Federal statistical agencies collect survey data that are used for dispersing funds, informing policy, and aiding research. The collection of accurate data is crucial for these activities, but survey response rates have been declining for three decades. We provide new evidence on privacy and confidentiality concerns as a possible cause for declining response rates. First, we present a model in which individuals choose whether to respond to a federal agency's survey while also interacting with a firm that relies in part on the agency's published data to imperfectly price discriminate. The model demonstrates how privacy loss risk impacts survey response and provides some testable implications. Next, we empirically test the relationship between privacy and survey response using the staggered rollout of broadband internet across the United States as a technology shock that increased individuals' privacy loss risk. Refusal in the Current Population Survey increased immediately after broadband services entered a county. Broadband rollout can explain nearly all of the increase in refusal from 1995-2008. Consistent with the model, we find the impact of broadband rollout on refusal was larger for counties with greater proxied household willingness to pay for an arbitrary good or service and, among households who did respond, broadband rollout increased question refusal for topics that reveal more private and sensitive information. Finally, we present some back-of-the-envelope calculations for the implied change in privacy loss risk due to the rollout of broadband internet.