Incidental data from sources such as social media, internet use, or mobile phones are generated at a staggering rate by a large proportion of society. Although such data contain considerable measurement error, they may still allow us to see – however dimly – into parts of social reality where traditional sources are blind. For example, they may record the impact of sudden events such as terrorist attacks, or reveal behaviors that self-reports would obscure. Use of incidental data in social research has skyrocketed—largely because of its predictive power: to call elections from tweets, flu cases from Google searches, and unemployment rates from mobile phone usage. But while prediction is certainly useful, social science is also concerned with advancing social theory by answering causal questions. In contrast with prediction, however, this requires valid measures of its variables – a requirement that incidental data, in spite of their exciting and potentially groundbreaking opportunities, simply do not fill. Thus, several decades of promise have revealed that progress in the use of incidental data for social theory is blocked by ubiquitous measurement errors. The aim of this project is to develop novel statistical methods that unlock incidental data. It will: 1. Develop a novel statistical framework to estimate and correct for measurement error in incidental data; 2. Provide the social science community with user-friendly software that can generate "virtual panel data” from sources such as Facebook, Google, Twitter, or LinkedIn; 3. Apply the developed novel techniques to answer substantive questions on basic human values, protests and repression, and environmentally friendly behavior. This will allow social scientists, for the first time, to reach into incidental data’s growing treasure trove and make valid comparative and longitudinal inferences on research questions of theoretical interest.