizes the frequency of surveys on health behavior and diarrhea incidence, in a context in which other water quality interventions are also randomly provided. More-frequent surveying leads to lower reported child diarrhea and cleaner water (as measured by the presence of detectable chlorine in household drinking water). We posit that frequent surveying serves as a reminder to invest in water purification, although because sub...