A lot. We derive common and conflicting predictions from models in which agents face information constraints and then assess their validity using surveys of consumers, firms, central bankers, and professional forecasters. We document that mean forecasts fail to completely adjust on impact to shocks, leading to statistically and economically significant deviations from the null of full information. The dynamics of forecast errors after shocks are consistent with the predictions of models with information rigidities. The conditional responses of forecast errors and disagreement among agents can also be used to differentiate between some of the most prominent models of information rigidities.